repo
stringlengths
1
99
file
stringlengths
13
215
code
stringlengths
12
59.2M
file_length
int64
12
59.2M
avg_line_length
float64
3.82
1.48M
max_line_length
int64
12
2.51M
extension_type
stringclasses
1 value
adversarial_ntk_evolution
adversarial_ntk_evolution-master/data.py
import torch import numpy as np import torchvision transform_train = torchvision.transforms.Compose([ torchvision.transforms.ToTensor(), ]) def get_loader(dataset_name, train = True, batch_size = 128, shuffle = True): if dataset_name == 'cifar10': dataset = torchvision.datasets.CIFAR10(root='./data', train=train, download=True, transform=transform_train) return torch.utils.data.DataLoader(dataset, batch_size=batch_size, shuffle=shuffle, num_workers=4) elif dataset_name == 'cifar100': dataset = torchvision.datasets.CIFAR100(root='./data', train=train, download=True, transform=transform_train) return torch.utils.data.DataLoader(dataset, batch_size=batch_size, shuffle=shuffle, num_workers=4) def get_data_and_labels(dataset_name): if dataset_name == 'cifar10': dataset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform_train) elif dataset_name == 'cifar100': dataset = torchvision.datasets.CIFAR100(root='./data', train=True, download=True, transform=transform_train) return torch.tensor(np.transpose(dataset.data, [0, 3, 1, 2]) / 255), torch.tensor(dataset.targets) def get_n_classes(dataset_name): if dataset_name == 'cifar10': return 10 elif dataset_name == 'cifar100': return 100
1,368
41.78125
117
py
adversarial_ntk_evolution
adversarial_ntk_evolution-master/eval_and_make_adv.py
import jax import haiku as hk import jax.numpy as jnp from jax.example_libraries import optimizers import torch import torchvision import torchvision.transforms as transforms from torch.utils.data import Dataset import numpy as np import neural_tangents as nt import functools import operator import optax import copy import models import pickle from utils import bind, _add, _sub import os from test_functions import do_perturbation_step_l_inf, do_perturbation_step_l_2, perturb, test, loss_fn import numpy as np import argparse import time import data def main(): parser = argparse.ArgumentParser() parser.add_argument('--model_path', type=str, default = '', help = 'where the model is saved') parser.add_argument('--save_name', type=str, default = 'test_results', help = 'extra specifier to include when saving results') parser.add_argument('--linear', action='store_true', help = 'whether the loaded model is in linearized dynamics') parser.add_argument('--centering', action='store_true', help = 'whether the loaded model is in centered dynamics') parser.add_argument('--test_path', type=str, default = '', help = 'where the test images are. Set it to empty to use standard cifar10/cifar100 images. This is used for adversarial transferability testing') parser.add_argument('--bonus_dir', type=str, default = '.', help = 'include an extra dir the save path for more specific save locations') parser.add_argument('--no_adv', action='store_true', help = 'whether to skip making adversarial examples') parser.add_argument('--save_examples', action='store_true', help = 'whether to save the adversarial examples') parser.add_argument('--short', action='store_true', help = 'dont set this. basically just for debugging') parser.add_argument('--random_seed', type = int, default = 0, help = 'random seed') parser.add_argument('--dataset', type=str, default = 'cifar10', help = 'cifar10/cifar100') parser.add_argument('--model', type=str, default = 'resnet18', help = 'model') parser.add_argument('--eps', type=float, default = 4.00, help = 'eps value for l-inf adversarial attacks. scaled by 1/255') args = parser.parse_args() transform_test = transforms.Compose([ transforms.ToTensor(), ]) class TensorDataset(Dataset): def __init__(self, *tensors, transform=None): assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors) self.tensors = tensors self.transform = transform def __getitem__(self, index): im, targ = tuple(tensor[index] for tensor in self.tensors) if self.transform: real_transform = transforms.Compose([ transforms.ToPILImage(), self.transform ]) im = real_transform(im) return im, targ def __len__(self): return self.tensors[0].size(0) rng = jax.random.PRNGKey(args.random_seed) net_forward_init, net_forward_apply = models.get_model(args.model, data.get_n_classes(args.dataset)) checkpoint = pickle.load(open('./{}'.format(args.model_path), 'rb')) params = checkpoint['params'] lin_params = checkpoint['lin_params'] net_state = checkpoint['net_state'] if len(args.test_path) > 0: test_stuff = pickle.load(open('./{}'.format(args.test_path), 'rb')) test_data = torch.tensor(test_stuff['images']).cpu() print(test_data.shape) test_labels = torch.tensor(test_stuff['labels']).cpu() print(test_labels) test_dataset = TensorDataset(test_data, test_labels, transform=transform_test) elif args.dataset == 'cifar10': test_dataset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform_test) test_labels = np.array(test_dataset.targets) print(test_labels) elif args.dataset == 'cifar100': test_dataset = torchvision.datasets.CIFAR100(root='./data', train=False, download=True, transform=transform_test) test_labels = np.array(test_dataset.targets) print(test_labels) test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=100, shuffle=False, num_workers=4) acc_clean, acc_dirty, adv_examples, predictions_clean, predictions_dirty, components_clean, components_dirty = test(params, lin_params, net_state, net_forward_apply, rng, test_loader, make_adv_examples = not args.no_adv, linear = args.linear, centering = args.centering, attack = 'linf', return_examples = True, short = args.short, return_components = True, adv_eps = args.eps) base_path = os.path.dirname(args.model_path) results_dict = { 'acc_clean': acc_clean, 'acc_dirty': acc_dirty, 'predictions_clean': predictions_clean, 'predictions_dirty': predictions_dirty, 'components_clean': components_clean, 'components_dirty': components_dirty, } print('clean: {:.2f} dirty: {:.2f}'.format(100 * acc_clean, 100 * acc_dirty)) if not os.path.isdir('./{}/{}/'.format(base_path, args.bonus_dir)): os.mkdir('./{}/{}/'.format(base_path, args.bonus_dir)) pickle.dump(results_dict, open('./{}/{}/test_results_{}.pkl'.format(base_path, args.bonus_dir, args.save_name),'wb')) if args. save_examples: pickle.dump({'images': np.transpose(adv_examples, [0, 3, 1, 2]), 'labels': test_labels[:adv_examples.shape[0]]}, open('./{}/{}/adv_examples_{}.pkl'.format(base_path, args.bonus_dir, args.save_name),'wb')) if __name__ == '__main__': main()
5,638
43.401575
381
py
adversarial_ntk_evolution
adversarial_ntk_evolution-master/models.py
import jax import haiku as hk import jax.numpy as jnp from jax.example_libraries import optimizers import torch import torchvision import torchvision.transforms as transforms from torch.utils.data import Dataset import numpy as np import neural_tangents as nt import functools import operator import optax import copy import models import sys from utils import bind, _sub, _add import modified_resnets @functools.partial(jax.jit, static_argnums=(3, 6, 7, 8)) def linear_forward(params, params2, state, net_fn, rng, images, is_training = False, centering = False, return_components = False): dparams = _sub(params2, params) f_0, df_0, state = jax.jvp(lambda param: net_fn(param, state, rng, images, is_training = is_training), (params,), (dparams,), has_aux = True) if return_components: if centering: return df_0, {'state': state, 'f': f_0, 'df': df_0} return _add(f_0, df_0), {'state': state, 'f': f_0, 'df': df_0} if centering: return df_0, state return _add(f_0, df_0), state def get_resnet(n_classes): def _forward_resnet18(x, is_training): #use a 3x3 kernel size with stride 1 for the first layer because we are using 32x32 images net = modified_resnets.ResNet18(n_classes, initial_conv_config = {'kernel_shape': 3, 'stride': 1}) return net(x, is_training) net_forward = hk.transform_with_state(_forward_resnet18) return net_forward.init, net_forward.apply def _forward_narrow_mlp(x, is_training): mlp = hk.Sequential([ hk.Flatten(), hk.Linear(256), jax.nn.relu, hk.Linear(256), jax.nn.relu, hk.Linear(10), ]) return mlp(x) def _forward_wide_mlp(x, is_training): mlp = hk.Sequential([ hk.Flatten(), hk.Linear(2048), jax.nn.relu, hk.Linear(2048), jax.nn.relu, hk.Linear(10), ]) return mlp(x) def get_narrow_mlp(): net_forward = hk.transform_with_state(_forward_narrow_mlp) return net_forward.init, net_forward.apply def get_wide_mlp(): net_forward = hk.transform_with_state(_forward_wide_mlp) return net_forward.init, net_forward.apply def get_model(model_name, n_classes): if model_name == 'resnet18': return get_resnet(n_classes) elif model_name == 'mlp_skinny': return get_narrow_mlp() elif model_name == 'mlp_wide': return get_wide_mlp() else: print("Invalid model: {}".format(model_name)) sys.exit()
2,491
29.024096
145
py
adversarial_ntk_evolution
adversarial_ntk_evolution-master/modified_resnets.py
#Copied from Deepmind Haiku library """Resnet.""" import types from typing import Mapping, Optional, Sequence, Union, Any from haiku._src import basic from haiku._src import batch_norm from haiku._src import conv from haiku._src import module from haiku._src import pool import jax import jax.numpy as jnp # If forking replace this block with `import haiku as hk`. hk = types.ModuleType("haiku") hk.Module = module.Module hk.BatchNorm = batch_norm.BatchNorm hk.Conv2D = conv.Conv2D hk.Linear = basic.Linear hk.max_pool = pool.max_pool del basic, batch_norm, conv, module, pool FloatStrOrBool = Union[str, float, bool] class BlockV1(hk.Module): """ResNet V1 block with optional bottleneck.""" def __init__( self, channels: int, stride: Union[int, Sequence[int]], use_projection: bool, bn_config: Mapping[str, FloatStrOrBool], bottleneck: bool, name: Optional[str] = None, ): super().__init__(name=name) self.use_projection = use_projection bn_config = dict(bn_config) bn_config.setdefault("create_scale", True) bn_config.setdefault("create_offset", True) bn_config.setdefault("decay_rate", 0.999) if self.use_projection: self.proj_conv = hk.Conv2D( output_channels=channels, kernel_shape=1, stride=stride, with_bias=False, padding="SAME", name="shortcut_conv", ) self.proj_batchnorm = hk.BatchNorm(name="shortcut_batchnorm", **bn_config) channel_div = 4 if bottleneck else 1 conv_0 = hk.Conv2D( output_channels=channels // channel_div, kernel_shape=1 if bottleneck else 3, stride=1 if bottleneck else stride, with_bias=False, padding="SAME", name="conv_0", ) bn_0 = hk.BatchNorm(name="batchnorm_0", **bn_config) conv_1 = hk.Conv2D( output_channels=channels // channel_div, kernel_shape=3, stride=stride if bottleneck else 1, with_bias=False, padding="SAME", name="conv_1", ) bn_1 = hk.BatchNorm(name="batchnorm_1", **bn_config) layers = ((conv_0, bn_0), (conv_1, bn_1)) if bottleneck: conv_2 = hk.Conv2D( output_channels=channels, kernel_shape=1, stride=1, with_bias=False, padding="SAME", name="conv_2", ) bn_2 = hk.BatchNorm(name="batchnorm_2", scale_init=jnp.zeros, **bn_config) layers = layers + ((conv_2, bn_2),) self.layers = layers def __call__(self, inputs, is_training, test_local_stats): out = shortcut = inputs if self.use_projection: shortcut = self.proj_conv(shortcut) shortcut = self.proj_batchnorm(shortcut, is_training, test_local_stats) for i, (conv_i, bn_i) in enumerate(self.layers): out = conv_i(out) out = bn_i(out, is_training, test_local_stats) if i < len(self.layers) - 1: # Don't apply relu on last layer out = jax.nn.relu(out) return jax.nn.relu(out + shortcut) class BlockV2(hk.Module): """ResNet V2 block with optional bottleneck.""" def __init__( self, channels: int, stride: Union[int, Sequence[int]], use_projection: bool, bn_config: Mapping[str, FloatStrOrBool], bottleneck: bool, name: Optional[str] = None, ): super().__init__(name=name) self.use_projection = use_projection bn_config = dict(bn_config) bn_config.setdefault("create_scale", True) bn_config.setdefault("create_offset", True) if self.use_projection: self.proj_conv = hk.Conv2D( output_channels=channels, kernel_shape=1, stride=stride, with_bias=False, padding="SAME", name="shortcut_conv", ) channel_div = 4 if bottleneck else 1 conv_0 = hk.Conv2D( output_channels=channels // channel_div, kernel_shape=1 if bottleneck else 3, stride=1 if bottleneck else stride, with_bias=False, padding="SAME", name="conv_0", ) bn_0 = hk.BatchNorm(name="batchnorm_0", **bn_config) conv_1 = hk.Conv2D( output_channels=channels // channel_div, kernel_shape=3, stride=stride if bottleneck else 1, with_bias=False, padding="SAME", name="conv_1", ) bn_1 = hk.BatchNorm(name="batchnorm_1", **bn_config) layers = ((conv_0, bn_0), (conv_1, bn_1)) if bottleneck: conv_2 = hk.Conv2D( output_channels=channels, kernel_shape=1, stride=1, with_bias=False, padding="SAME", name="conv_2", ) # NOTE: Some implementations of ResNet50 v2 suggest initializing # gamma/scale here to zeros. bn_2 = hk.BatchNorm(name="batchnorm_2", **bn_config) layers = layers + ((conv_2, bn_2),) self.layers = layers def __call__(self, inputs, is_training, test_local_stats): x = shortcut = inputs for i, (conv_i, bn_i) in enumerate(self.layers): x = bn_i(x, is_training, test_local_stats) x = jax.nn.relu(x) if i == 0 and self.use_projection: shortcut = self.proj_conv(x) x = conv_i(x) return x + shortcut class BlockGroup(hk.Module): """Higher level block for ResNet implementation.""" def __init__( self, channels: int, num_blocks: int, stride: Union[int, Sequence[int]], bn_config: Mapping[str, FloatStrOrBool], resnet_v2: bool, bottleneck: bool, use_projection: bool, name: Optional[str] = None, ): super().__init__(name=name) block_cls = BlockV2 if resnet_v2 else BlockV1 self.blocks = [] for i in range(num_blocks): self.blocks.append( block_cls( channels=channels, stride=(1 if i else stride), use_projection=(i == 0 and use_projection), bottleneck=bottleneck, bn_config=bn_config, name="block_%d" % (i), ) ) def __call__(self, inputs, is_training, test_local_stats): out = inputs for block in self.blocks: out = block(out, is_training, test_local_stats) return out def check_length(length, value, name): if len(value) != length: raise ValueError(f"`{name}` must be of length 4 not {len(value)}") class ResNet(hk.Module): """ResNet model.""" CONFIGS = { 18: { "blocks_per_group": (2, 2, 2, 2), "bottleneck": False, "channels_per_group": (64, 128, 256, 512), "use_projection": (False, True, True, True), }, 34: { "blocks_per_group": (3, 4, 6, 3), "bottleneck": False, "channels_per_group": (64, 128, 256, 512), "use_projection": (False, True, True, True), }, 50: { "blocks_per_group": (3, 4, 6, 3), "bottleneck": True, "channels_per_group": (256, 512, 1024, 2048), "use_projection": (True, True, True, True), }, 101: { "blocks_per_group": (3, 4, 23, 3), "bottleneck": True, "channels_per_group": (256, 512, 1024, 2048), "use_projection": (True, True, True, True), }, 152: { "blocks_per_group": (3, 8, 36, 3), "bottleneck": True, "channels_per_group": (256, 512, 1024, 2048), "use_projection": (True, True, True, True), }, 200: { "blocks_per_group": (3, 24, 36, 3), "bottleneck": True, "channels_per_group": (256, 512, 1024, 2048), "use_projection": (True, True, True, True), }, } BlockGroup = BlockGroup # pylint: disable=invalid-name BlockV1 = BlockV1 # pylint: disable=invalid-name BlockV2 = BlockV2 # pylint: disable=invalid-name def __init__( self, blocks_per_group: Sequence[int], num_classes: int, bn_config: Optional[Mapping[str, FloatStrOrBool]] = None, resnet_v2: bool = False, bottleneck: bool = True, channels_per_group: Sequence[int] = (256, 512, 1024, 2048), use_projection: Sequence[bool] = (True, True, True, True), logits_config: Optional[Mapping[str, Any]] = None, name: Optional[str] = None, initial_conv_config: Optional[Mapping[str, FloatStrOrBool]] = None, strides: Sequence[int] = (1, 2, 2, 2), ): """Constructs a ResNet model. Args: blocks_per_group: A sequence of length 4 that indicates the number of blocks created in each group. num_classes: The number of classes to classify the inputs into. bn_config: A dictionary of two elements, ``decay_rate`` and ``eps`` to be passed on to the :class:`~haiku.BatchNorm` layers. By default the ``decay_rate`` is ``0.9`` and ``eps`` is ``1e-5``. resnet_v2: Whether to use the v1 or v2 ResNet implementation. Defaults to ``False``. bottleneck: Whether the block should bottleneck or not. Defaults to ``True``. channels_per_group: A sequence of length 4 that indicates the number of channels used for each block in each group. use_projection: A sequence of length 4 that indicates whether each residual block should use projection. logits_config: A dictionary of keyword arguments for the logits layer. name: Name of the module. initial_conv_config: Keyword arguments passed to the constructor of the initial :class:`~haiku.Conv2D` module. strides: A sequence of length 4 that indicates the size of stride of convolutions for each block in each group. """ super().__init__(name=name) self.resnet_v2 = resnet_v2 bn_config = dict(bn_config or {}) bn_config.setdefault("decay_rate", 0.9) bn_config.setdefault("eps", 1e-5) bn_config.setdefault("create_scale", True) bn_config.setdefault("create_offset", True) logits_config = dict(logits_config or {}) logits_config.setdefault("w_init", jnp.zeros) logits_config.setdefault("name", "logits") # Number of blocks in each group for ResNet. check_length(4, blocks_per_group, "blocks_per_group") check_length(4, channels_per_group, "channels_per_group") check_length(4, strides, "strides") initial_conv_config = dict(initial_conv_config or {}) initial_conv_config.setdefault("output_channels", 64) initial_conv_config.setdefault("kernel_shape", 7) initial_conv_config.setdefault("stride", 2) initial_conv_config.setdefault("with_bias", False) initial_conv_config.setdefault("padding", "SAME") initial_conv_config.setdefault("name", "initial_conv") self.initial_conv = hk.Conv2D(**initial_conv_config) if not self.resnet_v2: self.initial_batchnorm = hk.BatchNorm(name="initial_batchnorm", **bn_config) self.block_groups = [] for i, stride in enumerate(strides): self.block_groups.append( BlockGroup( channels=channels_per_group[i], num_blocks=blocks_per_group[i], stride=stride, bn_config=bn_config, resnet_v2=resnet_v2, bottleneck=bottleneck, use_projection=use_projection[i], name="block_group_%d" % (i), ) ) if self.resnet_v2: self.final_batchnorm = hk.BatchNorm(name="final_batchnorm", **bn_config) self.logits = hk.Linear(num_classes, **logits_config) def __call__(self, inputs, is_training, test_local_stats=False): out = inputs out = self.initial_conv(out) if not self.resnet_v2: out = self.initial_batchnorm(out, is_training, test_local_stats) out = jax.nn.relu(out) #We remove the final max pool because we are dealing with smaller 32x32 images instead of the standard 256x256 imagenet images that this model was designed for # out = hk.max_pool(out, # window_shape=(1, 3, 3, 1), # strides=(1, 2, 2, 1), # padding="SAME") for block_group in self.block_groups: out = block_group(out, is_training, test_local_stats) if self.resnet_v2: out = self.final_batchnorm(out, is_training, test_local_stats) out = jax.nn.relu(out) out = jnp.mean(out, axis=(1, 2)) return self.logits(out) class ResNet18(ResNet): """ResNet18.""" def __init__( self, num_classes: int, bn_config: Optional[Mapping[str, FloatStrOrBool]] = None, resnet_v2: bool = False, logits_config: Optional[Mapping[str, Any]] = None, name: Optional[str] = None, initial_conv_config: Optional[Mapping[str, FloatStrOrBool]] = None, strides: Sequence[int] = (1, 2, 2, 2), ): """Constructs a ResNet model. Args: num_classes: The number of classes to classify the inputs into. bn_config: A dictionary of two elements, ``decay_rate`` and ``eps`` to be passed on to the :class:`~haiku.BatchNorm` layers. resnet_v2: Whether to use the v1 or v2 ResNet implementation. Defaults to ``False``. logits_config: A dictionary of keyword arguments for the logits layer. name: Name of the module. initial_conv_config: Keyword arguments passed to the constructor of the initial :class:`~haiku.Conv2D` module. strides: A sequence of length 4 that indicates the size of stride of convolutions for each block in each group. """ super().__init__( num_classes=num_classes, bn_config=bn_config, initial_conv_config=initial_conv_config, resnet_v2=resnet_v2, strides=strides, logits_config=logits_config, name=name, **ResNet.CONFIGS[18], ) class ResNet34(ResNet): """ResNet34.""" def __init__( self, num_classes: int, bn_config: Optional[Mapping[str, FloatStrOrBool]] = None, resnet_v2: bool = False, logits_config: Optional[Mapping[str, Any]] = None, name: Optional[str] = None, initial_conv_config: Optional[Mapping[str, FloatStrOrBool]] = None, strides: Sequence[int] = (1, 2, 2, 2), ): """Constructs a ResNet model. Args: num_classes: The number of classes to classify the inputs into. bn_config: A dictionary of two elements, ``decay_rate`` and ``eps`` to be passed on to the :class:`~haiku.BatchNorm` layers. resnet_v2: Whether to use the v1 or v2 ResNet implementation. Defaults to ``False``. logits_config: A dictionary of keyword arguments for the logits layer. name: Name of the module. initial_conv_config: Keyword arguments passed to the constructor of the initial :class:`~haiku.Conv2D` module. strides: A sequence of length 4 that indicates the size of stride of convolutions for each block in each group. """ super().__init__( num_classes=num_classes, bn_config=bn_config, initial_conv_config=initial_conv_config, resnet_v2=resnet_v2, strides=strides, logits_config=logits_config, name=name, **ResNet.CONFIGS[34], ) class ResNet50(ResNet): """ResNet50.""" def __init__( self, num_classes: int, bn_config: Optional[Mapping[str, FloatStrOrBool]] = None, resnet_v2: bool = False, logits_config: Optional[Mapping[str, Any]] = None, name: Optional[str] = None, initial_conv_config: Optional[Mapping[str, FloatStrOrBool]] = None, strides: Sequence[int] = (1, 2, 2, 2), ): """Constructs a ResNet model. Args: num_classes: The number of classes to classify the inputs into. bn_config: A dictionary of two elements, ``decay_rate`` and ``eps`` to be passed on to the :class:`~haiku.BatchNorm` layers. resnet_v2: Whether to use the v1 or v2 ResNet implementation. Defaults to ``False``. logits_config: A dictionary of keyword arguments for the logits layer. name: Name of the module. initial_conv_config: Keyword arguments passed to the constructor of the initial :class:`~haiku.Conv2D` module. strides: A sequence of length 4 that indicates the size of stride of convolutions for each block in each group. """ super().__init__( num_classes=num_classes, bn_config=bn_config, initial_conv_config=initial_conv_config, resnet_v2=resnet_v2, strides=strides, logits_config=logits_config, name=name, **ResNet.CONFIGS[50], ) class ResNet101(ResNet): """ResNet101.""" def __init__( self, num_classes: int, bn_config: Optional[Mapping[str, FloatStrOrBool]] = None, resnet_v2: bool = False, logits_config: Optional[Mapping[str, Any]] = None, name: Optional[str] = None, initial_conv_config: Optional[Mapping[str, FloatStrOrBool]] = None, strides: Sequence[int] = (1, 2, 2, 2), ): """Constructs a ResNet model. Args: num_classes: The number of classes to classify the inputs into. bn_config: A dictionary of two elements, ``decay_rate`` and ``eps`` to be passed on to the :class:`~haiku.BatchNorm` layers. resnet_v2: Whether to use the v1 or v2 ResNet implementation. Defaults to ``False``. logits_config: A dictionary of keyword arguments for the logits layer. name: Name of the module. initial_conv_config: Keyword arguments passed to the constructor of the initial :class:`~haiku.Conv2D` module. strides: A sequence of length 4 that indicates the size of stride of convolutions for each block in each group. """ super().__init__( num_classes=num_classes, bn_config=bn_config, initial_conv_config=initial_conv_config, resnet_v2=resnet_v2, strides=strides, logits_config=logits_config, name=name, **ResNet.CONFIGS[101], ) class ResNet152(ResNet): """ResNet152.""" def __init__( self, num_classes: int, bn_config: Optional[Mapping[str, FloatStrOrBool]] = None, resnet_v2: bool = False, logits_config: Optional[Mapping[str, Any]] = None, name: Optional[str] = None, initial_conv_config: Optional[Mapping[str, FloatStrOrBool]] = None, strides: Sequence[int] = (1, 2, 2, 2), ): """Constructs a ResNet model. Args: num_classes: The number of classes to classify the inputs into. bn_config: A dictionary of two elements, ``decay_rate`` and ``eps`` to be passed on to the :class:`~haiku.BatchNorm` layers. resnet_v2: Whether to use the v1 or v2 ResNet implementation. Defaults to ``False``. logits_config: A dictionary of keyword arguments for the logits layer. name: Name of the module. initial_conv_config: Keyword arguments passed to the constructor of the initial :class:`~haiku.Conv2D` module. strides: A sequence of length 4 that indicates the size of stride of convolutions for each block in each group. """ super().__init__( num_classes=num_classes, bn_config=bn_config, initial_conv_config=initial_conv_config, resnet_v2=resnet_v2, strides=strides, logits_config=logits_config, name=name, **ResNet.CONFIGS[152], ) class ResNet200(ResNet): """ResNet200.""" def __init__( self, num_classes: int, bn_config: Optional[Mapping[str, FloatStrOrBool]] = None, resnet_v2: bool = False, logits_config: Optional[Mapping[str, Any]] = None, name: Optional[str] = None, initial_conv_config: Optional[Mapping[str, FloatStrOrBool]] = None, strides: Sequence[int] = (1, 2, 2, 2), ): """Constructs a ResNet model. Args: num_classes: The number of classes to classify the inputs into. bn_config: A dictionary of two elements, ``decay_rate`` and ``eps`` to be passed on to the :class:`~haiku.BatchNorm` layers. resnet_v2: Whether to use the v1 or v2 ResNet implementation. Defaults to ``False``. logits_config: A dictionary of keyword arguments for the logits layer. name: Name of the module. initial_conv_config: Keyword arguments passed to the constructor of the initial :class:`~haiku.Conv2D` module. strides: A sequence of length 4 that indicates the size of stride of convolutions for each block in each group. """ super().__init__( num_classes=num_classes, bn_config=bn_config, initial_conv_config=initial_conv_config, resnet_v2=resnet_v2, strides=strides, logits_config=logits_config, name=name, **ResNet.CONFIGS[200], )
22,689
34.676101
167
py
big_transfer
big_transfer-master/bit_jax/models.py
# Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import jax import jax.numpy as jnp import flax.nn as nn def fixed_padding(x, kernel_size): pad_total = kernel_size - 1 pad_beg = pad_total // 2 pad_end = pad_total - pad_beg x = jax.lax.pad(x, 0.0, ((0, 0, 0), (pad_beg, pad_end, 0), (pad_beg, pad_end, 0), (0, 0, 0))) return x def standardize(x, axis, eps): x = x - jnp.mean(x, axis=axis, keepdims=True) x = x / jnp.sqrt(jnp.mean(jnp.square(x), axis=axis, keepdims=True) + eps) return x class GroupNorm(nn.Module): """Group normalization (arxiv.org/abs/1803.08494).""" def apply(self, x, num_groups=32): input_shape = x.shape group_shape = x.shape[:-1] + (num_groups, x.shape[-1] // num_groups) x = x.reshape(group_shape) # Standardize along spatial and group dimensions x = standardize(x, axis=[1, 2, 4], eps=1e-5) x = x.reshape(input_shape) bias_scale_shape = tuple([1, 1, 1] + [input_shape[-1]]) x = x * self.param('scale', bias_scale_shape, nn.initializers.ones) x = x + self.param('bias', bias_scale_shape, nn.initializers.zeros) return x class StdConv(nn.Conv): def param(self, name, shape, initializer): param = super().param(name, shape, initializer) if name == 'kernel': param = standardize(param, axis=[0, 1, 2], eps=1e-10) return param class RootBlock(nn.Module): def apply(self, x, width): x = fixed_padding(x, 7) x = StdConv(x, width, (7, 7), (2, 2), padding="VALID", bias=False, name="conv_root") x = fixed_padding(x, 3) x = nn.max_pool(x, (3, 3), strides=(2, 2), padding="VALID") return x class ResidualUnit(nn.Module): """Bottleneck ResNet block.""" def apply(self, x, nout, strides=(1, 1)): x_shortcut = x needs_projection = x.shape[-1] != nout * 4 or strides != (1, 1) group_norm = GroupNorm conv = StdConv.partial(bias=False) x = group_norm(x, name="gn1") x = nn.relu(x) if needs_projection: x_shortcut = conv(x, nout * 4, (1, 1), strides, name="conv_proj") x = conv(x, nout, (1, 1), name="conv1") x = group_norm(x, name="gn2") x = nn.relu(x) x = fixed_padding(x, 3) x = conv(x, nout, (3, 3), strides, name="conv2", padding='VALID') x = group_norm(x, name="gn3") x = nn.relu(x) x = conv(x, nout * 4, (1, 1), name="conv3") return x + x_shortcut class ResidualBlock(nn.Module): def apply(self, x, block_size, nout, first_stride): x = ResidualUnit( x, nout, strides=first_stride, name="unit01") for i in range(1, block_size): x = ResidualUnit( x, nout, strides=(1, 1), name=f"unit{i+1:02d}") return x class ResNet(nn.Module): """ResNetV2.""" def apply(self, x, num_classes=1000, width_factor=1, num_layers=50): block_sizes = _block_sizes[num_layers] width = 64 * width_factor root_block = RootBlock.partial(width=width) x = root_block(x, name='root_block') # Blocks for i, block_size in enumerate(block_sizes): x = ResidualBlock(x, block_size, width * 2 ** i, first_stride=(1, 1) if i == 0 else (2, 2), name=f"block{i + 1}") # Pre-head x = GroupNorm(x, name='norm-pre-head') x = nn.relu(x) x = jnp.mean(x, axis=(1, 2)) # Head x = nn.Dense(x, num_classes, name="conv_head", kernel_init=nn.initializers.zeros) return x.astype(jnp.float32) _block_sizes = { 50: [3, 4, 6, 3], 101: [3, 4, 23, 3], 152: [3, 8, 36, 3], } KNOWN_MODELS = dict( [(bit + f'-R{l}x{w}', ResNet.partial(num_layers=l, width_factor=w)) for bit in ['BiT-S', 'BiT-M'] for l, w in [(50, 1), (50, 3), (101, 1), (152, 2), (101, 3), (152, 4)]] )
4,397
25.493976
75
py
big_transfer
big_transfer-master/bit_jax/train.py
# Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import os from functools import partial import numpy as np import jax import jax.numpy as jnp import flax.optim as optim import flax.jax_utils as flax_utils import input_pipeline_tf2_or_jax as input_pipeline import bit_jax.models as models import bit_jax.tf2jax as tf2jax import bit_common import bit_hyperrule def main(args): logger = bit_common.setup_logger(args) logger.info(f'Available devices: {jax.devices()}') model = models.KNOWN_MODELS[args.model] # Load weigths of a BiT model bit_model_file = os.path.join(args.bit_pretrained_dir, f'{args.model}.npz') if not os.path.exists(bit_model_file): raise FileNotFoundError( f'Model file is not found in "{args.bit_pretrained_dir}" directory.') with open(bit_model_file, 'rb') as f: params_tf = np.load(f) params_tf = dict(zip(params_tf.keys(), params_tf.values())) resize_size, crop_size = bit_hyperrule.get_resolution_from_dataset( args.dataset) # Setup input pipeline dataset_info = input_pipeline.get_dataset_info( args.dataset, 'train', args.examples_per_class) data_train = input_pipeline.get_data( dataset=args.dataset, mode='train', repeats=None, batch_size=args.batch, resize_size=resize_size, crop_size=crop_size, examples_per_class=args.examples_per_class, examples_per_class_seed=args.examples_per_class_seed, mixup_alpha=bit_hyperrule.get_mixup(dataset_info['num_examples']), num_devices=jax.local_device_count(), tfds_manual_dir=args.tfds_manual_dir) logger.info(data_train) data_test = input_pipeline.get_data( dataset=args.dataset, mode='test', repeats=1, batch_size=args.batch_eval, resize_size=resize_size, crop_size=crop_size, examples_per_class=None, examples_per_class_seed=0, mixup_alpha=None, num_devices=jax.local_device_count(), tfds_manual_dir=args.tfds_manual_dir) logger.info(data_test) # Build ResNet architecture ResNet = model.partial(num_classes=dataset_info['num_classes']) _, params = ResNet.init_by_shape( jax.random.PRNGKey(0), [([1, crop_size, crop_size, 3], jnp.float32)]) resnet_fn = ResNet.call # pmap replicates the models over all GPUs resnet_fn_repl = jax.pmap(ResNet.call) def cross_entropy_loss(*, logits, labels): logp = jax.nn.log_softmax(logits) return -jnp.mean(jnp.sum(logp * labels, axis=1)) def loss_fn(params, images, labels): logits = resnet_fn(params, images) return cross_entropy_loss(logits=logits, labels=labels) # Update step, replicated over all GPUs @partial(jax.pmap, axis_name='batch') def update_fn(opt, lr, batch): l, g = jax.value_and_grad(loss_fn)(opt.target, batch['image'], batch['label']) g = jax.tree_map(lambda x: jax.lax.pmean(x, axis_name='batch'), g) opt = opt.apply_gradient(g, learning_rate=lr) return opt # In-place update of randomly initialized weights by BiT weigths tf2jax.transform_params(params, params_tf, num_classes=dataset_info['num_classes']) # Create optimizer and replicate it over all GPUs opt = optim.Momentum(beta=0.9).create(params) opt_repl = flax_utils.replicate(opt) # Delete referenes to the objects that are not needed anymore del opt del params total_steps = bit_hyperrule.get_schedule(dataset_info['num_examples'])[-1] # Run training loop for step, batch in zip(range(1, total_steps + 1), data_train.as_numpy_iterator()): lr = bit_hyperrule.get_lr(step - 1, dataset_info['num_examples'], args.base_lr) opt_repl = update_fn(opt_repl, flax_utils.replicate(lr), batch) # Run eval step if ((args.eval_every and step % args.eval_every == 0) or (step == total_steps)): accuracy_test = np.mean([ c for batch in data_test.as_numpy_iterator() for c in ( np.argmax(resnet_fn_repl(opt_repl.target, batch['image']), axis=2) == np.argmax(batch['label'], axis=2)).ravel()]) logger.info( f'Step: {step}, ' f'learning rate: {lr:.07f}, ' f'Test accuracy: {accuracy_test:0.3f}') if __name__ == "__main__": parser = bit_common.argparser(models.KNOWN_MODELS.keys()) parser.add_argument("--tfds_manual_dir", default=None, help="Path to maually downloaded dataset.") parser.add_argument("--batch_eval", default=32, type=int, help="Eval batch size.") main(parser.parse_args())
5,183
32.662338
79
py
big_transfer
big_transfer-master/bit_pytorch/models.py
# Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Lint as: python3 """Bottleneck ResNet v2 with GroupNorm and Weight Standardization.""" from collections import OrderedDict # pylint: disable=g-importing-member import torch import torch.nn as nn import torch.nn.functional as F class StdConv2d(nn.Conv2d): def forward(self, x): w = self.weight v, m = torch.var_mean(w, dim=[1, 2, 3], keepdim=True, unbiased=False) w = (w - m) / torch.sqrt(v + 1e-10) return F.conv2d(x, w, self.bias, self.stride, self.padding, self.dilation, self.groups) def conv3x3(cin, cout, stride=1, groups=1, bias=False): return StdConv2d(cin, cout, kernel_size=3, stride=stride, padding=1, bias=bias, groups=groups) def conv1x1(cin, cout, stride=1, bias=False): return StdConv2d(cin, cout, kernel_size=1, stride=stride, padding=0, bias=bias) def tf2th(conv_weights): """Possibly convert HWIO to OIHW.""" if conv_weights.ndim == 4: conv_weights = conv_weights.transpose([3, 2, 0, 1]) return torch.from_numpy(conv_weights) class PreActBottleneck(nn.Module): """Pre-activation (v2) bottleneck block. Follows the implementation of "Identity Mappings in Deep Residual Networks": https://github.com/KaimingHe/resnet-1k-layers/blob/master/resnet-pre-act.lua Except it puts the stride on 3x3 conv when available. """ def __init__(self, cin, cout=None, cmid=None, stride=1): super().__init__() cout = cout or cin cmid = cmid or cout//4 self.gn1 = nn.GroupNorm(32, cin) self.conv1 = conv1x1(cin, cmid) self.gn2 = nn.GroupNorm(32, cmid) self.conv2 = conv3x3(cmid, cmid, stride) # Original code has it on conv1!! self.gn3 = nn.GroupNorm(32, cmid) self.conv3 = conv1x1(cmid, cout) self.relu = nn.ReLU(inplace=True) if (stride != 1 or cin != cout): # Projection also with pre-activation according to paper. self.downsample = conv1x1(cin, cout, stride) def forward(self, x): out = self.relu(self.gn1(x)) # Residual branch residual = x if hasattr(self, 'downsample'): residual = self.downsample(out) # Unit's branch out = self.conv1(out) out = self.conv2(self.relu(self.gn2(out))) out = self.conv3(self.relu(self.gn3(out))) return out + residual def load_from(self, weights, prefix=''): convname = 'standardized_conv2d' with torch.no_grad(): self.conv1.weight.copy_(tf2th(weights[f'{prefix}a/{convname}/kernel'])) self.conv2.weight.copy_(tf2th(weights[f'{prefix}b/{convname}/kernel'])) self.conv3.weight.copy_(tf2th(weights[f'{prefix}c/{convname}/kernel'])) self.gn1.weight.copy_(tf2th(weights[f'{prefix}a/group_norm/gamma'])) self.gn2.weight.copy_(tf2th(weights[f'{prefix}b/group_norm/gamma'])) self.gn3.weight.copy_(tf2th(weights[f'{prefix}c/group_norm/gamma'])) self.gn1.bias.copy_(tf2th(weights[f'{prefix}a/group_norm/beta'])) self.gn2.bias.copy_(tf2th(weights[f'{prefix}b/group_norm/beta'])) self.gn3.bias.copy_(tf2th(weights[f'{prefix}c/group_norm/beta'])) if hasattr(self, 'downsample'): w = weights[f'{prefix}a/proj/{convname}/kernel'] self.downsample.weight.copy_(tf2th(w)) class ResNetV2(nn.Module): """Implementation of Pre-activation (v2) ResNet mode.""" def __init__(self, block_units, width_factor, head_size=21843, zero_head=False): super().__init__() wf = width_factor # shortcut 'cause we'll use it a lot. # The following will be unreadable if we split lines. # pylint: disable=line-too-long self.root = nn.Sequential(OrderedDict([ ('conv', StdConv2d(3, 64*wf, kernel_size=7, stride=2, padding=3, bias=False)), ('pad', nn.ConstantPad2d(1, 0)), ('pool', nn.MaxPool2d(kernel_size=3, stride=2, padding=0)), # The following is subtly not the same! # ('pool', nn.MaxPool2d(kernel_size=3, stride=2, padding=1)), ])) self.body = nn.Sequential(OrderedDict([ ('block1', nn.Sequential(OrderedDict( [('unit01', PreActBottleneck(cin=64*wf, cout=256*wf, cmid=64*wf))] + [(f'unit{i:02d}', PreActBottleneck(cin=256*wf, cout=256*wf, cmid=64*wf)) for i in range(2, block_units[0] + 1)], ))), ('block2', nn.Sequential(OrderedDict( [('unit01', PreActBottleneck(cin=256*wf, cout=512*wf, cmid=128*wf, stride=2))] + [(f'unit{i:02d}', PreActBottleneck(cin=512*wf, cout=512*wf, cmid=128*wf)) for i in range(2, block_units[1] + 1)], ))), ('block3', nn.Sequential(OrderedDict( [('unit01', PreActBottleneck(cin=512*wf, cout=1024*wf, cmid=256*wf, stride=2))] + [(f'unit{i:02d}', PreActBottleneck(cin=1024*wf, cout=1024*wf, cmid=256*wf)) for i in range(2, block_units[2] + 1)], ))), ('block4', nn.Sequential(OrderedDict( [('unit01', PreActBottleneck(cin=1024*wf, cout=2048*wf, cmid=512*wf, stride=2))] + [(f'unit{i:02d}', PreActBottleneck(cin=2048*wf, cout=2048*wf, cmid=512*wf)) for i in range(2, block_units[3] + 1)], ))), ])) # pylint: enable=line-too-long self.zero_head = zero_head self.head = nn.Sequential(OrderedDict([ ('gn', nn.GroupNorm(32, 2048*wf)), ('relu', nn.ReLU(inplace=True)), ('avg', nn.AdaptiveAvgPool2d(output_size=1)), ('conv', nn.Conv2d(2048*wf, head_size, kernel_size=1, bias=True)), ])) def forward(self, x): x = self.head(self.body(self.root(x))) assert x.shape[-2:] == (1, 1) # We should have no spatial shape left. return x[...,0,0] def load_from(self, weights, prefix='resnet/'): with torch.no_grad(): self.root.conv.weight.copy_(tf2th(weights[f'{prefix}root_block/standardized_conv2d/kernel'])) # pylint: disable=line-too-long self.head.gn.weight.copy_(tf2th(weights[f'{prefix}group_norm/gamma'])) self.head.gn.bias.copy_(tf2th(weights[f'{prefix}group_norm/beta'])) if self.zero_head: nn.init.zeros_(self.head.conv.weight) nn.init.zeros_(self.head.conv.bias) else: self.head.conv.weight.copy_(tf2th(weights[f'{prefix}head/conv2d/kernel'])) # pylint: disable=line-too-long self.head.conv.bias.copy_(tf2th(weights[f'{prefix}head/conv2d/bias'])) for bname, block in self.body.named_children(): for uname, unit in block.named_children(): unit.load_from(weights, prefix=f'{prefix}{bname}/{uname}/') KNOWN_MODELS = OrderedDict([ ('BiT-M-R50x1', lambda *a, **kw: ResNetV2([3, 4, 6, 3], 1, *a, **kw)), ('BiT-M-R50x3', lambda *a, **kw: ResNetV2([3, 4, 6, 3], 3, *a, **kw)), ('BiT-M-R101x1', lambda *a, **kw: ResNetV2([3, 4, 23, 3], 1, *a, **kw)), ('BiT-M-R101x3', lambda *a, **kw: ResNetV2([3, 4, 23, 3], 3, *a, **kw)), ('BiT-M-R152x2', lambda *a, **kw: ResNetV2([3, 8, 36, 3], 2, *a, **kw)), ('BiT-M-R152x4', lambda *a, **kw: ResNetV2([3, 8, 36, 3], 4, *a, **kw)), ('BiT-S-R50x1', lambda *a, **kw: ResNetV2([3, 4, 6, 3], 1, *a, **kw)), ('BiT-S-R50x3', lambda *a, **kw: ResNetV2([3, 4, 6, 3], 3, *a, **kw)), ('BiT-S-R101x1', lambda *a, **kw: ResNetV2([3, 4, 23, 3], 1, *a, **kw)), ('BiT-S-R101x3', lambda *a, **kw: ResNetV2([3, 4, 23, 3], 3, *a, **kw)), ('BiT-S-R152x2', lambda *a, **kw: ResNetV2([3, 8, 36, 3], 2, *a, **kw)), ('BiT-S-R152x4', lambda *a, **kw: ResNetV2([3, 8, 36, 3], 4, *a, **kw)), ])
7,962
40.691099
132
py
big_transfer
big_transfer-master/bit_pytorch/fewshot.py
# Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Lint as: python3 """Utility to find k-shot dataset indices, outputs the indices on stdout.""" #!/usr/bin/env python3 # coding: utf-8 from collections import * from functools import * import random import sys import torch import torchvision as tv class AddIndexIter(torch.utils.data.dataloader._SingleProcessDataLoaderIter): def _next_data(self): index = self._next_index() # may raise StopIteration data = self._dataset_fetcher.fetch(index) # may raise StopIteration if self._pin_memory: data = torch.utils.data._utils.pin_memory.pin_memory(data) return index, data def find_indices_loader(loader, n_shots, n_classes): per_label_indices = defaultdict(partial(deque, maxlen=n_shots)) for ibatch, (indices, (images, labels)) in enumerate(AddIndexIter(loader)): for idx, lbl in zip(indices, labels): per_label_indices[lbl.item()].append(idx) findings = sum(map(len, per_label_indices.values())) if findings == n_shots * n_classes: return per_label_indices raise RuntimeError("Unable to find enough examples!") def find_fewshot_indices(dataset, n_shots): n_classes = len(dataset.classes) orig_transform = dataset.transform dataset.transform = tv.transforms.Compose([ tv.transforms.CenterCrop(1), tv.transforms.ToTensor() ]) # TODO(lbeyer): if dataset isinstance DatasetFolder, we can (maybe?) do much better! loader = torch.utils.data.DataLoader(dataset, batch_size=1024, shuffle=True, num_workers=0) per_label_indices = find_indices_loader(loader, n_shots, n_classes) all_indices = [i for indices in per_label_indices.values() for i in indices] random.shuffle(all_indices) dataset.transform = orig_transform return all_indices if __name__ == "__main__": dataset = tv.datasets.ImageFolder(sys.argv[2], preprocess) all_indices = find_fewshot_indices(dataset, int(sys.argv[1])) for i in all_indices: print(i)
2,508
31.584416
93
py
big_transfer
big_transfer-master/bit_pytorch/train.py
# Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Lint as: python3 """Fine-tune a BiT model on some downstream dataset.""" #!/usr/bin/env python3 # coding: utf-8 from os.path import join as pjoin # pylint: disable=g-importing-member import time import numpy as np import torch import torchvision as tv import bit_pytorch.fewshot as fs import bit_pytorch.lbtoolbox as lb import bit_pytorch.models as models import bit_common import bit_hyperrule def topk(output, target, ks=(1,)): """Returns one boolean vector for each k, whether the target is within the output's top-k.""" _, pred = output.topk(max(ks), 1, True, True) pred = pred.t() correct = pred.eq(target.view(1, -1).expand_as(pred)) return [correct[:k].max(0)[0] for k in ks] def recycle(iterable): """Variant of itertools.cycle that does not save iterates.""" while True: for i in iterable: yield i def mktrainval(args, logger): """Returns train and validation datasets.""" precrop, crop = bit_hyperrule.get_resolution_from_dataset(args.dataset) train_tx = tv.transforms.Compose([ tv.transforms.Resize((precrop, precrop)), tv.transforms.RandomCrop((crop, crop)), tv.transforms.RandomHorizontalFlip(), tv.transforms.ToTensor(), tv.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), ]) val_tx = tv.transforms.Compose([ tv.transforms.Resize((crop, crop)), tv.transforms.ToTensor(), tv.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), ]) if args.dataset == "cifar10": train_set = tv.datasets.CIFAR10(args.datadir, transform=train_tx, train=True, download=True) valid_set = tv.datasets.CIFAR10(args.datadir, transform=val_tx, train=False, download=True) elif args.dataset == "cifar100": train_set = tv.datasets.CIFAR100(args.datadir, transform=train_tx, train=True, download=True) valid_set = tv.datasets.CIFAR100(args.datadir, transform=val_tx, train=False, download=True) elif args.dataset == "imagenet2012": train_set = tv.datasets.ImageFolder(pjoin(args.datadir, "train"), train_tx) valid_set = tv.datasets.ImageFolder(pjoin(args.datadir, "val"), val_tx) else: raise ValueError(f"Sorry, we have not spent time implementing the " f"{args.dataset} dataset in the PyTorch codebase. " f"In principle, it should be easy to add :)") if args.examples_per_class is not None: logger.info(f"Looking for {args.examples_per_class} images per class...") indices = fs.find_fewshot_indices(train_set, args.examples_per_class) train_set = torch.utils.data.Subset(train_set, indices=indices) logger.info(f"Using a training set with {len(train_set)} images.") logger.info(f"Using a validation set with {len(valid_set)} images.") micro_batch_size = args.batch // args.batch_split valid_loader = torch.utils.data.DataLoader( valid_set, batch_size=micro_batch_size, shuffle=False, num_workers=args.workers, pin_memory=True, drop_last=False) if micro_batch_size <= len(train_set): train_loader = torch.utils.data.DataLoader( train_set, batch_size=micro_batch_size, shuffle=True, num_workers=args.workers, pin_memory=True, drop_last=False) else: # In the few-shot cases, the total dataset size might be smaller than the batch-size. # In these cases, the default sampler doesn't repeat, so we need to make it do that # if we want to match the behaviour from the paper. train_loader = torch.utils.data.DataLoader( train_set, batch_size=micro_batch_size, num_workers=args.workers, pin_memory=True, sampler=torch.utils.data.RandomSampler(train_set, replacement=True, num_samples=micro_batch_size)) return train_set, valid_set, train_loader, valid_loader def run_eval(model, data_loader, device, chrono, logger, step): # switch to evaluate mode model.eval() logger.info("Running validation...") logger.flush() all_c, all_top1, all_top5 = [], [], [] end = time.time() for b, (x, y) in enumerate(data_loader): with torch.no_grad(): x = x.to(device, non_blocking=True) y = y.to(device, non_blocking=True) # measure data loading time chrono._done("eval load", time.time() - end) # compute output, measure accuracy and record loss. with chrono.measure("eval fprop"): logits = model(x) c = torch.nn.CrossEntropyLoss(reduction='none')(logits, y) top1, top5 = topk(logits, y, ks=(1, 5)) all_c.extend(c.cpu()) # Also ensures a sync point. all_top1.extend(top1.cpu()) all_top5.extend(top5.cpu()) # measure elapsed time end = time.time() model.train() logger.info(f"Validation@{step} loss {np.mean(all_c):.5f}, " f"top1 {np.mean(all_top1):.2%}, " f"top5 {np.mean(all_top5):.2%}") logger.flush() return all_c, all_top1, all_top5 def mixup_data(x, y, l): """Returns mixed inputs, pairs of targets, and lambda""" indices = torch.randperm(x.shape[0]).to(x.device) mixed_x = l * x + (1 - l) * x[indices] y_a, y_b = y, y[indices] return mixed_x, y_a, y_b def mixup_criterion(criterion, pred, y_a, y_b, l): return l * criterion(pred, y_a) + (1 - l) * criterion(pred, y_b) def main(args): logger = bit_common.setup_logger(args) # Lets cuDNN benchmark conv implementations and choose the fastest. # Only good if sizes stay the same within the main loop! torch.backends.cudnn.benchmark = True device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") logger.info(f"Going to train on {device}") train_set, valid_set, train_loader, valid_loader = mktrainval(args, logger) logger.info(f"Loading model from {args.model}.npz") model = models.KNOWN_MODELS[args.model](head_size=len(valid_set.classes), zero_head=True) model.load_from(np.load(f"{args.model}.npz")) logger.info("Moving model onto all GPUs") model = torch.nn.DataParallel(model) # Optionally resume from a checkpoint. # Load it to CPU first as we'll move the model to GPU later. # This way, we save a little bit of GPU memory when loading. step = 0 # Note: no weight-decay! optim = torch.optim.SGD(model.parameters(), lr=0.003, momentum=0.9) # Resume fine-tuning if we find a saved model. savename = pjoin(args.logdir, args.name, "bit.pth.tar") try: logger.info(f"Model will be saved in '{savename}'") checkpoint = torch.load(savename, map_location="cpu") logger.info(f"Found saved model to resume from at '{savename}'") step = checkpoint["step"] model.load_state_dict(checkpoint["model"]) optim.load_state_dict(checkpoint["optim"]) logger.info(f"Resumed at step {step}") except FileNotFoundError: logger.info("Fine-tuning from BiT") model = model.to(device) optim.zero_grad() model.train() mixup = bit_hyperrule.get_mixup(len(train_set)) cri = torch.nn.CrossEntropyLoss().to(device) logger.info("Starting training!") chrono = lb.Chrono() accum_steps = 0 mixup_l = np.random.beta(mixup, mixup) if mixup > 0 else 1 end = time.time() with lb.Uninterrupt() as u: for x, y in recycle(train_loader): # measure data loading time, which is spent in the `for` statement. chrono._done("load", time.time() - end) if u.interrupted: break # Schedule sending to GPU(s) x = x.to(device, non_blocking=True) y = y.to(device, non_blocking=True) # Update learning-rate, including stop training if over. lr = bit_hyperrule.get_lr(step, len(train_set), args.base_lr) if lr is None: break for param_group in optim.param_groups: param_group["lr"] = lr if mixup > 0.0: x, y_a, y_b = mixup_data(x, y, mixup_l) # compute output with chrono.measure("fprop"): logits = model(x) if mixup > 0.0: c = mixup_criterion(cri, logits, y_a, y_b, mixup_l) else: c = cri(logits, y) c_num = float(c.data.cpu().numpy()) # Also ensures a sync point. # Accumulate grads with chrono.measure("grads"): (c / args.batch_split).backward() accum_steps += 1 accstep = f" ({accum_steps}/{args.batch_split})" if args.batch_split > 1 else "" logger.info(f"[step {step}{accstep}]: loss={c_num:.5f} (lr={lr:.1e})") # pylint: disable=logging-format-interpolation logger.flush() # Update params if accum_steps == args.batch_split: with chrono.measure("update"): optim.step() optim.zero_grad() step += 1 accum_steps = 0 # Sample new mixup ratio for next batch mixup_l = np.random.beta(mixup, mixup) if mixup > 0 else 1 # Run evaluation and save the model. if args.eval_every and step % args.eval_every == 0: run_eval(model, valid_loader, device, chrono, logger, step) if args.save: torch.save({ "step": step, "model": model.state_dict(), "optim" : optim.state_dict(), }, savename) end = time.time() # Final eval at end of training. run_eval(model, valid_loader, device, chrono, logger, step='end') logger.info(f"Timings:\n{chrono}") if __name__ == "__main__": parser = bit_common.argparser(models.KNOWN_MODELS.keys()) parser.add_argument("--datadir", required=True, help="Path to the ImageNet data folder, preprocessed for torchvision.") parser.add_argument("--workers", type=int, default=8, help="Number of background threads used to load data.") parser.add_argument("--no-save", dest="save", action="store_false") main(parser.parse_args())
10,239
34.555556
124
py
big_transfer
big_transfer-master/bit_tf2/models.py
# Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ResNet architecture as used in BiT.""" import tensorflow.compat.v2 as tf from . import normalization def add_name_prefix(name, prefix=None): return prefix + "/" + name if prefix else name class ReLU(tf.keras.layers.ReLU): def compute_output_shape(self, input_shape): return tf.TensorShape(input_shape) class PaddingFromKernelSize(tf.keras.layers.Layer): """Layer that adds padding to an image taking into a given kernel size.""" def __init__(self, kernel_size, **kwargs): super(PaddingFromKernelSize, self).__init__(**kwargs) pad_total = kernel_size - 1 self._pad_beg = pad_total // 2 self._pad_end = pad_total - self._pad_beg def compute_output_shape(self, input_shape): batch_size, height, width, channels = tf.TensorShape(input_shape).as_list() if height is not None: height = height + self._pad_beg + self._pad_end if width is not None: width = width + self._pad_beg + self._pad_end return tf.TensorShape((batch_size, height, width, channels)) def call(self, x): padding = [ [0, 0], [self._pad_beg, self._pad_end], [self._pad_beg, self._pad_end], [0, 0]] return tf.pad(x, padding) class StandardizedConv2D(tf.keras.layers.Conv2D): """Implements the abs/1903.10520 technique (see go/dune-gn). You can simply replace any Conv2D with this one to use re-parametrized convolution operation in which the kernels are standardized before conv. Note that it does not come with extra learnable scale/bias parameters, as those used in "Weight normalization" (abs/1602.07868). This does not matter if combined with BN/GN/..., but it would matter if the convolution was used standalone. Author: Lucas Beyer """ def build(self, input_shape): super(StandardizedConv2D, self).build(input_shape) # Wrap a standardization around the conv OP. default_conv_op = self._convolution_op def standardized_conv_op(inputs, kernel): # Kernel has shape HWIO, normalize over HWI mean, var = tf.nn.moments(kernel, axes=[0, 1, 2], keepdims=True) # Author code uses std + 1e-5 return default_conv_op(inputs, (kernel - mean) / tf.sqrt(var + 1e-10)) self._convolution_op = standardized_conv_op self.built = True class BottleneckV2Unit(tf.keras.layers.Layer): """Implements a standard ResNet's unit (version 2). """ def __init__(self, num_filters, stride=1, **kwargs): """Initializer. Args: num_filters: number of filters in the bottleneck. stride: specifies block's stride. **kwargs: other tf.keras.layers.Layer keyword arguments. """ super(BottleneckV2Unit, self).__init__(**kwargs) self._num_filters = num_filters self._stride = stride self._proj = None self._unit_a = tf.keras.Sequential([ normalization.GroupNormalization(name="group_norm"), ReLU(), ], name="a") self._unit_a_conv = StandardizedConv2D( filters=num_filters, kernel_size=1, use_bias=False, padding="VALID", trainable=self.trainable, name="a/standardized_conv2d") self._unit_b = tf.keras.Sequential([ normalization.GroupNormalization(name="group_norm"), ReLU(), PaddingFromKernelSize(kernel_size=3), StandardizedConv2D( filters=num_filters, kernel_size=3, strides=stride, use_bias=False, padding="VALID", trainable=self.trainable, name="standardized_conv2d") ], name="b") self._unit_c = tf.keras.Sequential([ normalization.GroupNormalization(name="group_norm"), ReLU(), StandardizedConv2D( filters=4 * num_filters, kernel_size=1, use_bias=False, padding="VALID", trainable=self.trainable, name="standardized_conv2d") ], name="c") def build(self, input_shape): input_shape = tf.TensorShape(input_shape).as_list() # Add projection layer if necessary. if (self._stride > 1) or (4 * self._num_filters != input_shape[-1]): self._proj = StandardizedConv2D( filters=4 * self._num_filters, kernel_size=1, strides=self._stride, use_bias=False, padding="VALID", trainable=self.trainable, name="a/proj/standardized_conv2d") self.built = True def compute_output_shape(self, input_shape): current_shape = self._unit_a.compute_output_shape(input_shape) current_shape = self._unit_a_conv.compute_output_shape(current_shape) current_shape = self._unit_b.compute_output_shape(current_shape) current_shape = self._unit_c.compute_output_shape(current_shape) return current_shape def call(self, x): x_shortcut = x # Unit "a". x = self._unit_a(x) if self._proj is not None: x_shortcut = self._proj(x) x = self._unit_a_conv(x) # Unit "b". x = self._unit_b(x) # Unit "c". x = self._unit_c(x) return x + x_shortcut class ResnetV2(tf.keras.Model): """Generic ResnetV2 architecture, as used in the BiT paper.""" def __init__(self, num_units=(3, 4, 6, 3), num_outputs=1000, filters_factor=4, strides=(1, 2, 2, 2), **kwargs): super(ResnetV2, self).__init__(**kwargs) num_blocks = len(num_units) num_filters = tuple(16 * filters_factor * 2**b for b in range(num_blocks)) self._root = self._create_root_block(num_filters=num_filters[0]) self._blocks = [] for b, (f, u, s) in enumerate(zip(num_filters, num_units, strides), 1): n = "block{}".format(b) self._blocks.append( self._create_block(num_units=u, num_filters=f, stride=s, name=n)) self._pre_head = [ normalization.GroupNormalization(name="group_norm"), ReLU(), tf.keras.layers.GlobalAveragePooling2D() ] self._head = None if num_outputs: self._head = tf.keras.layers.Dense( units=num_outputs, use_bias=True, kernel_initializer="zeros", trainable=self.trainable, name="head/dense") def _create_root_block(self, num_filters, conv_size=7, conv_stride=2, pool_size=3, pool_stride=2): layers = [ PaddingFromKernelSize(conv_size), StandardizedConv2D( filters=num_filters, kernel_size=conv_size, strides=conv_stride, trainable=self.trainable, use_bias=False, name="standardized_conv2d"), PaddingFromKernelSize(pool_size), tf.keras.layers.MaxPool2D( pool_size=pool_size, strides=pool_stride, padding="valid") ] return tf.keras.Sequential(layers, name="root_block") def _create_block(self, num_units, num_filters, stride, name): layers = [] for i in range(1, num_units + 1): layers.append( BottleneckV2Unit( num_filters=num_filters, stride=(stride if i == 1 else 1), name="unit%02d" % i)) return tf.keras.Sequential(layers, name=name) def compute_output_shape(self, input_shape): current_shape = self._root.compute_output_shape(input_shape) for block in self._blocks: current_shape = block.compute_output_shape(current_shape) for layer in self._pre_head: current_shape = layer.compute_output_shape(current_shape) if self._head is not None: batch_size, features = current_shape.as_list() current_shape = (batch_size, 1, 1, features) current_shape = self._head.compute_output_shape(current_shape).as_list() current_shape = (current_shape[0], current_shape[3]) return tf.TensorShape(current_shape) def call(self, x): x = self._root(x) for block in self._blocks: x = block(x) for layer in self._pre_head: x = layer(x) if self._head is not None: x = self._head(x) return x KNOWN_MODELS = { f'{bit}-R{l}x{w}': f'gs://bit_models/{bit}-R{l}x{w}.h5' for bit in ['BiT-S', 'BiT-M'] for l, w in [(50, 1), (50, 3), (101, 1), (101, 3), (152, 4)] } NUM_UNITS = { k: (3, 4, 6, 3) if 'R50' in k else (3, 4, 23, 3) if 'R101' in k else (3, 8, 36, 3) for k in KNOWN_MODELS }
9,000
31.261649
79
py
big_transfer
big_transfer-master/bit_tf2/normalization.py
# Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """Group normalization.""" import tensorflow.compat.v2 as tf def group_normalize(x, gamma, beta, num_groups=None, group_size=None, eps=1e-5): """Applies group-normalization to NHWC `x` (see abs/1803.08494, go/dune-gn). This function just does the math, if you want a "layer" that creates the necessary variables etc., see `group_norm` below. You must either specify a fixed number of groups `num_groups`, which will automatically select a corresponding group size depending on the input's number of channels, or you must specify a `group_size`, which leads to an automatic number of groups depending on the input's number of channels. Args: x: N..C-tensor, the input to group-normalize. For images, this would be a NHWC-tensor, for time-series a NTC, for videos a NHWTC or NTHWC, all of them work, as normalization includes everything between N and C. Even just NC shape works, as C is grouped and normalized. gamma: tensor with C entries, learnable scale after normalization. beta: tensor with C entries, learnable bias after normalization. num_groups: int, number of groups to normalize over (divides C). group_size: int, size of the groups to normalize over (divides C). eps: float, a small additive constant to avoid /sqrt(0). Returns: Group-normalized `x`, of the same shape and type as `x`. Author: Lucas Beyer """ assert x.shape.ndims >= 2, ( "Less than 2-dim Tensor passed to GroupNorm. Something's fishy.") num_channels = x.shape[-1] assert num_channels is not None, "Cannot apply GroupNorm on dynamic channels." assert (num_groups is None) != (group_size is None), ( "You must specify exactly one of `num_groups`, `group_size`") if group_size is not None: num_groups = num_channels // group_size assert num_channels % num_groups == 0, ( "GroupNorm: {} not divisible by {}".format(num_channels, num_groups)) orig_shape = tf.shape(x) # This shape is NHWGS where G is #groups and S is group-size. extra_shape = [num_groups, num_channels // num_groups] group_shape = tf.concat([orig_shape[:-1], extra_shape], axis=-1) x = tf.reshape(x, group_shape) # The dimensions to normalize over: HWS for images, but more generally all # dimensions except N (batch, first) and G (cross-groups, next-to-last). # So more visually, normdims are the dots in N......G. (note the last one is # also a dot, not a full-stop, argh!) normdims = list(range(1, x.shape.ndims - 2)) + [x.shape.ndims - 1] mean, var = tf.nn.moments(x, normdims, keepdims=True) # Interestingly, we don't have a beta/gamma per group, but still one per # channel, at least according to the original paper. Reshape such that they # broadcast correctly. beta = tf.reshape(beta, extra_shape) gamma = tf.reshape(gamma, extra_shape) x = tf.nn.batch_normalization(x, mean, var, beta, gamma, eps) return tf.reshape(x, orig_shape) class GroupNormalization(tf.keras.layers.Layer): """A group-norm "layer" (see abs/1803.08494 go/dune-gn). This function creates beta/gamma variables in a name_scope, and uses them to apply `group_normalize` on the input `x`. You can either specify a fixed number of groups `num_groups`, which will automatically select a corresponding group size depending on the input's number of channels, or you must specify a `group_size`, which leads to an automatic number of groups depending on the input's number of channels. If you specify neither, the paper's recommended `num_groups=32` is used. Authors: Lucas Beyer, Joan Puigcerver. """ def __init__(self, num_groups=None, group_size=None, eps=1e-5, beta_init=tf.zeros_initializer(), gamma_init=tf.ones_initializer(), **kwargs): """Initializer. Args: num_groups: int, the number of channel-groups to normalize over. group_size: int, size of the groups to normalize over. eps: float, a small additive constant to avoid /sqrt(0). beta_init: initializer for bias, defaults to zeros. gamma_init: initializer for scale, defaults to ones. **kwargs: other tf.keras.layers.Layer arguments. """ super(GroupNormalization, self).__init__(**kwargs) if num_groups is None and group_size is None: num_groups = 32 self._num_groups = num_groups self._group_size = group_size self._eps = eps self._beta_init = beta_init self._gamma_init = gamma_init def build(self, input_size): channels = input_size[-1] assert channels is not None, "Cannot apply GN on dynamic channels." self._gamma = self.add_weight( name="gamma", shape=(channels,), initializer=self._gamma_init, dtype=self.dtype) self._beta = self.add_weight( name="beta", shape=(channels,), initializer=self._beta_init, dtype=self.dtype) super(GroupNormalization, self).build(input_size) def call(self, x): return group_normalize(x, self._gamma, self._beta, self._num_groups, self._group_size, self._eps)
5,692
39.664286
80
py
big_transfer
big_transfer-master/bit_tf2/train.py
# Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Lint as: python3 # coding: utf-8 from functools import partial import time import os import tensorflow.compat.v2 as tf tf.enable_v2_behavior() import bit_common import bit_hyperrule import bit_tf2.models as models import input_pipeline_tf2_or_jax as input_pipeline def reshape_for_keras(features, batch_size, crop_size): features["image"] = tf.reshape(features["image"], (batch_size, crop_size, crop_size, 3)) features["label"] = tf.reshape(features["label"], (batch_size, -1)) return (features["image"], features["label"]) class BiTLRSched(tf.keras.callbacks.Callback): def __init__(self, base_lr, num_samples): self.step = 0 self.base_lr = base_lr self.num_samples = num_samples def on_train_batch_begin(self, batch, logs=None): lr = bit_hyperrule.get_lr(self.step, self.num_samples, self.base_lr) tf.keras.backend.set_value(self.model.optimizer.lr, lr) self.step += 1 def main(args): tf.io.gfile.makedirs(args.logdir) logger = bit_common.setup_logger(args) logger.info(f'Available devices: {tf.config.list_physical_devices()}') tf.io.gfile.makedirs(args.bit_pretrained_dir) bit_model_file = os.path.join(args.bit_pretrained_dir, f'{args.model}.h5') if not tf.io.gfile.exists(bit_model_file): model_url = models.KNOWN_MODELS[args.model] logger.info(f'Downloading the model from {model_url}...') tf.io.gfile.copy(model_url, bit_model_file) # Set up input pipeline dataset_info = input_pipeline.get_dataset_info( args.dataset, 'train', args.examples_per_class) # Distribute training strategy = tf.distribute.MirroredStrategy() num_devices = strategy.num_replicas_in_sync print('Number of devices: {}'.format(num_devices)) resize_size, crop_size = bit_hyperrule.get_resolution_from_dataset(args.dataset) data_train = input_pipeline.get_data( dataset=args.dataset, mode='train', repeats=None, batch_size=args.batch, resize_size=resize_size, crop_size=crop_size, examples_per_class=args.examples_per_class, examples_per_class_seed=args.examples_per_class_seed, mixup_alpha=bit_hyperrule.get_mixup(dataset_info['num_examples']), num_devices=num_devices, tfds_manual_dir=args.tfds_manual_dir) data_test = input_pipeline.get_data( dataset=args.dataset, mode='test', repeats=1, batch_size=args.batch, resize_size=resize_size, crop_size=crop_size, examples_per_class=1, examples_per_class_seed=0, mixup_alpha=None, num_devices=num_devices, tfds_manual_dir=args.tfds_manual_dir) data_train = data_train.map(lambda x: reshape_for_keras( x, batch_size=args.batch, crop_size=crop_size)) data_test = data_test.map(lambda x: reshape_for_keras( x, batch_size=args.batch, crop_size=crop_size)) with strategy.scope(): filters_factor = int(args.model[-1])*4 model = models.ResnetV2( num_units=models.NUM_UNITS[args.model], num_outputs=21843, filters_factor=filters_factor, name="resnet", trainable=True, dtype=tf.float32) model.build((None, None, None, 3)) logger.info(f'Loading weights...') model.load_weights(bit_model_file) logger.info(f'Weights loaded into model!') model._head = tf.keras.layers.Dense( units=dataset_info['num_classes'], use_bias=True, kernel_initializer="zeros", trainable=True, name="head/dense") lr_supports = bit_hyperrule.get_schedule(dataset_info['num_examples']) schedule_length = lr_supports[-1] # NOTE: Let's not do that unless verified necessary and we do the same # across all three codebases. # schedule_length = schedule_length * 512 / args.batch optimizer = tf.keras.optimizers.SGD(momentum=0.9) loss_fn = tf.keras.losses.CategoricalCrossentropy(from_logits=True) model.compile(optimizer=optimizer, loss=loss_fn, metrics=['accuracy']) logger.info(f'Fine-tuning the model...') steps_per_epoch = args.eval_every or schedule_length history = model.fit( data_train, steps_per_epoch=steps_per_epoch, epochs=schedule_length // steps_per_epoch, validation_data=data_test, # here we are only using # this data to evaluate our performance callbacks=[BiTLRSched(args.base_lr, dataset_info['num_examples'])], ) for epoch, accu in enumerate(history.history['val_accuracy']): logger.info( f'Step: {epoch * args.eval_every}, ' f'Test accuracy: {accu:0.3f}') if __name__ == "__main__": parser = bit_common.argparser(models.KNOWN_MODELS.keys()) parser.add_argument("--tfds_manual_dir", default=None, help="Path to maually downloaded dataset.") parser.add_argument("--batch_eval", default=32, type=int, help="Eval batch size.") main(parser.parse_args())
5,417
34.181818
90
py
tirg
tirg-master/main.py
# Copyright 2019 Google Inc. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """Main method to train the model.""" #!/usr/bin/python import argparse import sys import time import datasets import img_text_composition_models import numpy as np from tensorboardX import SummaryWriter import test_retrieval import torch import torch.utils.data import torchvision from tqdm import tqdm as tqdm torch.set_num_threads(3) def parse_opt(): """Parses the input arguments.""" parser = argparse.ArgumentParser() parser.add_argument('-f', type=str, default='') parser.add_argument('--comment', type=str, default='test_notebook') parser.add_argument('--dataset', type=str, default='css3d') parser.add_argument( '--dataset_path', type=str, default='../imgcomsearch/CSSDataset/output') parser.add_argument('--model', type=str, default='tirg') parser.add_argument('--embed_dim', type=int, default=512) parser.add_argument('--learning_rate', type=float, default=1e-2) parser.add_argument( '--learning_rate_decay_frequency', type=int, default=9999999) parser.add_argument('--batch_size', type=int, default=32) parser.add_argument('--weight_decay', type=float, default=1e-6) parser.add_argument('--num_iters', type=int, default=210000) parser.add_argument('--loss', type=str, default='soft_triplet') parser.add_argument('--loader_num_workers', type=int, default=4) args = parser.parse_args() return args def load_dataset(opt): """Loads the input datasets.""" print('Reading dataset ', opt.dataset) if opt.dataset == 'css3d': trainset = datasets.CSSDataset( path=opt.dataset_path, split='train', transform=torchvision.transforms.Compose([ torchvision.transforms.ToTensor(), torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ])) testset = datasets.CSSDataset( path=opt.dataset_path, split='test', transform=torchvision.transforms.Compose([ torchvision.transforms.ToTensor(), torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ])) elif opt.dataset == 'fashion200k': trainset = datasets.Fashion200k( path=opt.dataset_path, split='train', transform=torchvision.transforms.Compose([ torchvision.transforms.Resize(224), torchvision.transforms.CenterCrop(224), torchvision.transforms.ToTensor(), torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ])) testset = datasets.Fashion200k( path=opt.dataset_path, split='test', transform=torchvision.transforms.Compose([ torchvision.transforms.Resize(224), torchvision.transforms.CenterCrop(224), torchvision.transforms.ToTensor(), torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ])) elif opt.dataset == 'mitstates': trainset = datasets.MITStates( path=opt.dataset_path, split='train', transform=torchvision.transforms.Compose([ torchvision.transforms.Resize(224), torchvision.transforms.CenterCrop(224), torchvision.transforms.ToTensor(), torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ])) testset = datasets.MITStates( path=opt.dataset_path, split='test', transform=torchvision.transforms.Compose([ torchvision.transforms.Resize(224), torchvision.transforms.CenterCrop(224), torchvision.transforms.ToTensor(), torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ])) else: print('Invalid dataset', opt.dataset) sys.exit() print('trainset size:', len(trainset)) print('testset size:', len(testset)) return trainset, testset def create_model_and_optimizer(opt, texts): """Builds the model and related optimizer.""" print('Creating model and optimizer for', opt.model) if opt.model == 'imgonly': model = img_text_composition_models.SimpleModelImageOnly( texts, embed_dim=opt.embed_dim) elif opt.model == 'textonly': model = img_text_composition_models.SimpleModelTextOnly( texts, embed_dim=opt.embed_dim) elif opt.model == 'concat': model = img_text_composition_models.Concat(texts, embed_dim=opt.embed_dim) elif opt.model == 'tirg': model = img_text_composition_models.TIRG(texts, embed_dim=opt.embed_dim) elif opt.model == 'tirg_lastconv': model = img_text_composition_models.TIRGLastConv( texts, embed_dim=opt.embed_dim) else: print('Invalid model', opt.model) print('available: imgonly, textonly, concat, tirg or tirg_lastconv') sys.exit() model = model.cuda() # create optimizer params = [] # low learning rate for pretrained layers on real image datasets if opt.dataset != 'css3d': params.append({ 'params': [p for p in model.img_model.fc.parameters()], 'lr': opt.learning_rate }) params.append({ 'params': [p for p in model.img_model.parameters()], 'lr': 0.1 * opt.learning_rate }) params.append({'params': [p for p in model.parameters()]}) for _, p1 in enumerate(params): # remove duplicated params for _, p2 in enumerate(params): if p1 is not p2: for p11 in p1['params']: for j, p22 in enumerate(p2['params']): if p11 is p22: p2['params'][j] = torch.tensor(0.0, requires_grad=True) optimizer = torch.optim.SGD( params, lr=opt.learning_rate, momentum=0.9, weight_decay=opt.weight_decay) return model, optimizer def train_loop(opt, logger, trainset, testset, model, optimizer): """Function for train loop""" print('Begin training') losses_tracking = {} it = 0 epoch = -1 tic = time.time() while it < opt.num_iters: epoch += 1 # show/log stats print('It', it, 'epoch', epoch, 'Elapsed time', round(time.time() - tic, 4), opt.comment) tic = time.time() for loss_name in losses_tracking: avg_loss = np.mean(losses_tracking[loss_name][-len(trainloader):]) print(' Loss', loss_name, round(avg_loss, 4)) logger.add_scalar(loss_name, avg_loss, it) logger.add_scalar('learning_rate', optimizer.param_groups[0]['lr'], it) # test if epoch % 3 == 1: tests = [] for name, dataset in [('train', trainset), ('test', testset)]: t = test_retrieval.test(opt, model, dataset) tests += [(name + ' ' + metric_name, metric_value) for metric_name, metric_value in t] for metric_name, metric_value in tests: logger.add_scalar(metric_name, metric_value, it) print(' ', metric_name, round(metric_value, 4)) # save checkpoint torch.save({ 'it': it, 'opt': opt, 'model_state_dict': model.state_dict(), }, logger.file_writer.get_logdir() + '/latest_checkpoint.pth') # run trainning for 1 epoch model.train() trainloader = trainset.get_loader( batch_size=opt.batch_size, shuffle=True, drop_last=True, num_workers=opt.loader_num_workers) def training_1_iter(data): assert type(data) is list img1 = np.stack([d['source_img_data'] for d in data]) img1 = torch.from_numpy(img1).float() img1 = torch.autograd.Variable(img1).cuda() img2 = np.stack([d['target_img_data'] for d in data]) img2 = torch.from_numpy(img2).float() img2 = torch.autograd.Variable(img2).cuda() mods = [str(d['mod']['str']) for d in data] # compute loss losses = [] if opt.loss == 'soft_triplet': loss_value = model.compute_loss( img1, mods, img2, soft_triplet_loss=True) elif opt.loss == 'batch_based_classification': loss_value = model.compute_loss( img1, mods, img2, soft_triplet_loss=False) else: print('Invalid loss function', opt.loss) sys.exit() loss_name = opt.loss loss_weight = 1.0 losses += [(loss_name, loss_weight, loss_value)] total_loss = sum([ loss_weight * loss_value for loss_name, loss_weight, loss_value in losses ]) assert not torch.isnan(total_loss) losses += [('total training loss', None, total_loss)] # track losses for loss_name, loss_weight, loss_value in losses: if loss_name not in losses_tracking: losses_tracking[loss_name] = [] losses_tracking[loss_name].append(float(loss_value)) # gradient descend optimizer.zero_grad() total_loss.backward() optimizer.step() for data in tqdm(trainloader, desc='Training for epoch ' + str(epoch)): it += 1 training_1_iter(data) # decay learing rate if it >= opt.learning_rate_decay_frequency and it % opt.learning_rate_decay_frequency == 0: for g in optimizer.param_groups: g['lr'] *= 0.1 print('Finished training') def main(): opt = parse_opt() print('Arguments:') for k in opt.__dict__.keys(): print(' ', k, ':', str(opt.__dict__[k])) logger = SummaryWriter(comment=opt.comment) print('Log files saved to', logger.file_writer.get_logdir()) for k in opt.__dict__.keys(): logger.add_text(k, str(opt.__dict__[k])) trainset, testset = load_dataset(opt) model, optimizer = create_model_and_optimizer(opt, trainset.get_all_texts()) train_loop(opt, logger, trainset, testset, model, optimizer) logger.close() if __name__ == '__main__': main()
10,540
34.372483
97
py
tirg
tirg-master/test_retrieval.py
# Copyright 2018 Google Inc. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """Evaluates the retrieval model.""" import numpy as np import torch from tqdm import tqdm as tqdm def test(opt, model, testset): """Tests a model over the given testset.""" model.eval() test_queries = testset.get_test_queries() all_imgs = [] all_captions = [] all_queries = [] all_target_captions = [] if test_queries: # compute test query features imgs = [] mods = [] for t in tqdm(test_queries): imgs += [testset.get_img(t['source_img_id'])] mods += [t['mod']['str']] if len(imgs) >= opt.batch_size or t is test_queries[-1]: if 'torch' not in str(type(imgs[0])): imgs = [torch.from_numpy(d).float() for d in imgs] imgs = torch.stack(imgs).float() imgs = torch.autograd.Variable(imgs).cuda() f = model.compose_img_text(imgs, mods).data.cpu().numpy() all_queries += [f] imgs = [] mods = [] all_queries = np.concatenate(all_queries) all_target_captions = [t['target_caption'] for t in test_queries] # compute all image features imgs = [] for i in tqdm(range(len(testset.imgs))): imgs += [testset.get_img(i)] if len(imgs) >= opt.batch_size or i == len(testset.imgs) - 1: if 'torch' not in str(type(imgs[0])): imgs = [torch.from_numpy(d).float() for d in imgs] imgs = torch.stack(imgs).float() imgs = torch.autograd.Variable(imgs).cuda() imgs = model.extract_img_feature(imgs).data.cpu().numpy() all_imgs += [imgs] imgs = [] all_imgs = np.concatenate(all_imgs) all_captions = [img['captions'][0] for img in testset.imgs] else: # use training queries to approximate training retrieval performance imgs0 = [] imgs = [] mods = [] for i in range(10000): item = testset[i] imgs += [item['source_img_data']] mods += [item['mod']['str']] if len(imgs) > opt.batch_size or i == 9999: imgs = torch.stack(imgs).float() imgs = torch.autograd.Variable(imgs) f = model.compose_img_text(imgs.cuda(), mods).data.cpu().numpy() all_queries += [f] imgs = [] mods = [] imgs0 += [item['target_img_data']] if len(imgs0) > opt.batch_size or i == 9999: imgs0 = torch.stack(imgs0).float() imgs0 = torch.autograd.Variable(imgs0) imgs0 = model.extract_img_feature(imgs0.cuda()).data.cpu().numpy() all_imgs += [imgs0] imgs0 = [] all_captions += [item['target_caption']] all_target_captions += [item['target_caption']] all_imgs = np.concatenate(all_imgs) all_queries = np.concatenate(all_queries) # feature normalization for i in range(all_queries.shape[0]): all_queries[i, :] /= np.linalg.norm(all_queries[i, :]) for i in range(all_imgs.shape[0]): all_imgs[i, :] /= np.linalg.norm(all_imgs[i, :]) # match test queries to target images, get nearest neighbors nn_result = [] for i in tqdm(range(all_queries.shape[0])): sims = all_queries[i:(i+1), :].dot(all_imgs.T) if test_queries: sims[0, test_queries[i]['source_img_id']] = -10e10 # remove query image nn_result.append(np.argsort(-sims[0, :])[:110]) # compute recalls out = [] nn_result = [[all_captions[nn] for nn in nns] for nns in nn_result] for k in [1, 5, 10, 50, 100]: r = 0.0 for i, nns in enumerate(nn_result): if all_target_captions[i] in nns[:k]: r += 1 r /= len(nn_result) out += [('recall_top' + str(k) + '_correct_composition', r)] if opt.dataset == 'mitstates': r = 0.0 for i, nns in enumerate(nn_result): if all_target_captions[i].split()[0] in [c.split()[0] for c in nns[:k]]: r += 1 r /= len(nn_result) out += [('recall_top' + str(k) + '_correct_adj', r)] r = 0.0 for i, nns in enumerate(nn_result): if all_target_captions[i].split()[1] in [c.split()[1] for c in nns[:k]]: r += 1 r /= len(nn_result) out += [('recall_top' + str(k) + '_correct_noun', r)] return out
4,735
34.343284
80
py
tirg
tirg-master/text_model.py
# Copyright 2018 Google Inc. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """Class for text data.""" import string import numpy as np import torch class SimpleVocab(object): def __init__(self): super(SimpleVocab, self).__init__() self.word2id = {} self.wordcount = {} self.word2id['<UNK>'] = 0 self.wordcount['<UNK>'] = 9e9 def tokenize_text(self, text): text = text.encode('ascii', 'ignore').decode('ascii') tokens = str(text).lower() tokens = tokens.translate(str.maketrans('','',string.punctuation)) tokens = tokens.strip().split() return tokens def build(self, texts): for text in texts: tokens = self.tokenize_text(text) for token in tokens: if token not in self.wordcount: self.wordcount[token] = 0 self.wordcount[token] += 1 for token in sorted(list(self.wordcount.keys())): if token not in self.word2id: self.word2id[token] = len(self.word2id) def threshold_rare_words(self, wordcount_threshold=5): for w in self.word2id: if self.wordcount[w] < wordcount_threshold: self.word2id[w] = 0 def encode_text(self, text): tokens = self.tokenize_text(text) x = [self.word2id.get(t, 0) for t in tokens] return x def get_size(self): return len(self.word2id) class TextLSTMModel(torch.nn.Module): def __init__(self, texts_to_build_vocab, word_embed_dim=512, lstm_hidden_dim=512): super(TextLSTMModel, self).__init__() self.vocab = SimpleVocab() self.vocab.build(texts_to_build_vocab) vocab_size = self.vocab.get_size() self.word_embed_dim = word_embed_dim self.lstm_hidden_dim = lstm_hidden_dim self.embedding_layer = torch.nn.Embedding(vocab_size, word_embed_dim) self.lstm = torch.nn.LSTM(word_embed_dim, lstm_hidden_dim) self.fc_output = torch.nn.Sequential( torch.nn.Dropout(p=0.1), torch.nn.Linear(lstm_hidden_dim, lstm_hidden_dim), ) def forward(self, x): """ input x: list of strings""" if type(x) is list: if type(x[0]) is str or type(x[0]) is unicode: x = [self.vocab.encode_text(text) for text in x] assert type(x) is list assert type(x[0]) is list assert type(x[0][0]) is int return self.forward_encoded_texts(x) def forward_encoded_texts(self, texts): # to tensor lengths = [len(t) for t in texts] itexts = torch.zeros((np.max(lengths), len(texts))).long() for i in range(len(texts)): itexts[:lengths[i], i] = torch.tensor(texts[i]) # embed words itexts = torch.autograd.Variable(itexts).cuda() etexts = self.embedding_layer(itexts) # lstm lstm_output, _ = self.forward_lstm_(etexts) # get last output (using length) text_features = [] for i in range(len(texts)): text_features.append(lstm_output[lengths[i] - 1, i, :]) # output text_features = torch.stack(text_features) text_features = self.fc_output(text_features) return text_features def forward_lstm_(self, etexts): batch_size = etexts.shape[1] first_hidden = (torch.zeros(1, batch_size, self.lstm_hidden_dim), torch.zeros(1, batch_size, self.lstm_hidden_dim)) first_hidden = (first_hidden[0].cuda(), first_hidden[1].cuda()) lstm_output, last_hidden = self.lstm(etexts, first_hidden) return lstm_output, last_hidden
4,013
30.359375
80
py
tirg
tirg-master/img_text_composition_models.py
# Copyright 2019 Google Inc. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """Models for Text and Image Composition.""" import numpy as np import torch import torchvision import torch.nn.functional as F import text_model import torch_functions class ConCatModule(torch.nn.Module): def __init__(self): super(ConCatModule, self).__init__() def forward(self, x): x = torch.cat(x, dim=1) return x class ImgTextCompositionBase(torch.nn.Module): """Base class for image + text composition.""" def __init__(self): super(ImgTextCompositionBase, self).__init__() self.normalization_layer = torch_functions.NormalizationLayer( normalize_scale=4.0, learn_scale=True) self.soft_triplet_loss = torch_functions.TripletLoss() def extract_img_feature(self, imgs): raise NotImplementedError def extract_text_feature(self, texts): raise NotImplementedError def compose_img_text(self, imgs, texts): raise NotImplementedError def compute_loss(self, imgs_query, modification_texts, imgs_target, soft_triplet_loss=True): mod_img1 = self.compose_img_text(imgs_query, modification_texts) mod_img1 = self.normalization_layer(mod_img1) img2 = self.extract_img_feature(imgs_target) img2 = self.normalization_layer(img2) assert (mod_img1.shape[0] == img2.shape[0] and mod_img1.shape[1] == img2.shape[1]) if soft_triplet_loss: return self.compute_soft_triplet_loss_(mod_img1, img2) else: return self.compute_batch_based_classification_loss_(mod_img1, img2) def compute_soft_triplet_loss_(self, mod_img1, img2): triplets = [] labels = range(mod_img1.shape[0]) + range(img2.shape[0]) for i in range(len(labels)): triplets_i = [] for j in range(len(labels)): if labels[i] == labels[j] and i != j: for k in range(len(labels)): if labels[i] != labels[k]: triplets_i.append([i, j, k]) np.random.shuffle(triplets_i) triplets += triplets_i[:3] assert (triplets and len(triplets) < 2000) return self.soft_triplet_loss(torch.cat([mod_img1, img2]), triplets) def compute_batch_based_classification_loss_(self, mod_img1, img2): x = torch.mm(mod_img1, img2.transpose(0, 1)) labels = torch.tensor(range(x.shape[0])).long() labels = torch.autograd.Variable(labels).cuda() return F.cross_entropy(x, labels) class ImgEncoderTextEncoderBase(ImgTextCompositionBase): """Base class for image and text encoder.""" def __init__(self, texts, embed_dim): super(ImgEncoderTextEncoderBase, self).__init__() # img model img_model = torchvision.models.resnet18(pretrained=True) class GlobalAvgPool2d(torch.nn.Module): def forward(self, x): return F.adaptive_avg_pool2d(x, (1, 1)) img_model.avgpool = GlobalAvgPool2d() img_model.fc = torch.nn.Sequential(torch.nn.Linear(512, embed_dim)) self.img_model = img_model # text model self.text_model = text_model.TextLSTMModel( texts_to_build_vocab=texts, word_embed_dim=embed_dim, lstm_hidden_dim=embed_dim) def extract_img_feature(self, imgs): return self.img_model(imgs) def extract_text_feature(self, texts): return self.text_model(texts) class SimpleModelImageOnly(ImgEncoderTextEncoderBase): def compose_img_text(self, imgs, texts): return self.extract_img_feature(imgs) class SimpleModelTextOnly(ImgEncoderTextEncoderBase): def compose_img_text(self, imgs, texts): return self.extract_text_feature(texts) class Concat(ImgEncoderTextEncoderBase): """Concatenation model.""" def __init__(self, texts, embed_dim): super(Concat, self).__init__(texts, embed_dim) # composer class Composer(torch.nn.Module): """Inner composer class.""" def __init__(self): super(Composer, self).__init__() self.m = torch.nn.Sequential( torch.nn.BatchNorm1d(2 * embed_dim), torch.nn.ReLU(), torch.nn.Linear(2 * embed_dim, 2 * embed_dim), torch.nn.BatchNorm1d(2 * embed_dim), torch.nn.ReLU(), torch.nn.Dropout(0.1), torch.nn.Linear(2 * embed_dim, embed_dim)) def forward(self, x): f = torch.cat(x, dim=1) f = self.m(f) return f self.composer = Composer() def compose_img_text(self, imgs, texts): img_features = self.extract_img_feature(imgs) text_features = self.extract_text_feature(texts) return self.compose_img_text_features(img_features, text_features) def compose_img_text_features(self, img_features, text_features): return self.composer((img_features, text_features)) class TIRG(ImgEncoderTextEncoderBase): """The TIGR model. The method is described in Nam Vo, Lu Jiang, Chen Sun, Kevin Murphy, Li-Jia Li, Li Fei-Fei, James Hays. "Composing Text and Image for Image Retrieval - An Empirical Odyssey" CVPR 2019. arXiv:1812.07119 """ def __init__(self, texts, embed_dim): super(TIRG, self).__init__(texts, embed_dim) self.a = torch.nn.Parameter(torch.tensor([1.0, 10.0, 1.0, 1.0])) self.gated_feature_composer = torch.nn.Sequential( ConCatModule(), torch.nn.BatchNorm1d(2 * embed_dim), torch.nn.ReLU(), torch.nn.Linear(2 * embed_dim, embed_dim)) self.res_info_composer = torch.nn.Sequential( ConCatModule(), torch.nn.BatchNorm1d(2 * embed_dim), torch.nn.ReLU(), torch.nn.Linear(2 * embed_dim, 2 * embed_dim), torch.nn.ReLU(), torch.nn.Linear(2 * embed_dim, embed_dim)) def compose_img_text(self, imgs, texts): img_features = self.extract_img_feature(imgs) text_features = self.extract_text_feature(texts) return self.compose_img_text_features(img_features, text_features) def compose_img_text_features(self, img_features, text_features): f1 = self.gated_feature_composer((img_features, text_features)) f2 = self.res_info_composer((img_features, text_features)) f = F.sigmoid(f1) * img_features * self.a[0] + f2 * self.a[1] return f class TIRGLastConv(ImgEncoderTextEncoderBase): """The TIGR model with spatial modification over the last conv layer. The method is described in Nam Vo, Lu Jiang, Chen Sun, Kevin Murphy, Li-Jia Li, Li Fei-Fei, James Hays. "Composing Text and Image for Image Retrieval - An Empirical Odyssey" CVPR 2019. arXiv:1812.07119 """ def __init__(self, texts, embed_dim): super(TIRGLastConv, self).__init__(texts, embed_dim) self.a = torch.nn.Parameter(torch.tensor([1.0, 10.0, 1.0, 1.0])) self.mod2d = torch.nn.Sequential( torch.nn.BatchNorm2d(512 + embed_dim), torch.nn.Conv2d(512 + embed_dim, 512 + embed_dim, [3, 3], padding=1), torch.nn.ReLU(), torch.nn.Conv2d(512 + embed_dim, 512, [3, 3], padding=1), ) self.mod2d_gate = torch.nn.Sequential( torch.nn.BatchNorm2d(512 + embed_dim), torch.nn.Conv2d(512 + embed_dim, 512 + embed_dim, [3, 3], padding=1), torch.nn.ReLU(), torch.nn.Conv2d(512 + embed_dim, 512, [3, 3], padding=1), ) def compose_img_text(self, imgs, texts): text_features = self.extract_text_feature(texts) x = imgs x = self.img_model.conv1(x) x = self.img_model.bn1(x) x = self.img_model.relu(x) x = self.img_model.maxpool(x) x = self.img_model.layer1(x) x = self.img_model.layer2(x) x = self.img_model.layer3(x) x = self.img_model.layer4(x) # mod y = text_features y = y.reshape((y.shape[0], y.shape[1], 1, 1)).repeat( 1, 1, x.shape[2], x.shape[3]) z = torch.cat((x, y), dim=1) t = self.mod2d(z) tgate = self.mod2d_gate(z) x = self.a[0] * F.sigmoid(tgate) * x + self.a[1] * t x = self.img_model.avgpool(x) x = x.view(x.size(0), -1) x = self.img_model.fc(x) return x
8,478
32.121094
80
py
tirg
tirg-master/torch_functions.py
# TODO(lujiang): put it into the third-party # MIT License # Copyright (c) 2018 Nam Vo # Permission is hereby granted, free of charge, to any person obtaining a copy # of this software and associated documentation files (the "Software"), to deal # in the Software without restriction, including without limitation the rights # to use, copy, modify, merge, publish, distribute, sublicense, and/or sell # copies of the Software, and to permit persons to whom the Software is # furnished to do so, subject to the following conditions: # The above copyright notice and this permission notice shall be included in all # copies or substantial portions of the Software. """Metric learning functions. Codes are modified from: https://github.com/lugiavn/generalization-dml/blob/master/nams.py """ import numpy as np import torch import torchvision def pairwise_distances(x, y=None): """ Input: x is a Nxd matrix y is an optional Mxd matirx Output: dist is a NxM matrix where dist[i,j] is the square norm between x[i,:] and y[j,:] if y is not given then use 'y=x'. i.e. dist[i,j] = ||x[i,:]-y[j,:]||^2 source: https://discuss.pytorch.org/t/efficient-distance-matrix-computation/9065/2 """ x_norm = (x**2).sum(1).view(-1, 1) if y is not None: y_t = torch.transpose(y, 0, 1) y_norm = (y**2).sum(1).view(1, -1) else: y_t = torch.transpose(x, 0, 1) y_norm = x_norm.view(1, -1) dist = x_norm + y_norm - 2.0 * torch.mm(x, y_t) # Ensure diagonal is zero if x=y # if y is None: # dist = dist - torch.diag(dist.diag) return torch.clamp(dist, 0.0, np.inf) class MyTripletLossFunc(torch.autograd.Function): def __init__(self, triplets): super(MyTripletLossFunc, self).__init__() self.triplets = triplets self.triplet_count = len(triplets) def forward(self, features): self.save_for_backward(features) self.distances = pairwise_distances(features).cpu().numpy() loss = 0.0 triplet_count = 0.0 correct_count = 0.0 for i, j, k in self.triplets: w = 1.0 triplet_count += w loss += w * np.log(1 + np.exp(self.distances[i, j] - self.distances[i, k])) if self.distances[i, j] < self.distances[i, k]: correct_count += 1 loss /= triplet_count return torch.FloatTensor((loss,)) def backward(self, grad_output): features, = self.saved_tensors features_np = features.cpu().numpy() grad_features = features.clone() * 0.0 grad_features_np = grad_features.cpu().numpy() for i, j, k in self.triplets: w = 1.0 f = 1.0 - 1.0 / ( 1.0 + np.exp(self.distances[i, j] - self.distances[i, k])) grad_features_np[i, :] += w * f * ( features_np[i, :] - features_np[j, :]) / self.triplet_count grad_features_np[j, :] += w * f * ( features_np[j, :] - features_np[i, :]) / self.triplet_count grad_features_np[i, :] += -w * f * ( features_np[i, :] - features_np[k, :]) / self.triplet_count grad_features_np[k, :] += -w * f * ( features_np[k, :] - features_np[i, :]) / self.triplet_count for i in range(features_np.shape[0]): grad_features[i, :] = torch.from_numpy(grad_features_np[i, :]) grad_features *= float(grad_output.data[0]) return grad_features class TripletLoss(torch.nn.Module): """Class for the triplet loss.""" def __init__(self, pre_layer=None): super(TripletLoss, self).__init__() self.pre_layer = pre_layer def forward(self, x, triplets): if self.pre_layer is not None: x = self.pre_layer(x) loss = MyTripletLossFunc(triplets)(x) return loss class NormalizationLayer(torch.nn.Module): """Class for normalization layer.""" def __init__(self, normalize_scale=1.0, learn_scale=True): super(NormalizationLayer, self).__init__() self.norm_s = float(normalize_scale) if learn_scale: self.norm_s = torch.nn.Parameter(torch.FloatTensor((self.norm_s,))) def forward(self, x): features = self.norm_s * x / torch.norm(x, dim=1, keepdim=True).expand_as(x) return features
4,133
31.046512
80
py
tirg
tirg-master/datasets.py
# Copyright 2019 Google Inc. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """Provides data for training and testing.""" import numpy as np import PIL import skimage.io import torch import json import torch.utils.data import torchvision import warnings import random class BaseDataset(torch.utils.data.Dataset): """Base class for a dataset.""" def __init__(self): super(BaseDataset, self).__init__() self.imgs = [] self.test_queries = [] def get_loader(self, batch_size, shuffle=False, drop_last=False, num_workers=0): return torch.utils.data.DataLoader( self, batch_size=batch_size, shuffle=shuffle, num_workers=num_workers, drop_last=drop_last, collate_fn=lambda i: i) def get_test_queries(self): return self.test_queries def get_all_texts(self): raise NotImplementedError def __getitem__(self, idx): return self.generate_random_query_target() def generate_random_query_target(self): raise NotImplementedError def get_img(self, idx, raw_img=False): raise NotImplementedError class CSSDataset(BaseDataset): """CSS dataset.""" def __init__(self, path, split='train', transform=None): super(CSSDataset, self).__init__() self.img_path = path + '/images/' self.transform = transform self.split = split self.data = np.load(path + '/css_toy_dataset_novel2_small.dup.npy').item() self.mods = self.data[self.split]['mods'] self.imgs = [] for objects in self.data[self.split]['objects_img']: label = len(self.imgs) if 'labels' in self.data[self.split]: label = self.data[self.split]['labels'][label] self.imgs += [{ 'objects': objects, 'label': label, 'captions': [str(label)] }] self.imgid2modtarget = {} for i in range(len(self.imgs)): self.imgid2modtarget[i] = [] for i, mod in enumerate(self.mods): for k in range(len(mod['from'])): f = mod['from'][k] t = mod['to'][k] self.imgid2modtarget[f] += [(i, t)] self.generate_test_queries_() def generate_test_queries_(self): test_queries = [] for mod in self.mods: for i, j in zip(mod['from'], mod['to']): test_queries += [{ 'source_img_id': i, 'target_caption': self.imgs[j]['captions'][0], 'mod': { 'str': mod['to_str'] } }] self.test_queries = test_queries def get_1st_training_query(self): i = np.random.randint(0, len(self.mods)) mod = self.mods[i] j = np.random.randint(0, len(mod['from'])) self.last_from = mod['from'][j] self.last_mod = [i] return mod['from'][j], i, mod['to'][j] def get_2nd_training_query(self): modid, new_to = random.choice(self.imgid2modtarget[self.last_from]) while modid in self.last_mod: modid, new_to = random.choice(self.imgid2modtarget[self.last_from]) self.last_mod += [modid] # mod = self.mods[modid] return self.last_from, modid, new_to def generate_random_query_target(self): try: if len(self.last_mod) < 2: img1id, modid, img2id = self.get_2nd_training_query() else: img1id, modid, img2id = self.get_1st_training_query() except: img1id, modid, img2id = self.get_1st_training_query() out = {} out['source_img_id'] = img1id out['source_img_data'] = self.get_img(img1id) out['target_img_id'] = img2id out['target_img_data'] = self.get_img(img2id) out['mod'] = {'id': modid, 'str': self.mods[modid]['to_str']} return out def __len__(self): return len(self.imgs) def get_all_texts(self): return [mod['to_str'] for mod in self.mods] def get_img(self, idx, raw_img=False, get_2d=False): """Gets CSS images.""" def generate_2d_image(objects): img = np.ones((64, 64, 3)) colortext2values = { 'gray': [87, 87, 87], 'red': [244, 35, 35], 'blue': [42, 75, 215], 'green': [29, 205, 20], 'brown': [129, 74, 25], 'purple': [129, 38, 192], 'cyan': [41, 208, 208], 'yellow': [255, 238, 51] } for obj in objects: s = 4.0 if obj['size'] == 'large': s *= 2 c = [0, 0, 0] for j in range(3): c[j] = 1.0 * colortext2values[obj['color']][j] / 255.0 y = obj['pos'][0] * img.shape[0] x = obj['pos'][1] * img.shape[1] if obj['shape'] == 'rectangle': img[int(y - s):int(y + s), int(x - s):int(x + s), :] = c if obj['shape'] == 'circle': for y0 in range(int(y - s), int(y + s) + 1): x0 = x + (abs(y0 - y) - s) x1 = 2 * x - x0 img[y0, int(x0):int(x1), :] = c if obj['shape'] == 'triangle': for y0 in range(int(y - s), int(y + s)): x0 = x + (y0 - y + s) / 2 x1 = 2 * x - x0 x0, x1 = min(x0, x1), max(x0, x1) img[y0, int(x0):int(x1), :] = c return img if self.img_path is None or get_2d: img = generate_2d_image(self.imgs[idx]['objects']) else: img_path = self.img_path + ('/css_%s_%06d.png' % (self.split, int(idx))) with open(img_path, 'rb') as f: img = PIL.Image.open(f) img = img.convert('RGB') if raw_img: return img if self.transform: img = self.transform(img) return img class Fashion200k(BaseDataset): """Fashion200k dataset.""" def __init__(self, path, split='train', transform=None): super(Fashion200k, self).__init__() self.split = split self.transform = transform self.img_path = path + '/' # get label files for the split label_path = path + '/labels/' from os import listdir from os.path import isfile from os.path import join label_files = [ f for f in listdir(label_path) if isfile(join(label_path, f)) ] label_files = [f for f in label_files if split in f] # read image info from label files self.imgs = [] def caption_post_process(s): return s.strip().replace('.', 'dotmark').replace('?', 'questionmark').replace( '&', 'andmark').replace('*', 'starmark') for filename in label_files: print('read ' + filename) with open(label_path + '/' + filename) as f: lines = f.readlines() for line in lines: line = line.split(' ') img = { 'file_path': line[0], 'detection_score': line[1], 'captions': [caption_post_process(line[2])], 'split': split, 'modifiable': False } self.imgs += [img] print('Fashion200k:', len(self.imgs), 'images') # generate query for training or testing if split == 'train': self.caption_index_init_() else: self.generate_test_queries_() def get_different_word(self, source_caption, target_caption): source_words = source_caption.split() target_words = target_caption.split() for source_word in source_words: if source_word not in target_words: break for target_word in target_words: if target_word not in source_words: break mod_str = 'replace ' + source_word + ' with ' + target_word return source_word, target_word, mod_str def generate_test_queries_(self): file2imgid = {} for i, img in enumerate(self.imgs): file2imgid[img['file_path']] = i with open(self.img_path + '/test_queries.txt') as f: lines = f.readlines() self.test_queries = [] for line in lines: source_file, target_file = line.split() idx = file2imgid[source_file] target_idx = file2imgid[target_file] source_caption = self.imgs[idx]['captions'][0] target_caption = self.imgs[target_idx]['captions'][0] source_word, target_word, mod_str = self.get_different_word( source_caption, target_caption) self.test_queries += [{ 'source_img_id': idx, 'source_caption': source_caption, 'target_caption': target_caption, 'mod': { 'str': mod_str } }] def caption_index_init_(self): """ index caption to generate training query-target example on the fly later""" # index caption 2 caption_id and caption 2 image_ids caption2id = {} id2caption = {} caption2imgids = {} for i, img in enumerate(self.imgs): for c in img['captions']: if c not in caption2id: id2caption[len(caption2id)] = c caption2id[c] = len(caption2id) caption2imgids[c] = [] caption2imgids[c].append(i) self.caption2imgids = caption2imgids print(len(caption2imgids), 'unique cations') # parent captions are 1-word shorter than their children parent2children_captions = {} for c in caption2id.keys(): for w in c.split(): p = c.replace(w, '') p = p.replace(' ', ' ').strip() if p not in parent2children_captions: parent2children_captions[p] = [] if c not in parent2children_captions[p]: parent2children_captions[p].append(c) self.parent2children_captions = parent2children_captions # identify parent captions for each image for img in self.imgs: img['modifiable'] = False img['parent_captions'] = [] for p in parent2children_captions: if len(parent2children_captions[p]) >= 2: for c in parent2children_captions[p]: for imgid in caption2imgids[c]: self.imgs[imgid]['modifiable'] = True self.imgs[imgid]['parent_captions'] += [p] num_modifiable_imgs = 0 for img in self.imgs: if img['modifiable']: num_modifiable_imgs += 1 print('Modifiable images', num_modifiable_imgs) def caption_index_sample_(self, idx): while not self.imgs[idx]['modifiable']: idx = np.random.randint(0, len(self.imgs)) # find random target image (same parent) img = self.imgs[idx] while True: p = random.choice(img['parent_captions']) c = random.choice(self.parent2children_captions[p]) if c not in img['captions']: break target_idx = random.choice(self.caption2imgids[c]) # find the word difference between query and target (not in parent caption) source_caption = self.imgs[idx]['captions'][0] target_caption = self.imgs[target_idx]['captions'][0] source_word, target_word, mod_str = self.get_different_word( source_caption, target_caption) return idx, target_idx, source_word, target_word, mod_str def get_all_texts(self): texts = [] for img in self.imgs: for c in img['captions']: texts.append(c) return texts def __len__(self): return len(self.imgs) def __getitem__(self, idx): idx, target_idx, source_word, target_word, mod_str = self.caption_index_sample_( idx) out = {} out['source_img_id'] = idx out['source_img_data'] = self.get_img(idx) out['source_caption'] = self.imgs[idx]['captions'][0] out['target_img_id'] = target_idx out['target_img_data'] = self.get_img(target_idx) out['target_caption'] = self.imgs[target_idx]['captions'][0] out['mod'] = {'str': mod_str} return out def get_img(self, idx, raw_img=False): img_path = self.img_path + self.imgs[idx]['file_path'] with open(img_path, 'rb') as f: img = PIL.Image.open(f) img = img.convert('RGB') if raw_img: return img if self.transform: img = self.transform(img) return img class MITStates(BaseDataset): """MITStates dataset.""" def __init__(self, path, split='train', transform=None): super(MITStates, self).__init__() self.path = path self.transform = transform self.split = split self.imgs = [] test_nouns = [ u'armor', u'bracelet', u'bush', u'camera', u'candy', u'castle', u'ceramic', u'cheese', u'clock', u'clothes', u'coffee', u'fan', u'fig', u'fish', u'foam', u'forest', u'fruit', u'furniture', u'garden', u'gate', u'glass', u'horse', u'island', u'laptop', u'lead', u'lightning', u'mirror', u'orange', u'paint', u'persimmon', u'plastic', u'plate', u'potato', u'road', u'rubber', u'sand', u'shell', u'sky', u'smoke', u'steel', u'stream', u'table', u'tea', u'tomato', u'vacuum', u'wax', u'wheel', u'window', u'wool' ] from os import listdir for f in listdir(path + '/images'): if ' ' not in f: continue adj, noun = f.split() if adj == 'adj': continue if split == 'train' and noun in test_nouns: continue if split == 'test' and noun not in test_nouns: continue for file_path in listdir(path + '/images/' + f): assert (file_path.endswith('jpg')) self.imgs += [{ 'file_path': path + '/images/' + f + '/' + file_path, 'captions': [f], 'adj': adj, 'noun': noun }] self.caption_index_init_() if split == 'test': self.generate_test_queries_() def get_all_texts(self): texts = [] for img in self.imgs: texts += img['captions'] return texts def __getitem__(self, idx): try: self.saved_item except: self.saved_item = None if self.saved_item is None: while True: idx, target_idx1 = self.caption_index_sample_(idx) idx, target_idx2 = self.caption_index_sample_(idx) if self.imgs[target_idx1]['adj'] != self.imgs[target_idx2]['adj']: break idx, target_idx = [idx, target_idx1] self.saved_item = [idx, target_idx2] else: idx, target_idx = self.saved_item self.saved_item = None mod_str = self.imgs[target_idx]['adj'] return { 'source_img_id': idx, 'source_img_data': self.get_img(idx), 'source_caption': self.imgs[idx]['captions'][0], 'target_img_id': target_idx, 'target_img_data': self.get_img(target_idx), 'target_caption': self.imgs[target_idx]['captions'][0], 'mod': { 'str': mod_str } } def caption_index_init_(self): self.caption2imgids = {} self.noun2adjs = {} for i, img in enumerate(self.imgs): cap = img['captions'][0] adj = img['adj'] noun = img['noun'] if cap not in self.caption2imgids.keys(): self.caption2imgids[cap] = [] if noun not in self.noun2adjs.keys(): self.noun2adjs[noun] = [] self.caption2imgids[cap].append(i) if adj not in self.noun2adjs[noun]: self.noun2adjs[noun].append(adj) for noun, adjs in self.noun2adjs.iteritems(): assert len(adjs) >= 2 def caption_index_sample_(self, idx): noun = self.imgs[idx]['noun'] # adj = self.imgs[idx]['adj'] target_adj = random.choice(self.noun2adjs[noun]) target_caption = target_adj + ' ' + noun target_idx = random.choice(self.caption2imgids[target_caption]) return idx, target_idx def generate_test_queries_(self): self.test_queries = [] for idx, img in enumerate(self.imgs): adj = img['adj'] noun = img['noun'] for target_adj in self.noun2adjs[noun]: if target_adj != adj: mod_str = target_adj self.test_queries += [{ 'source_img_id': idx, 'source_caption': adj + ' ' + noun, 'target_caption': target_adj + ' ' + noun, 'mod': { 'str': mod_str } }] print(len(self.test_queries), 'test queries') def __len__(self): return len(self.imgs) def get_img(self, idx, raw_img=False): img_path = self.imgs[idx]['file_path'] with open(img_path, 'rb') as f: img = PIL.Image.open(f) img = img.convert('RGB') if raw_img: return img if self.transform: img = self.transform(img) return img
16,642
30.520833
84
py
tirg
tirg-master/third_party/torch_functions.py
# Copyright (c) 2018 Nam Vo # Permission is hereby granted, free of charge, to any person obtaining a copy # of this software and associated documentation files (the "Software"), to deal # in the Software without restriction, including without limitation the rights # to use, copy, modify, merge, publish, distribute, sublicense, and/or sell # copies of the Software, and to permit persons to whom the Software is # furnished to do so, subject to the following conditions: # The above copyright notice and this permission notice shall be included in all # copies or substantial portions of the Software. """Metric learning functions. Codes are modified from: https://github.com/lugiavn/generalization-dml/blob/master/nams.py """ import numpy as np import torch import torchvision def pairwise_distances(x, y=None): """ Input: x is a Nxd matrix y is an optional Mxd matirx Output: dist is a NxM matrix where dist[i,j] is the square norm between x[i,:] and y[j,:] if y is not given then use 'y=x'. i.e. dist[i,j] = ||x[i,:]-y[j,:]||^2 source: https://discuss.pytorch.org/t/efficient-distance-matrix-computation/9065/2 """ x_norm = (x**2).sum(1).view(-1, 1) if y is not None: y_t = torch.transpose(y, 0, 1) y_norm = (y**2).sum(1).view(1, -1) else: y_t = torch.transpose(x, 0, 1) y_norm = x_norm.view(1, -1) dist = x_norm + y_norm - 2.0 * torch.mm(x, y_t) # Ensure diagonal is zero if x=y # if y is None: # dist = dist - torch.diag(dist.diag) return torch.clamp(dist, 0.0, np.inf) class MyTripletLossFunc(torch.autograd.Function): def __init__(self, triplets): super(MyTripletLossFunc, self).__init__() self.triplets = triplets self.triplet_count = len(triplets) def forward(self, features): self.save_for_backward(features) self.distances = pairwise_distances(features).cpu().numpy() loss = 0.0 triplet_count = 0.0 correct_count = 0.0 for i, j, k in self.triplets: w = 1.0 triplet_count += w loss += w * np.log(1 + np.exp(self.distances[i, j] - self.distances[i, k])) if self.distances[i, j] < self.distances[i, k]: correct_count += 1 loss /= triplet_count return torch.FloatTensor((loss,)) def backward(self, grad_output): features, = self.saved_tensors features_np = features.cpu().numpy() grad_features = features.clone() * 0.0 grad_features_np = grad_features.cpu().numpy() for i, j, k in self.triplets: w = 1.0 f = 1.0 - 1.0 / ( 1.0 + np.exp(self.distances[i, j] - self.distances[i, k])) grad_features_np[i, :] += w * f * ( features_np[i, :] - features_np[j, :]) / self.triplet_count grad_features_np[j, :] += w * f * ( features_np[j, :] - features_np[i, :]) / self.triplet_count grad_features_np[i, :] += -w * f * ( features_np[i, :] - features_np[k, :]) / self.triplet_count grad_features_np[k, :] += -w * f * ( features_np[k, :] - features_np[i, :]) / self.triplet_count for i in range(features_np.shape[0]): grad_features[i, :] = torch.from_numpy(grad_features_np[i, :]) grad_features *= float(grad_output.data[0]) return grad_features class TripletLoss(torch.nn.Module): """Class for the triplet loss.""" def __init__(self, pre_layer=None): super(TripletLoss, self).__init__() self.pre_layer = pre_layer def forward(self, x, triplets): if self.pre_layer is not None: x = self.pre_layer(x) loss = MyTripletLossFunc(triplets)(x) return loss class NormalizationLayer(torch.nn.Module): """Class for normalization layer.""" def __init__(self, normalize_scale=1.0, learn_scale=True): super(NormalizationLayer, self).__init__() self.norm_s = float(normalize_scale) if learn_scale: self.norm_s = torch.nn.Parameter(torch.FloatTensor((self.norm_s,))) def forward(self, x): features = self.norm_s * x / torch.norm(x, dim=1, keepdim=True).expand_as(x) return features
4,072
31.584
80
py
2s-AGCN
2s-AGCN-master/main.py
#!/usr/bin/env python from __future__ import print_function import argparse import inspect import os import pickle import random import shutil import time from collections import OrderedDict import numpy as np # torch import torch import torch.backends.cudnn as cudnn import torch.nn as nn import torch.optim as optim import yaml from tensorboardX import SummaryWriter from torch.autograd import Variable from torch.optim.lr_scheduler import _LRScheduler from tqdm import tqdm class GradualWarmupScheduler(_LRScheduler): def __init__(self, optimizer, total_epoch, after_scheduler=None): self.total_epoch = total_epoch self.after_scheduler = after_scheduler self.finished = False self.last_epoch = -1 super().__init__(optimizer) def get_lr(self): return [base_lr * (self.last_epoch + 1) / self.total_epoch for base_lr in self.base_lrs] def step(self, epoch=None, metric=None): if self.last_epoch >= self.total_epoch - 1: if metric is None: return self.after_scheduler.step(epoch) else: return self.after_scheduler.step(metric, epoch) else: return super(GradualWarmupScheduler, self).step(epoch) def init_seed(_): torch.cuda.manual_seed_all(1) torch.manual_seed(1) np.random.seed(1) random.seed(1) # torch.backends.cudnn.enabled = False torch.backends.cudnn.deterministic = True torch.backends.cudnn.benchmark = False def get_parser(): # parameter priority: command line > config > default parser = argparse.ArgumentParser( description='Spatial Temporal Graph Convolution Network') parser.add_argument( '--work-dir', default='./work_dir/temp', help='the work folder for storing results') parser.add_argument('-model_saved_name', default='') parser.add_argument( '--config', default='./config/nturgbd-cross-view/test_bone.yaml', help='path to the configuration file') # processor parser.add_argument( '--phase', default='train', help='must be train or test') parser.add_argument( '--save-score', type=str2bool, default=False, help='if ture, the classification score will be stored') # visulize and debug parser.add_argument( '--seed', type=int, default=1, help='random seed for pytorch') parser.add_argument( '--log-interval', type=int, default=100, help='the interval for printing messages (#iteration)') parser.add_argument( '--save-interval', type=int, default=2, help='the interval for storing models (#iteration)') parser.add_argument( '--eval-interval', type=int, default=5, help='the interval for evaluating models (#iteration)') parser.add_argument( '--print-log', type=str2bool, default=True, help='print logging or not') parser.add_argument( '--show-topk', type=int, default=[1, 5], nargs='+', help='which Top K accuracy will be shown') # feeder parser.add_argument( '--feeder', default='feeder.feeder', help='data loader will be used') parser.add_argument( '--num-worker', type=int, default=32, help='the number of worker for data loader') parser.add_argument( '--train-feeder-args', default=dict(), help='the arguments of data loader for training') parser.add_argument( '--test-feeder-args', default=dict(), help='the arguments of data loader for test') # model parser.add_argument('--model', default=None, help='the model will be used') parser.add_argument( '--model-args', type=dict, default=dict(), help='the arguments of model') parser.add_argument( '--weights', default=None, help='the weights for network initialization') parser.add_argument( '--ignore-weights', type=str, default=[], nargs='+', help='the name of weights which will be ignored in the initialization') # optim parser.add_argument( '--base-lr', type=float, default=0.01, help='initial learning rate') parser.add_argument( '--step', type=int, default=[20, 40, 60], nargs='+', help='the epoch where optimizer reduce the learning rate') parser.add_argument( '--device', type=int, default=0, nargs='+', help='the indexes of GPUs for training or testing') parser.add_argument('--optimizer', default='SGD', help='type of optimizer') parser.add_argument( '--nesterov', type=str2bool, default=False, help='use nesterov or not') parser.add_argument( '--batch-size', type=int, default=256, help='training batch size') parser.add_argument( '--test-batch-size', type=int, default=256, help='test batch size') parser.add_argument( '--start-epoch', type=int, default=0, help='start training from which epoch') parser.add_argument( '--num-epoch', type=int, default=80, help='stop training in which epoch') parser.add_argument( '--weight-decay', type=float, default=0.0005, help='weight decay for optimizer') parser.add_argument('--only_train_part', default=False) parser.add_argument('--only_train_epoch', default=0) parser.add_argument('--warm_up_epoch', default=0) return parser class Processor(): """ Processor for Skeleton-based Action Recgnition """ def __init__(self, arg): self.arg = arg self.save_arg() if arg.phase == 'train': if not arg.train_feeder_args['debug']: if os.path.isdir(arg.model_saved_name): print('log_dir: ', arg.model_saved_name, 'already exist') answer = input('delete it? y/n:') if answer == 'y': shutil.rmtree(arg.model_saved_name) print('Dir removed: ', arg.model_saved_name) input('Refresh the website of tensorboard by pressing any keys') else: print('Dir not removed: ', arg.model_saved_name) self.train_writer = SummaryWriter(os.path.join(arg.model_saved_name, 'train'), 'train') self.val_writer = SummaryWriter(os.path.join(arg.model_saved_name, 'val'), 'val') else: self.train_writer = self.val_writer = SummaryWriter(os.path.join(arg.model_saved_name, 'test'), 'test') self.global_step = 0 self.load_model() self.load_optimizer() self.load_data() self.lr = self.arg.base_lr self.best_acc = 0 def load_data(self): Feeder = import_class(self.arg.feeder) self.data_loader = dict() if self.arg.phase == 'train': self.data_loader['train'] = torch.utils.data.DataLoader( dataset=Feeder(**self.arg.train_feeder_args), batch_size=self.arg.batch_size, shuffle=True, num_workers=self.arg.num_worker, drop_last=True, worker_init_fn=init_seed) self.data_loader['test'] = torch.utils.data.DataLoader( dataset=Feeder(**self.arg.test_feeder_args), batch_size=self.arg.test_batch_size, shuffle=False, num_workers=self.arg.num_worker, drop_last=False, worker_init_fn=init_seed) def load_model(self): output_device = self.arg.device[0] if type(self.arg.device) is list else self.arg.device self.output_device = output_device Model = import_class(self.arg.model) shutil.copy2(inspect.getfile(Model), self.arg.work_dir) print(Model) self.model = Model(**self.arg.model_args).cuda(output_device) print(self.model) self.loss = nn.CrossEntropyLoss().cuda(output_device) if self.arg.weights: self.global_step = int(arg.weights[:-3].split('-')[-1]) self.print_log('Load weights from {}.'.format(self.arg.weights)) if '.pkl' in self.arg.weights: with open(self.arg.weights, 'r') as f: weights = pickle.load(f) else: weights = torch.load(self.arg.weights) weights = OrderedDict( [[k.split('module.')[-1], v.cuda(output_device)] for k, v in weights.items()]) keys = list(weights.keys()) for w in self.arg.ignore_weights: for key in keys: if w in key: if weights.pop(key, None) is not None: self.print_log('Sucessfully Remove Weights: {}.'.format(key)) else: self.print_log('Can Not Remove Weights: {}.'.format(key)) try: self.model.load_state_dict(weights) except: state = self.model.state_dict() diff = list(set(state.keys()).difference(set(weights.keys()))) print('Can not find these weights:') for d in diff: print(' ' + d) state.update(weights) self.model.load_state_dict(state) if type(self.arg.device) is list: if len(self.arg.device) > 1: self.model = nn.DataParallel( self.model, device_ids=self.arg.device, output_device=output_device) def load_optimizer(self): if self.arg.optimizer == 'SGD': self.optimizer = optim.SGD( self.model.parameters(), lr=self.arg.base_lr, momentum=0.9, nesterov=self.arg.nesterov, weight_decay=self.arg.weight_decay) elif self.arg.optimizer == 'Adam': self.optimizer = optim.Adam( self.model.parameters(), lr=self.arg.base_lr, weight_decay=self.arg.weight_decay) else: raise ValueError() lr_scheduler_pre = optim.lr_scheduler.MultiStepLR( self.optimizer, milestones=self.arg.step, gamma=0.1) self.lr_scheduler = GradualWarmupScheduler(self.optimizer, total_epoch=self.arg.warm_up_epoch, after_scheduler=lr_scheduler_pre) self.print_log('using warm up, epoch: {}'.format(self.arg.warm_up_epoch)) def save_arg(self): # save arg arg_dict = vars(self.arg) if not os.path.exists(self.arg.work_dir): os.makedirs(self.arg.work_dir) with open('{}/config.yaml'.format(self.arg.work_dir), 'w') as f: yaml.dump(arg_dict, f) def adjust_learning_rate(self, epoch): if self.arg.optimizer == 'SGD' or self.arg.optimizer == 'Adam': if epoch < self.arg.warm_up_epoch: lr = self.arg.base_lr * (epoch + 1) / self.arg.warm_up_epoch else: lr = self.arg.base_lr * ( 0.1 ** np.sum(epoch >= np.array(self.arg.step))) for param_group in self.optimizer.param_groups: param_group['lr'] = lr return lr else: raise ValueError() def print_time(self): localtime = time.asctime(time.localtime(time.time())) self.print_log("Local current time : " + localtime) def print_log(self, str, print_time=True): if print_time: localtime = time.asctime(time.localtime(time.time())) str = "[ " + localtime + ' ] ' + str print(str) if self.arg.print_log: with open('{}/log.txt'.format(self.arg.work_dir), 'a') as f: print(str, file=f) def record_time(self): self.cur_time = time.time() return self.cur_time def split_time(self): split_time = time.time() - self.cur_time self.record_time() return split_time def train(self, epoch, save_model=False): self.model.train() self.print_log('Training epoch: {}'.format(epoch + 1)) loader = self.data_loader['train'] self.adjust_learning_rate(epoch) # for name, param in self.model.named_parameters(): # self.train_writer.add_histogram(name, param.clone().cpu().data.numpy(), epoch) loss_value = [] self.train_writer.add_scalar('epoch', epoch, self.global_step) self.record_time() timer = dict(dataloader=0.001, model=0.001, statistics=0.001) process = tqdm(loader) if self.arg.only_train_part: if epoch > self.arg.only_train_epoch: print('only train part, require grad') for key, value in self.model.named_parameters(): if 'PA' in key: value.requires_grad = True # print(key + '-require grad') else: print('only train part, do not require grad') for key, value in self.model.named_parameters(): if 'PA' in key: value.requires_grad = False # print(key + '-not require grad') for batch_idx, (data, label, index) in enumerate(process): self.global_step += 1 # get data data = Variable(data.float().cuda(self.output_device), requires_grad=False) label = Variable(label.long().cuda(self.output_device), requires_grad=False) timer['dataloader'] += self.split_time() # forward output = self.model(data) # if batch_idx == 0 and epoch == 0: # self.train_writer.add_graph(self.model, output) if isinstance(output, tuple): output, l1 = output l1 = l1.mean() else: l1 = 0 loss = self.loss(output, label) + l1 # backward self.optimizer.zero_grad() loss.backward() self.optimizer.step() loss_value.append(loss.data.item()) timer['model'] += self.split_time() value, predict_label = torch.max(output.data, 1) acc = torch.mean((predict_label == label.data).float()) self.train_writer.add_scalar('acc', acc, self.global_step) self.train_writer.add_scalar('loss', loss.data.item(), self.global_step) self.train_writer.add_scalar('loss_l1', l1, self.global_step) # self.train_writer.add_scalar('batch_time', process.iterable.last_duration, self.global_step) # statistics self.lr = self.optimizer.param_groups[0]['lr'] self.train_writer.add_scalar('lr', self.lr, self.global_step) # if self.global_step % self.arg.log_interval == 0: # self.print_log( # '\tBatch({}/{}) done. Loss: {:.4f} lr:{:.6f}'.format( # batch_idx, len(loader), loss.data[0], lr)) timer['statistics'] += self.split_time() # statistics of time consumption and loss proportion = { k: '{:02d}%'.format(int(round(v * 100 / sum(timer.values())))) for k, v in timer.items() } self.print_log( '\tMean training loss: {:.4f}.'.format(np.mean(loss_value))) self.print_log( '\tTime consumption: [Data]{dataloader}, [Network]{model}'.format( **proportion)) if save_model: state_dict = self.model.state_dict() weights = OrderedDict([[k.split('module.')[-1], v.cpu()] for k, v in state_dict.items()]) torch.save(weights, self.arg.model_saved_name + '-' + str(epoch) + '-' + str(int(self.global_step)) + '.pt') def eval(self, epoch, save_score=False, loader_name=['test'], wrong_file=None, result_file=None): if wrong_file is not None: f_w = open(wrong_file, 'w') if result_file is not None: f_r = open(result_file, 'w') self.model.eval() self.print_log('Eval epoch: {}'.format(epoch + 1)) for ln in loader_name: loss_value = [] score_frag = [] right_num_total = 0 total_num = 0 loss_total = 0 step = 0 process = tqdm(self.data_loader[ln]) for batch_idx, (data, label, index) in enumerate(process): with torch.no_grad(): data = Variable( data.float().cuda(self.output_device), requires_grad=False, volatile=True) label = Variable( label.long().cuda(self.output_device), requires_grad=False, volatile=True) output = self.model(data) if isinstance(output, tuple): output, l1 = output l1 = l1.mean() else: l1 = 0 loss = self.loss(output, label) score_frag.append(output.data.cpu().numpy()) loss_value.append(loss.data.item()) _, predict_label = torch.max(output.data, 1) step += 1 if wrong_file is not None or result_file is not None: predict = list(predict_label.cpu().numpy()) true = list(label.data.cpu().numpy()) for i, x in enumerate(predict): if result_file is not None: f_r.write(str(x) + ',' + str(true[i]) + '\n') if x != true[i] and wrong_file is not None: f_w.write(str(index[i]) + ',' + str(x) + ',' + str(true[i]) + '\n') score = np.concatenate(score_frag) loss = np.mean(loss_value) accuracy = self.data_loader[ln].dataset.top_k(score, 1) if accuracy > self.best_acc: self.best_acc = accuracy # self.lr_scheduler.step(loss) print('Accuracy: ', accuracy, ' model: ', self.arg.model_saved_name) if self.arg.phase == 'train': self.val_writer.add_scalar('loss', loss, self.global_step) self.val_writer.add_scalar('loss_l1', l1, self.global_step) self.val_writer.add_scalar('acc', accuracy, self.global_step) score_dict = dict( zip(self.data_loader[ln].dataset.sample_name, score)) self.print_log('\tMean {} loss of {} batches: {}.'.format( ln, len(self.data_loader[ln]), np.mean(loss_value))) for k in self.arg.show_topk: self.print_log('\tTop{}: {:.2f}%'.format( k, 100 * self.data_loader[ln].dataset.top_k(score, k))) if save_score: with open('{}/epoch{}_{}_score.pkl'.format( self.arg.work_dir, epoch + 1, ln), 'wb') as f: pickle.dump(score_dict, f) def start(self): if self.arg.phase == 'train': self.print_log('Parameters:\n{}\n'.format(str(vars(self.arg)))) self.global_step = self.arg.start_epoch * len(self.data_loader['train']) / self.arg.batch_size for epoch in range(self.arg.start_epoch, self.arg.num_epoch): if self.lr < 1e-3: break save_model = ((epoch + 1) % self.arg.save_interval == 0) or ( epoch + 1 == self.arg.num_epoch) self.train(epoch, save_model=save_model) self.eval( epoch, save_score=self.arg.save_score, loader_name=['test']) print('best accuracy: ', self.best_acc, ' model_name: ', self.arg.model_saved_name) elif self.arg.phase == 'test': if not self.arg.test_feeder_args['debug']: wf = self.arg.model_saved_name + '_wrong.txt' rf = self.arg.model_saved_name + '_right.txt' else: wf = rf = None if self.arg.weights is None: raise ValueError('Please appoint --weights.') self.arg.print_log = False self.print_log('Model: {}.'.format(self.arg.model)) self.print_log('Weights: {}.'.format(self.arg.weights)) self.eval(epoch=0, save_score=self.arg.save_score, loader_name=['test'], wrong_file=wf, result_file=rf) self.print_log('Done.\n') def str2bool(v): if v.lower() in ('yes', 'true', 't', 'y', '1'): return True elif v.lower() in ('no', 'false', 'f', 'n', '0'): return False else: raise argparse.ArgumentTypeError('Boolean value expected.') def import_class(name): components = name.split('.') mod = __import__(components[0]) # import return model for comp in components[1:]: mod = getattr(mod, comp) return mod if __name__ == '__main__': parser = get_parser() # load arg form config file p = parser.parse_args() if p.config is not None: with open(p.config, 'r') as f: default_arg = yaml.load(f) key = vars(p).keys() for k in default_arg.keys(): if k not in key: print('WRONG ARG: {}'.format(k)) assert (k in key) parser.set_defaults(**default_arg) arg = parser.parse_args() init_seed(0) processor = Processor(arg) processor.start()
22,084
37.143351
120
py
2s-AGCN
2s-AGCN-master/model/agcn.py
import math import numpy as np import torch import torch.nn as nn from torch.autograd import Variable def import_class(name): components = name.split('.') mod = __import__(components[0]) for comp in components[1:]: mod = getattr(mod, comp) return mod def conv_branch_init(conv, branches): weight = conv.weight n = weight.size(0) k1 = weight.size(1) k2 = weight.size(2) nn.init.normal_(weight, 0, math.sqrt(2. / (n * k1 * k2 * branches))) nn.init.constant_(conv.bias, 0) def conv_init(conv): nn.init.kaiming_normal_(conv.weight, mode='fan_out') nn.init.constant_(conv.bias, 0) def bn_init(bn, scale): nn.init.constant_(bn.weight, scale) nn.init.constant_(bn.bias, 0) class unit_tcn(nn.Module): def __init__(self, in_channels, out_channels, kernel_size=9, stride=1): super(unit_tcn, self).__init__() pad = int((kernel_size - 1) / 2) self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=(kernel_size, 1), padding=(pad, 0), stride=(stride, 1)) self.bn = nn.BatchNorm2d(out_channels) self.relu = nn.ReLU() conv_init(self.conv) bn_init(self.bn, 1) def forward(self, x): x = self.bn(self.conv(x)) return x class unit_gcn(nn.Module): def __init__(self, in_channels, out_channels, A, coff_embedding=4, num_subset=3): super(unit_gcn, self).__init__() inter_channels = out_channels // coff_embedding self.inter_c = inter_channels self.PA = nn.Parameter(torch.from_numpy(A.astype(np.float32))) nn.init.constant_(self.PA, 1e-6) self.A = Variable(torch.from_numpy(A.astype(np.float32)), requires_grad=False) self.num_subset = num_subset self.conv_a = nn.ModuleList() self.conv_b = nn.ModuleList() self.conv_d = nn.ModuleList() for i in range(self.num_subset): self.conv_a.append(nn.Conv2d(in_channels, inter_channels, 1)) self.conv_b.append(nn.Conv2d(in_channels, inter_channels, 1)) self.conv_d.append(nn.Conv2d(in_channels, out_channels, 1)) if in_channels != out_channels: self.down = nn.Sequential( nn.Conv2d(in_channels, out_channels, 1), nn.BatchNorm2d(out_channels) ) else: self.down = lambda x: x self.bn = nn.BatchNorm2d(out_channels) self.soft = nn.Softmax(-2) self.relu = nn.ReLU() for m in self.modules(): if isinstance(m, nn.Conv2d): conv_init(m) elif isinstance(m, nn.BatchNorm2d): bn_init(m, 1) bn_init(self.bn, 1e-6) for i in range(self.num_subset): conv_branch_init(self.conv_d[i], self.num_subset) def forward(self, x): N, C, T, V = x.size() A = self.A.cuda(x.get_device()) A = A + self.PA y = None for i in range(self.num_subset): A1 = self.conv_a[i](x).permute(0, 3, 1, 2).contiguous().view(N, V, self.inter_c * T) A2 = self.conv_b[i](x).view(N, self.inter_c * T, V) A1 = self.soft(torch.matmul(A1, A2) / A1.size(-1)) # N V V A1 = A1 + A[i] A2 = x.view(N, C * T, V) z = self.conv_d[i](torch.matmul(A2, A1).view(N, C, T, V)) y = z + y if y is not None else z y = self.bn(y) y += self.down(x) return self.relu(y) class TCN_GCN_unit(nn.Module): def __init__(self, in_channels, out_channels, A, stride=1, residual=True): super(TCN_GCN_unit, self).__init__() self.gcn1 = unit_gcn(in_channels, out_channels, A) self.tcn1 = unit_tcn(out_channels, out_channels, stride=stride) self.relu = nn.ReLU() if not residual: self.residual = lambda x: 0 elif (in_channels == out_channels) and (stride == 1): self.residual = lambda x: x else: self.residual = unit_tcn(in_channels, out_channels, kernel_size=1, stride=stride) def forward(self, x): x = self.tcn1(self.gcn1(x)) + self.residual(x) return self.relu(x) class Model(nn.Module): def __init__(self, num_class=60, num_point=25, num_person=2, graph=None, graph_args=dict(), in_channels=3): super(Model, self).__init__() if graph is None: raise ValueError() else: Graph = import_class(graph) self.graph = Graph(**graph_args) A = self.graph.A self.data_bn = nn.BatchNorm1d(num_person * in_channels * num_point) self.l1 = TCN_GCN_unit(3, 64, A, residual=False) self.l2 = TCN_GCN_unit(64, 64, A) self.l3 = TCN_GCN_unit(64, 64, A) self.l4 = TCN_GCN_unit(64, 64, A) self.l5 = TCN_GCN_unit(64, 128, A, stride=2) self.l6 = TCN_GCN_unit(128, 128, A) self.l7 = TCN_GCN_unit(128, 128, A) self.l8 = TCN_GCN_unit(128, 256, A, stride=2) self.l9 = TCN_GCN_unit(256, 256, A) self.l10 = TCN_GCN_unit(256, 256, A) self.fc = nn.Linear(256, num_class) nn.init.normal_(self.fc.weight, 0, math.sqrt(2. / num_class)) bn_init(self.data_bn, 1) def forward(self, x): N, C, T, V, M = x.size() x = x.permute(0, 4, 3, 1, 2).contiguous().view(N, M * V * C, T) x = self.data_bn(x) x = x.view(N, M, V, C, T).permute(0, 1, 3, 4, 2).contiguous().view(N * M, C, T, V) x = self.l1(x) x = self.l2(x) x = self.l3(x) x = self.l4(x) x = self.l5(x) x = self.l6(x) x = self.l7(x) x = self.l8(x) x = self.l9(x) x = self.l10(x) # N*M,C,T,V c_new = x.size(1) x = x.view(N, M, c_new, -1) x = x.mean(3).mean(1) return self.fc(x)
5,882
30.972826
111
py
2s-AGCN
2s-AGCN-master/model/aagcn.py
import math import numpy as np import torch import torch.nn as nn from torch.autograd import Variable def import_class(name): components = name.split('.') mod = __import__(components[0]) for comp in components[1:]: mod = getattr(mod, comp) return mod def conv_branch_init(conv, branches): weight = conv.weight n = weight.size(0) k1 = weight.size(1) k2 = weight.size(2) nn.init.normal_(weight, 0, math.sqrt(2. / (n * k1 * k2 * branches))) nn.init.constant_(conv.bias, 0) def conv_init(conv): nn.init.kaiming_normal_(conv.weight, mode='fan_out') nn.init.constant_(conv.bias, 0) def bn_init(bn, scale): nn.init.constant_(bn.weight, scale) nn.init.constant_(bn.bias, 0) class unit_tcn(nn.Module): def __init__(self, in_channels, out_channels, kernel_size=9, stride=1): super(unit_tcn, self).__init__() pad = int((kernel_size - 1) / 2) self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=(kernel_size, 1), padding=(pad, 0), stride=(stride, 1)) self.bn = nn.BatchNorm2d(out_channels) self.relu = nn.ReLU(inplace=True) conv_init(self.conv) bn_init(self.bn, 1) def forward(self, x): x = self.bn(self.conv(x)) return x class unit_gcn(nn.Module): def __init__(self, in_channels, out_channels, A, coff_embedding=4, num_subset=3, adaptive=True, attention=True): super(unit_gcn, self).__init__() inter_channels = out_channels // coff_embedding self.inter_c = inter_channels self.out_c = out_channels self.in_c = in_channels self.num_subset = num_subset num_jpts = A.shape[-1] self.conv_d = nn.ModuleList() for i in range(self.num_subset): self.conv_d.append(nn.Conv2d(in_channels, out_channels, 1)) if adaptive: self.PA = nn.Parameter(torch.from_numpy(A.astype(np.float32))) self.alpha = nn.Parameter(torch.zeros(1)) # self.beta = nn.Parameter(torch.ones(1)) # nn.init.constant_(self.PA, 1e-6) # self.A = Variable(torch.from_numpy(A.astype(np.float32)), requires_grad=False) # self.A = self.PA self.conv_a = nn.ModuleList() self.conv_b = nn.ModuleList() for i in range(self.num_subset): self.conv_a.append(nn.Conv2d(in_channels, inter_channels, 1)) self.conv_b.append(nn.Conv2d(in_channels, inter_channels, 1)) else: self.A = Variable(torch.from_numpy(A.astype(np.float32)), requires_grad=False) self.adaptive = adaptive if attention: # self.beta = nn.Parameter(torch.zeros(1)) # self.gamma = nn.Parameter(torch.zeros(1)) # unified attention # self.Attention = nn.Parameter(torch.ones(num_jpts)) # temporal attention self.conv_ta = nn.Conv1d(out_channels, 1, 9, padding=4) nn.init.constant_(self.conv_ta.weight, 0) nn.init.constant_(self.conv_ta.bias, 0) # s attention ker_jpt = num_jpts - 1 if not num_jpts % 2 else num_jpts pad = (ker_jpt - 1) // 2 self.conv_sa = nn.Conv1d(out_channels, 1, ker_jpt, padding=pad) nn.init.xavier_normal_(self.conv_sa.weight) nn.init.constant_(self.conv_sa.bias, 0) # channel attention rr = 2 self.fc1c = nn.Linear(out_channels, out_channels // rr) self.fc2c = nn.Linear(out_channels // rr, out_channels) nn.init.kaiming_normal_(self.fc1c.weight) nn.init.constant_(self.fc1c.bias, 0) nn.init.constant_(self.fc2c.weight, 0) nn.init.constant_(self.fc2c.bias, 0) # self.bn = nn.BatchNorm2d(out_channels) # bn_init(self.bn, 1) self.attention = attention if in_channels != out_channels: self.down = nn.Sequential( nn.Conv2d(in_channels, out_channels, 1), nn.BatchNorm2d(out_channels) ) else: self.down = lambda x: x self.bn = nn.BatchNorm2d(out_channels) self.soft = nn.Softmax(-2) self.tan = nn.Tanh() self.sigmoid = nn.Sigmoid() self.relu = nn.ReLU(inplace=True) for m in self.modules(): if isinstance(m, nn.Conv2d): conv_init(m) elif isinstance(m, nn.BatchNorm2d): bn_init(m, 1) bn_init(self.bn, 1e-6) for i in range(self.num_subset): conv_branch_init(self.conv_d[i], self.num_subset) def forward(self, x): N, C, T, V = x.size() y = None if self.adaptive: A = self.PA # A = A + self.PA for i in range(self.num_subset): A1 = self.conv_a[i](x).permute(0, 3, 1, 2).contiguous().view(N, V, self.inter_c * T) A2 = self.conv_b[i](x).view(N, self.inter_c * T, V) A1 = self.tan(torch.matmul(A1, A2) / A1.size(-1)) # N V V A1 = A[i] + A1 * self.alpha A2 = x.view(N, C * T, V) z = self.conv_d[i](torch.matmul(A2, A1).view(N, C, T, V)) y = z + y if y is not None else z else: A = self.A.cuda(x.get_device()) * self.mask for i in range(self.num_subset): A1 = A[i] A2 = x.view(N, C * T, V) z = self.conv_d[i](torch.matmul(A2, A1).view(N, C, T, V)) y = z + y if y is not None else z y = self.bn(y) y += self.down(x) y = self.relu(y) if self.attention: # spatial attention se = y.mean(-2) # N C V se1 = self.sigmoid(self.conv_sa(se)) y = y * se1.unsqueeze(-2) + y # a1 = se1.unsqueeze(-2) # temporal attention se = y.mean(-1) se1 = self.sigmoid(self.conv_ta(se)) y = y * se1.unsqueeze(-1) + y # a2 = se1.unsqueeze(-1) # channel attention se = y.mean(-1).mean(-1) se1 = self.relu(self.fc1c(se)) se2 = self.sigmoid(self.fc2c(se1)) y = y * se2.unsqueeze(-1).unsqueeze(-1) + y # a3 = se2.unsqueeze(-1).unsqueeze(-1) # unified attention # y = y * self.Attention + y # y = y + y * ((a2 + a3) / 2) # y = self.bn(y) return y class TCN_GCN_unit(nn.Module): def __init__(self, in_channels, out_channels, A, stride=1, residual=True, adaptive=True, attention=True): super(TCN_GCN_unit, self).__init__() self.gcn1 = unit_gcn(in_channels, out_channels, A, adaptive=adaptive, attention=attention) self.tcn1 = unit_tcn(out_channels, out_channels, stride=stride) self.relu = nn.ReLU(inplace=True) # if attention: # self.alpha = nn.Parameter(torch.zeros(1)) # self.beta = nn.Parameter(torch.ones(1)) # temporal attention # self.conv_ta1 = nn.Conv1d(out_channels, out_channels//rt, 9, padding=4) # self.bn = nn.BatchNorm2d(out_channels) # bn_init(self.bn, 1) # self.conv_ta2 = nn.Conv1d(out_channels, 1, 9, padding=4) # nn.init.kaiming_normal_(self.conv_ta1.weight) # nn.init.constant_(self.conv_ta1.bias, 0) # nn.init.constant_(self.conv_ta2.weight, 0) # nn.init.constant_(self.conv_ta2.bias, 0) # rt = 4 # self.inter_c = out_channels // rt # self.conv_ta1 = nn.Conv2d(out_channels, out_channels // rt, 1) # self.conv_ta2 = nn.Conv2d(out_channels, out_channels // rt, 1) # nn.init.constant_(self.conv_ta1.weight, 0) # nn.init.constant_(self.conv_ta1.bias, 0) # nn.init.constant_(self.conv_ta2.weight, 0) # nn.init.constant_(self.conv_ta2.bias, 0) # s attention # num_jpts = A.shape[-1] # ker_jpt = num_jpts - 1 if not num_jpts % 2 else num_jpts # pad = (ker_jpt - 1) // 2 # self.conv_sa = nn.Conv1d(out_channels, 1, ker_jpt, padding=pad) # nn.init.constant_(self.conv_sa.weight, 0) # nn.init.constant_(self.conv_sa.bias, 0) # channel attention # rr = 16 # self.fc1c = nn.Linear(out_channels, out_channels // rr) # self.fc2c = nn.Linear(out_channels // rr, out_channels) # nn.init.kaiming_normal_(self.fc1c.weight) # nn.init.constant_(self.fc1c.bias, 0) # nn.init.constant_(self.fc2c.weight, 0) # nn.init.constant_(self.fc2c.bias, 0) # # self.softmax = nn.Softmax(-2) # self.sigmoid = nn.Sigmoid() self.attention = attention if not residual: self.residual = lambda x: 0 elif (in_channels == out_channels) and (stride == 1): self.residual = lambda x: x else: self.residual = unit_tcn(in_channels, out_channels, kernel_size=1, stride=stride) def forward(self, x): if self.attention: y = self.relu(self.tcn1(self.gcn1(x)) + self.residual(x)) # spatial attention # se = y.mean(-2) # N C V # se1 = self.sigmoid(self.conv_sa(se)) # y = y * se1.unsqueeze(-2) + y # a1 = se1.unsqueeze(-2) # temporal attention # se = y.mean(-1) # N C T # # se1 = self.relu(self.bn(self.conv_ta1(se))) # se2 = self.sigmoid(self.conv_ta2(se)) # # y = y * se1.unsqueeze(-1) + y # a2 = se2.unsqueeze(-1) # se = y # NCTV # N, C, T, V = y.shape # se1 = self.conv_ta1(se).permute(0, 2, 1, 3).contiguous().view(N, T, self.inter_c * V) # NTCV # se2 = self.conv_ta2(se).permute(0, 1, 3, 2).contiguous().view(N, self.inter_c * V, T) # NCVT # a2 = self.softmax(torch.matmul(se1, se2) / np.sqrt(se1.size(-1))) # N T T # y = torch.matmul(y.permute(0, 1, 3, 2).contiguous().view(N, C * V, T), a2) \ # .view(N, C, V, T).permute(0, 1, 3, 2) * self.alpha + y # channel attention # se = y.mean(-1).mean(-1) # se1 = self.relu(self.fc1c(se)) # se2 = self.sigmoid(self.fc2c(se1)) # # y = y * se2.unsqueeze(-1).unsqueeze(-1) + y # a3 = se2.unsqueeze(-1).unsqueeze(-1) # # y = y * ((a2 + a3) / 2) + y # y = self.bn(y) else: y = self.relu(self.tcn1(self.gcn1(x)) + self.residual(x)) return y class Model(nn.Module): def __init__(self, num_class=60, num_point=25, num_person=2, graph=None, graph_args=dict(), in_channels=3, drop_out=0, adaptive=True, attention=True): super(Model, self).__init__() if graph is None: raise ValueError() else: Graph = import_class(graph) self.graph = Graph(**graph_args) A = self.graph.A self.num_class = num_class self.data_bn = nn.BatchNorm1d(num_person * in_channels * num_point) self.l1 = TCN_GCN_unit(3, 64, A, residual=False, adaptive=adaptive, attention=attention) self.l2 = TCN_GCN_unit(64, 64, A, adaptive=adaptive, attention=attention) self.l3 = TCN_GCN_unit(64, 64, A, adaptive=adaptive, attention=attention) self.l4 = TCN_GCN_unit(64, 64, A, adaptive=adaptive, attention=attention) self.l5 = TCN_GCN_unit(64, 128, A, stride=2, adaptive=adaptive, attention=attention) self.l6 = TCN_GCN_unit(128, 128, A, adaptive=adaptive, attention=attention) self.l7 = TCN_GCN_unit(128, 128, A, adaptive=adaptive, attention=attention) self.l8 = TCN_GCN_unit(128, 256, A, stride=2, adaptive=adaptive, attention=attention) self.l9 = TCN_GCN_unit(256, 256, A, adaptive=adaptive, attention=attention) self.l10 = TCN_GCN_unit(256, 256, A, adaptive=adaptive, attention=attention) self.fc = nn.Linear(256, num_class) nn.init.normal_(self.fc.weight, 0, math.sqrt(2. / num_class)) bn_init(self.data_bn, 1) if drop_out: self.drop_out = nn.Dropout(drop_out) else: self.drop_out = lambda x: x def forward(self, x): N, C, T, V, M = x.size() x = x.permute(0, 4, 3, 1, 2).contiguous().view(N, M * V * C, T) x = self.data_bn(x) x = x.view(N, M, V, C, T).permute(0, 1, 3, 4, 2).contiguous().view(N * M, C, T, V) x = self.l1(x) x = self.l2(x) x = self.l3(x) x = self.l4(x) x = self.l5(x) x = self.l6(x) x = self.l7(x) x = self.l8(x) x = self.l9(x) x = self.l10(x) # N*M,C,T,V c_new = x.size(1) x = x.view(N, M, c_new, -1) x = x.mean(3).mean(1) x = self.drop_out(x) return self.fc(x)
13,013
36.504323
116
py
2s-AGCN
2s-AGCN-master/data_gen/kinetics_gendata.py
import argparse import os import numpy as np import json from torch.utils.data import Dataset import pickle from tqdm import tqdm num_joint = 18 max_frame = 300 num_person_out = 2 num_person_in = 5 class Feeder_kinetics(Dataset): """ Feeder for skeleton-based action recognition in kinetics-skeleton dataset # Joint index: # {0, "Nose"} # {1, "Neck"}, # {2, "RShoulder"}, # {3, "RElbow"}, # {4, "RWrist"}, # {5, "LShoulder"}, # {6, "LElbow"}, # {7, "LWrist"}, # {8, "RHip"}, # {9, "RKnee"}, # {10, "RAnkle"}, # {11, "LHip"}, # {12, "LKnee"}, # {13, "LAnkle"}, # {14, "REye"}, # {15, "LEye"}, # {16, "REar"}, # {17, "LEar"}, Arguments: data_path: the path to '.npy' data, the shape of data should be (N, C, T, V, M) label_path: the path to label window_size: The length of the output sequence num_person_in: The number of people the feeder can observe in the input sequence num_person_out: The number of people the feeder in the output sequence debug: If true, only use the first 100 samples """ def __init__(self, data_path, label_path, ignore_empty_sample=True, window_size=-1, num_person_in=5, num_person_out=2): self.data_path = data_path self.label_path = label_path self.window_size = window_size self.num_person_in = num_person_in self.num_person_out = num_person_out self.ignore_empty_sample = ignore_empty_sample self.load_data() def load_data(self): # load file list self.sample_name = os.listdir(self.data_path) # load label label_path = self.label_path with open(label_path) as f: label_info = json.load(f) sample_id = [name.split('.')[0] for name in self.sample_name] self.label = np.array([label_info[id]['label_index'] for id in sample_id]) has_skeleton = np.array([label_info[id]['has_skeleton'] for id in sample_id]) # ignore the samples which does not has skeleton sequence if self.ignore_empty_sample: self.sample_name = [s for h, s in zip(has_skeleton, self.sample_name) if h] self.label = self.label[has_skeleton] # output data shape (N, C, T, V, M) self.N = len(self.sample_name) # sample self.C = 3 # channel self.T = max_frame # frame self.V = num_joint # joint self.M = self.num_person_out # person def __len__(self): return len(self.sample_name) def __iter__(self): return self def __getitem__(self, index): # output shape (C, T, V, M) # get data sample_name = self.sample_name[index] sample_path = os.path.join(self.data_path, sample_name) with open(sample_path, 'r') as f: video_info = json.load(f) # fill data_numpy data_numpy = np.zeros((self.C, self.T, self.V, self.num_person_in)) for frame_info in video_info['data']: frame_index = frame_info['frame_index'] for m, skeleton_info in enumerate(frame_info["skeleton"]): if m >= self.num_person_in: break pose = skeleton_info['pose'] score = skeleton_info['score'] data_numpy[0, frame_index, :, m] = pose[0::2] data_numpy[1, frame_index, :, m] = pose[1::2] data_numpy[2, frame_index, :, m] = score # centralization data_numpy[0:2] = data_numpy[0:2] - 0.5 data_numpy[1:2] = -data_numpy[1:2] data_numpy[0][data_numpy[2] == 0] = 0 data_numpy[1][data_numpy[2] == 0] = 0 # get & check label index label = video_info['label_index'] assert (self.label[index] == label) # sort by score sort_index = (-data_numpy[2, :, :, :].sum(axis=1)).argsort(axis=1) for t, s in enumerate(sort_index): data_numpy[:, t, :, :] = data_numpy[:, t, :, s].transpose((1, 2, 0)) data_numpy = data_numpy[:, :, :, 0:self.num_person_out] return data_numpy, label def gendata(data_path, label_path, data_out_path, label_out_path, num_person_in=num_person_in, # observe the first 5 persons num_person_out=num_person_out, # then choose 2 persons with the highest score max_frame=max_frame): feeder = Feeder_kinetics( data_path=data_path, label_path=label_path, num_person_in=num_person_in, num_person_out=num_person_out, window_size=max_frame) sample_name = feeder.sample_name sample_label = [] fp = np.zeros((len(sample_name), 3, max_frame, num_joint, num_person_out), dtype=np.float32) for i, s in enumerate(tqdm(sample_name)): data, label = feeder[i] fp[i, :, 0:data.shape[1], :, :] = data sample_label.append(label) with open(label_out_path, 'wb') as f: pickle.dump((sample_name, list(sample_label)), f) np.save(data_out_path, fp) if __name__ == '__main__': parser = argparse.ArgumentParser( description='Kinetics-skeleton Data Converter.') parser.add_argument( '--data_path', default='../data/kinetics_raw') parser.add_argument( '--out_folder', default='../data/kinetics') arg = parser.parse_args() part = ['val', 'train'] for p in part: print('kinetics ', p) if not os.path.exists(arg.out_folder): os.makedirs(arg.out_folder) data_path = '{}/kinetics_{}'.format(arg.data_path, p) label_path = '{}/kinetics_{}_label.json'.format(arg.data_path, p) data_out_path = '{}/{}_data_joint.npy'.format(arg.out_folder, p) label_out_path = '{}/{}_label.pkl'.format(arg.out_folder, p) gendata(data_path, label_path, data_out_path, label_out_path)
6,090
32.467033
96
py
2s-AGCN
2s-AGCN-master/feeders/feeder.py
import numpy as np import pickle import torch from torch.utils.data import Dataset import sys sys.path.extend(['../']) from feeders import tools class Feeder(Dataset): def __init__(self, data_path, label_path, random_choose=False, random_shift=False, random_move=False, window_size=-1, normalization=False, debug=False, use_mmap=True): """ :param data_path: :param label_path: :param random_choose: If true, randomly choose a portion of the input sequence :param random_shift: If true, randomly pad zeros at the begining or end of sequence :param random_move: :param window_size: The length of the output sequence :param normalization: If true, normalize input sequence :param debug: If true, only use the first 100 samples :param use_mmap: If true, use mmap mode to load data, which can save the running memory """ self.debug = debug self.data_path = data_path self.label_path = label_path self.random_choose = random_choose self.random_shift = random_shift self.random_move = random_move self.window_size = window_size self.normalization = normalization self.use_mmap = use_mmap self.load_data() if normalization: self.get_mean_map() def load_data(self): # data: N C V T M try: with open(self.label_path) as f: self.sample_name, self.label = pickle.load(f) except: # for pickle file from python2 with open(self.label_path, 'rb') as f: self.sample_name, self.label = pickle.load(f, encoding='latin1') # load data if self.use_mmap: self.data = np.load(self.data_path, mmap_mode='r') else: self.data = np.load(self.data_path) if self.debug: self.label = self.label[0:100] self.data = self.data[0:100] self.sample_name = self.sample_name[0:100] def get_mean_map(self): data = self.data N, C, T, V, M = data.shape self.mean_map = data.mean(axis=2, keepdims=True).mean(axis=4, keepdims=True).mean(axis=0) self.std_map = data.transpose((0, 2, 4, 1, 3)).reshape((N * T * M, C * V)).std(axis=0).reshape((C, 1, V, 1)) def __len__(self): return len(self.label) def __iter__(self): return self def __getitem__(self, index): data_numpy = self.data[index] label = self.label[index] data_numpy = np.array(data_numpy) if self.normalization: data_numpy = (data_numpy - self.mean_map) / self.std_map if self.random_shift: data_numpy = tools.random_shift(data_numpy) if self.random_choose: data_numpy = tools.random_choose(data_numpy, self.window_size) elif self.window_size > 0: data_numpy = tools.auto_pading(data_numpy, self.window_size) if self.random_move: data_numpy = tools.random_move(data_numpy) return data_numpy, label, index def top_k(self, score, top_k): rank = score.argsort() hit_top_k = [l in rank[i, -top_k:] for i, l in enumerate(self.label)] return sum(hit_top_k) * 1.0 / len(hit_top_k) def import_class(name): components = name.split('.') mod = __import__(components[0]) for comp in components[1:]: mod = getattr(mod, comp) return mod def test(data_path, label_path, vid=None, graph=None, is_3d=False): ''' vis the samples using matplotlib :param data_path: :param label_path: :param vid: the id of sample :param graph: :param is_3d: when vis NTU, set it True :return: ''' import matplotlib.pyplot as plt loader = torch.utils.data.DataLoader( dataset=Feeder(data_path, label_path), batch_size=64, shuffle=False, num_workers=2) if vid is not None: sample_name = loader.dataset.sample_name sample_id = [name.split('.')[0] for name in sample_name] index = sample_id.index(vid) data, label, index = loader.dataset[index] data = data.reshape((1,) + data.shape) # for batch_idx, (data, label) in enumerate(loader): N, C, T, V, M = data.shape plt.ion() fig = plt.figure() if is_3d: from mpl_toolkits.mplot3d import Axes3D ax = fig.add_subplot(111, projection='3d') else: ax = fig.add_subplot(111) if graph is None: p_type = ['b.', 'g.', 'r.', 'c.', 'm.', 'y.', 'k.', 'k.', 'k.', 'k.'] pose = [ ax.plot(np.zeros(V), np.zeros(V), p_type[m])[0] for m in range(M) ] ax.axis([-1, 1, -1, 1]) for t in range(T): for m in range(M): pose[m].set_xdata(data[0, 0, t, :, m]) pose[m].set_ydata(data[0, 1, t, :, m]) fig.canvas.draw() plt.pause(0.001) else: p_type = ['b-', 'g-', 'r-', 'c-', 'm-', 'y-', 'k-', 'k-', 'k-', 'k-'] import sys from os import path sys.path.append( path.dirname(path.dirname(path.dirname(path.abspath(__file__))))) G = import_class(graph)() edge = G.inward pose = [] for m in range(M): a = [] for i in range(len(edge)): if is_3d: a.append(ax.plot(np.zeros(3), np.zeros(3), p_type[m])[0]) else: a.append(ax.plot(np.zeros(2), np.zeros(2), p_type[m])[0]) pose.append(a) ax.axis([-1, 1, -1, 1]) if is_3d: ax.set_zlim3d(-1, 1) for t in range(T): for m in range(M): for i, (v1, v2) in enumerate(edge): x1 = data[0, :2, t, v1, m] x2 = data[0, :2, t, v2, m] if (x1.sum() != 0 and x2.sum() != 0) or v1 == 1 or v2 == 1: pose[m][i].set_xdata(data[0, 0, t, [v1, v2], m]) pose[m][i].set_ydata(data[0, 1, t, [v1, v2], m]) if is_3d: pose[m][i].set_3d_properties(data[0, 2, t, [v1, v2], m]) fig.canvas.draw() # plt.savefig('/home/lshi/Desktop/skeleton_sequence/' + str(t) + '.jpg') plt.pause(0.01) if __name__ == '__main__': import os os.environ['DISPLAY'] = 'localhost:10.0' data_path = "../data/ntu/xview/val_data_joint.npy" label_path = "../data/ntu/xview/val_label.pkl" graph = 'graph.ntu_rgb_d.Graph' test(data_path, label_path, vid='S004C001P003R001A032', graph=graph, is_3d=True) # data_path = "../data/kinetics/val_data.npy" # label_path = "../data/kinetics/val_label.pkl" # graph = 'graph.Kinetics' # test(data_path, label_path, vid='UOD7oll3Kqo', graph=graph)
7,161
34.81
116
py
GA_CARS_2020
GA_CARS_2020-master/GA_Family_Selection.py
from typing import List import tensorflow as tf import numpy as np import os from deap import base from deap import creator from deap import tools from keras.callbacks import EarlyStopping, ModelCheckpoint from sklearn.model_selection import train_test_split from tqdm import tqdm import time from sklearn.metrics import roc_auc_score import json from operator import attrgetter from keras.models import load_model import random import gc import keras.backend as K tqdm.pandas() os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" # The GPU id to use, usually either "0" or "1"; os.environ["CUDA_VISIBLE_DEVICES"] = "1" # Do other imports now... config = tf.compat.v1.ConfigProto() config.gpu_options.allow_growth = True sess = tf.compat.v1.Session(config=config) family_exp_weights = lambda x: (1 - np.exp(-x)) / (1 - np.exp(-1)) def create_output_folder(path: str) -> None: """ Creates an output folder in the given path :param path: :return: """ try: os.mkdir(path) except FileExistsError: pass def read_dict(name: str): """ :param name: path to a json dict :return: Dict """ with open(name, 'r') as f: return json.load(f) def single_point_mutation(individual): """ Currently not used Flips a random bit in the given individual :param individual: GA Individual :return: Mutated Ind """ size = len(individual) rand_index = random.randint(0, size - 1) # 0 if value was 1 , and 1 if value was 0 rand_value = abs(individual[rand_index] - 1) individual[rand_index] = rand_value return individual, def generate_child(ind1, ind2): """ N point Crossover between two individuals :param ind1: GA Parent 1 :param ind2: GA Parent 2 :return: 2 new children """ child = [] # I assume that len(ind1)=len(ind2) # 50% chance to take a bit from Parent-1 and 50% to take a bit from Parent-2 for i in range(len(ind1)): if random.random() < 0.5: child += [ind1[i]] else: child += [ind2[i]] return child # Generating two children based on n_point_crossover def n_point_crossover(ind1, ind2): return generate_child(ind1, ind2), generate_child(ind1, ind2) def five_point_crossover(ind1, ind2): """ 5 Point crossover between two individuals :param ind1: Parent-1 :param ind2: Parent-2 :return: Two generated children """ points = random.sample(range(1, len(ind1)), 4) points.sort() i1, i2, i3, i4 = points[0], points[1], points[2], points[3] ind1[i1:i2], ind1[i3:i4], ind2[i1:i2], ind2[i3:i4] = ind2[i1:i2], ind2[i3:i4], ind1[i1:i2], ind1[i3:i4] return ind1, ind2 def mixedCrossovers(ind1, ind2): """ Crossover used in this Expr. 50% chance for a 5-point crossover and 50% chance for n-point crossover. Notice that family-vector is not used in the crossover (stays as it was). :param ind1: Parent-1 :param ind2: Parent-2 :return: """ features1, family1 = ind1 features2, family2 = ind2 crossoverSelectionThres = 0.5 if random.random() < crossoverSelectionThres: new_features1, new_features2 = five_point_crossover(features1, features2) else: new_features1, new_features2 = n_point_crossover(features1, features2) new_ind1 = [new_features1, family1] new_ind2 = [new_features2, family2] return new_ind1, new_ind2 def save_list(lst, path): """ Saves a list in text-form, so results can be read mid-expr. :param lst: Some list :param path: path to save :return: None """ with open(path, "w") as f: for s in lst: f.write(str(s) + "\n") def get_active_families(ind): """ :param ind: feature-selection vector & family mask vector :return: Number of features that are active """ _, family = ind return sum(family) def set_fits_for_inds(inds, max_ones, max_family, auc_func): """ Sets the fitness for the given generation of individuals :param inds: GA Individuals :param max_ones: Max number of 1's in the given generation :param max_family: Max number of families in the given generation :param auc_func: Function that maps inds to their AUC score. :return: None """ # Weights for the fitness. AUC_WEIGHT = 0.8 FAMILY_WEIGHT = 0.05 ONES_WEIGHTS = 0.15 auc_mapper = lambda indv: 1 - auc_func(indv) METRIC_INDS = list(map(auc_mapper, inds)) max_metric_inds = max(METRIC_INDS) for ind, mse_ind in zip(inds, METRIC_INDS): fitness_score = AUC_WEIGHT * (mse_ind / max_metric_inds) + ONES_WEIGHTS * (sum(ind[0]) / max_ones) fitness_score += FAMILY_WEIGHT * family_exp_weights((sum(ind[1]) / max_family)) ind.fitness.values = fitness_score, def mutate_family_ind(ind, low, up, indpb_features, indpb_family): """ Mutation is to redraw a bit with some given prob :param ind: GA individual :param low: Minimum value in vector (0 in a binary vector) :param up: Maximum value in vector (1 in a binary vector) :param indpb_features: Prob to mutate the feature-vector :param indpb_family: Prob to mutate the family-mask-vector :return: Mutated ind """ features, family = ind for i in range(len(features)): if random.random() < indpb_features: features[i] = random.randint(low, up) for i in range(len(family)): if random.random() < indpb_family: family[i] = random.randint(low, up) return [features, family], def get_ind_code(ind): """ :param ind: GA Ind :return: Code that is used to save scores in the cache (dictionary key) """ features, family = ind code = tuple(features), tuple(family) return code def opt_tournament(individuals, k, auc_func, tournsize, fit_attr="fitness"): """ :param individuals: Population :param k: How many individuals to select :param auc_func: function to calculate the AUC of every individual :param tournsize: How many individuals per tournament :param fit_attr: key for selection :return: k selected individuals """ chosen = [] sum_first = lambda x: sum(x[0]) sum_second = lambda x: sum(x[1]) max_ones = max(list(map(sum_first, individuals))) max_family = max(list(map(sum_second, individuals))) # getting k inds for _ in range(k): comps = tools.selRandom(individuals, tournsize) set_fits_for_inds(comps, max_ones, max_family, auc_func) chosen.append(max(comps, key=attrgetter(fit_attr))) return chosen class FeatureFamilySelectionGA: def __init__(self, n_generations: int, population_size: int, mu_for_sampling: float, sigma_for_sampling: float, crossover_prob: float, mutation_prob: float, ind_length, family_length, family_intervals, random_state=42): """ :param n_generations: Number of generations to run :param population_size: Population Size :param mu_for_sampling: Used to initialize a normal distribution :param sigma_for_sampling: Used to initialize a normal distribution :param crossover_prob: Crossover probability :param mutation_prob: Mutation probability :param ind_length: Individual length :param random_state: Initial random seed """ assert 0 <= crossover_prob <= 1, "ILLEGAL CROSSOVER PROBABILITY" assert 0 <= mutation_prob <= 1, "ILLEGAL MUTATION PROBABILITY" assert population_size > 0, "Population size must be a positive integer" assert n_generations > 0, "Number of generations must be a positive integer" assert mu_for_sampling > 0 and sigma_for_sampling > 0, 'Illegal selection params' # params self.n_generations = n_generations self.population_size = population_size self.mu_for_sampling = mu_for_sampling self.sigma_for_sampling = sigma_for_sampling self.crossover_prob = crossover_prob self.mutation_prob = mutation_prob self.random_state = random_state self.family_length = family_length self.family_intervals = family_intervals # params # GA toolbox self.toolbox = base.Toolbox() # solving for minimum fitness creator.create("FitnessMin", base.Fitness, weights=(-1.0,)) creator.create("Individual", list, fitness=creator.FitnessMin) # probability to generate '1' in a random selection self.one_prob = None self.fitness_dict = {} self.mask_dicts = {} self.pop_dict = {} self.std_gen = [] self.mean_gen = [] self.median_gen = [] self.max_gen = [] self.min_gen = [] self.time_gen = [] self.__set_ind_generator(max_ind_size=ind_length, family_size=self.family_length) self.toolbox.register("individual", tools.initIterate, creator.Individual, self.toolbox.gen_ind) self.toolbox.register("population", tools.initRepeat, list, self.toolbox.individual) self.toolbox.register("evaluate", self.evalInd) self.toolbox.register("mate", mixedCrossovers) self.toolbox.register("mutate", mutate_family_ind, low=0, up=1, indpb_features=0.005, indpb_family=0.01) self.toolbox.register("select", opt_tournament, tournsize=5) def reset_stats(self): """ Resets saved-metrics. :return: None """ self.fitness_dict = {} self.pop_dict = {} self.std_gen = [] self.mean_gen = [] self.median_gen = [] self.max_gen = [] self.min_gen = [] self.time_gen = [] def save_all_data(self, curr_generation, output_folder): """ Saves metrics for the given generations in a given path. :param curr_generation: Current generation (int) :param output_folder: folder to save data in. :return: None """ written_dict = {str(k): v for k, v in self.fitness_dict.items()} with open(output_folder + "fitness_dict.json", 'w') as file: json.dump(written_dict, file) with open(output_folder + "gens_dict.json", 'w') as file: json.dump(self.pop_dict, file) with open(output_folder + "am_alive", "w") as f: f.write("Still running at generation :" + str(curr_generation) + "\n") save_list(self.std_gen, output_folder + "std.txt") save_list(self.mean_gen, output_folder + "mean.txt") save_list(self.median_gen, output_folder + "median.txt") save_list(self.max_gen, output_folder + "max.txt") save_list(self.min_gen, output_folder + "min.txt") save_list(self.time_gen, output_folder + "time.txt") def __init_normal_dist(self, length_to_gen: int, family_len: int) -> List[list]: """ Creates a random individual with a given length :param length_to_gen: How many '1/0' to generate :return: A random binary array of the given size """ family = [] feature_arr = [] while len(feature_arr) < length_to_gen: feature_arr += [int(random.random() <= self.one_prob)] family_one_prob = 0.8 while len(family) < family_len: family += [int(random.random() <= family_one_prob)] return [feature_arr, family] def __set_ind_generator(self, max_ind_size, family_size) -> None: """ Sets the individual generator. This is done by sampling from a normal distribution (from the given __init__ params) the probability of '1' to generate. For example: max_ind_size = 480 mu = 120 sigma = 40 Assume randomly chose 130, then one_probability = 0.2708 = (130 / 480) = (normal_dist_sample / max_ind_size) :param max_ind_size: Size of the individual to generate :return: None """ normal_dist_sample = np.random.normal(self.mu_for_sampling, self.sigma_for_sampling, 1000) self.one_prob = random.choice(normal_dist_sample) / max_ind_size self.toolbox.register("gen_ind", self.__init_normal_dist, max_ind_size, family_size) def __init_seeds(self) -> None: """ Resets seeds back to the initial seed :return: None """ tf.random.set_seed(self.random_state) np.random.seed(self.random_state) random.seed(self.random_state) os.environ['PYTHONHASHSEED'] = str(self.random_state) def learn_only_pred(self, output_folder, context_val, y_val, classifier_path) -> None: """ :param output_folder: Where to write the output to :param context_val: Validation Context [matrix] :param y_val: Target values for validation :param classifier_path: Path to a keras classifier that will be used used to load fresh weights for every fitness calculation (for retraining the network from scratch) :return: """ self.reset_stats() self.__init_seeds() create_output_folder(output_folder) assert len(context_val) == len(y_val) classifier = load_model(classifier_path) get_auc_func = lambda ind: self.only_predict_get_auc(ind, context_val, y_val, classifier) self.__run_gens(output_folder, get_auc_func) def learn_only_pred_full_arch(self, output_folder: str, val_input, y_val, classifier_path: str) -> None: """ Starts the learning process to the only-predict algorithm :param output_folder: Folder to save results in :param val_input: Validation Data :param y_val: Validation Labels :param classifier_path: Path to the Keras MLP classifier. :return: None """ self.reset_stats() self.__init_seeds() create_output_folder(output_folder) input_size = 3 # item, user and context (3) assert len(val_input) == input_size classifier = load_model(classifier_path) get_auc_func = lambda ind: self.only_predict_get_auc_full_arch(ind, val_input, y_val, classifier) self.__run_gens(output_folder, get_auc_func) def learn_full_train(self, output_folder, context_train, y_train, context_val, y_val, classifier_path, fresh_weights_path): """ Starts the learning process for the full-train algorithm :param output_folder: Folder to save results in :param context_train: Training Data :param y_train: Training labels :param context_val: Validation Data :param y_val: Validation Labels :param classifier_path: Path to the keras MLP classifier :param fresh_weights_path: Path to fresh-weights (non-trained) for the classifier. :return: None """ self.reset_stats() self.__init_seeds() create_output_folder(output_folder) assert len(context_val) == len(y_val) and len(context_train) == len(y_train) # classifier = load_model(classifier_path) density = sum(y_train) / len(y_train) ones_weight = (1 - density) / density zero_weight = 1 class_weights = {0: zero_weight, 1: ones_weight} get_auc_func = lambda ind: self.train_mlp_net_get_auc(ind, context_train, y_train, context_val, y_val, classifier_path, fresh_weights_path, class_weights) self.__run_gens(output_folder, get_auc_func) def transform_array_by_ind_and_family(self, ind, arr): """ Returns the input-arr (training/validation data) after applying the feature-selection via the feature-vector and the mask-vector. - Selected features are not modified. - Non selected features are replaced with 0 (Not removed to retain the architecture input dimensions) :param ind: Binary Vector for feature selection & family mask :param arr: Feature vector :return: The feature vector after performing feature-selection by the family mask and feature-selection vector """ # saving masks, in order to not recalculate it again features, family = ind code = tuple(family) if code in self.mask_dicts: mask = self.mask_dicts[code] else: mask = np.array([0] * len(features)) for family_mask_index in range(len(family)): family_interval = self.family_intervals[family_mask_index] for feature_index in family_interval: mask[feature_index] = family[family_mask_index] self.mask_dicts[code] = mask return np.multiply(np.multiply(features, arr), mask) def train_mlp_net_get_auc(self, individual, context_train, y_train, context_val, y_val, classifier_path, fresh_weights, class_weights): """ :param individual: Ind to evaluate :param context_train: Training data :param y_train: Training labels :param context_val: Validation Data :param y_val: Validation Labels :param classifier_path: Path to a MLP classifier :param fresh_weights: Path to fresh weights :param class_weights: Class weights for training :return: AUC Score """ if self.is_in_dict(individual): return self.get_fit(individual) best_model_path = '../data/expr_running_best_full_train.h5' es = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='min') mcp = ModelCheckpoint(best_model_path, save_best_only=True, mode='min') make_zero = lambda element: self.transform_array_by_ind_and_family(individual, element) x_val = np.array(list(map(make_zero, context_val))) x_train = np.array(list(map(make_zero, context_train))) classifier = load_model(classifier_path) classifier.load_weights(fresh_weights) classifier.fit(x_train, y_train, validation_data=(x_val, y_val), epochs=100, batch_size=256, verbose=0, callbacks=[es, mcp], class_weight=class_weights) best_classifier = load_model(best_model_path) val_predict = best_classifier.predict(x_val, batch_size=256) res_auc = roc_auc_score(y_val, val_predict) self.save_fitness(individual, res_auc) # Clearing model, to avoid a memory leak. del best_classifier, es, mcp, x_val, x_train, classifier, val_predict gc.collect() K.clear_session() tf.compat.v1.reset_default_graph() return res_auc def __run_gens(self, outputFolder: str, get_auc_func) -> None: """ :param outputFolder: Folder to save output data :param get_auc_func: Function that receives an individual and returns his AUC score :return: None """ # Crossover probability & Mutation Probability CXPB, MUTPB = self.crossover_prob, self.mutation_prob pop = self.toolbox.population(n=self.population_size) for g in tqdm(range(0, self.n_generations)): start_time = time.time() self.pop_dict[g] = [list(x).copy() for x in pop] # A new generation print("-- Generation %i --" % g) # Note : The fitness is calculated inside of the select function, for better performance. # Meaning that the fitness is calculated only for the selected individuals in the tournament. # Select the next generation individuals offspring = self.toolbox.select(pop, len(pop), get_auc_func) # Clone the selected individuals offspring = list(map(self.toolbox.clone, offspring)) fits = [ind.fitness.values[0] for ind in offspring] self.update_progress_arrays(offspring, fits, g, start_time, outputFolder) # Apply crossover and mutation on the offspring for child1, child2 in zip(offspring[::2], offspring[1::2]): if random.random() < CXPB: self.toolbox.mate(child1, child2) del child1.fitness.values del child2.fitness.values for mutant in offspring: if random.random() < MUTPB: self.toolbox.mutate(mutant) del mutant.fitness.values # Changed to offsprings because fitness changes for the entire generation pop[:] = offspring def only_predict_get_auc_full_arch(self, individual, input_val, y_val, fully_trained_mlp) -> float: """ Calculates the fitness of a given individual by predicting on the fully trained arch :param individual: Ind to predict on :param input_val: a tuple of 3 elements - [item_val, user_val, context_val] :param y_val: Validation Target :param fully_trained_mlp: Fully trained arch classifier :return: AUC of the prediction """ if self.is_in_dict(individual): return self.get_fit(individual) item_val, user_val, con_val = input_val make_zero = lambda element: self.transform_array_by_ind_and_family(individual, element) x_cv = np.array(list(map(make_zero, con_val))) input_val_ind = [item_val, user_val, x_cv] val_predict = fully_trained_mlp.predict(input_val_ind) res_auc = roc_auc_score(y_val, val_predict) self.save_fitness(individual, res_auc) return res_auc def only_predict_get_auc(self, individual, context_val, y_val, fully_trained_mlp) -> float: """ Calculates the fitness of a given individual by predicting on the fully trained mlp :param individual: Ind to predict on :param context_val: Validation context :param y_val: Validation Target :param fully_trained_mlp: Fully trained classifier :return: AUC of the prediction """ if self.is_in_dict(individual): return self.get_fit(individual) make_zero = lambda element: self.transform_array_by_ind_and_family(individual, element) x_cv = np.array(list(map(make_zero, context_val))) val_predict = fully_trained_mlp.predict(x_cv) res_auc = roc_auc_score(y_val, val_predict) self.save_fitness(individual, res_auc) return res_auc def save_fitness(self, ind, val): """ :param ind: GA individual to save :param val: metric to save :return: None """ code = get_ind_code(ind) self.fitness_dict[code] = val def is_in_dict(self, ind) -> bool: """ :param ind: GA individual :return: True iff the given individual has a value store in the cache (past metric) """ code = get_ind_code(ind) return code in self.fitness_dict def get_fit(self, ind): """ :param ind: GA individual :return: Saved metric """ code = get_ind_code(ind) if code in self.fitness_dict: return self.fitness_dict[code] """ Not used on purpose, evaluation of individuals is done inside the selection function (for performance reasons). Instead of evaluating the entire generation, we evaluate only the individuals that were randomly selected in the tournament selection, thus saving unnecessary training of the MLP network . """ def evalInd(self, individual, METRIC): pass def update_progress_arrays(self, pop, fits, g, gen_start, outputpath): """ :param pop: current generation of GA individuals :param fits: list of fitness scores :param g: current generation number :param gen_start: time that this generation started :param outputpath: output to save the data :return: None """ length = len(pop) mean = sum(fits) / length sum2 = sum(x * x for x in fits) std = abs(sum2 / length - mean ** 2) ** 0.5 min_feat = min(fits) max_feat = max(fits) median_feat = np.median(fits) self.min_gen += [min_feat] self.max_gen += [max_feat] self.mean_gen += [mean] self.std_gen += [std] self.median_gen += [median_feat] self.time_gen += [time.time() - gen_start] self.save_all_data(g, outputpath) def get_train_test_val_split(arr): x_train, x_test = train_test_split(arr, test_size=0.3, shuffle=False) x_val, x_test = train_test_split(x_test, test_size=0.66, shuffle=False) return x_train, x_test, x_val if __name__ == '__main__': # Family intervals for dataset family_intervals_ = read_dict('../data_hero_ds/family_intervals.json') index_to_name = read_dict('../data_hero_ds/Family_Indexes.json') index_to_interval = [] for key in index_to_name.keys(): index_to_interval += [family_intervals_[index_to_name[key]]] n_generations_ = 300 n_population_size = 100 sigma_for_init = 40 CXPB_ = 0.65 MUTPB_ = 1 ind_length_ = 661 number_of_families = len(family_intervals_.keys()) mu_for_init = 0.25 * ind_length_ fs = FeatureFamilySelectionGA(n_generations_, n_population_size, mu_for_init, sigma_for_init, CXPB_, MUTPB_, ind_length_, number_of_families, index_to_interval) context_train_, _, context_val_ = get_train_test_val_split(np.load('../data_hero_ds/X.npy')) y_train_, _, y_val_ = get_train_test_val_split(np.load('../data_hero_ds/y.npy')) assert len(context_train_) == len(y_train_) and len(context_val_) == len(y_val_) mlp_path = '../fresh_mlp_data/HEARO_fresh_mlp_200_100_50_1.h5' fresh_weights_ = '../fresh_mlp_data/HEARO_fresh_weights_200_100_50_1.h5' output_folder_ = '../RESULTS_HERO_DS/GA_FAMILY/GA_RES_20_5_FULL_TRAIN/' fs.learn_full_train(output_folder_, context_train_, y_train_, context_val_, y_val_, mlp_path, fresh_weights_)
26,115
34.196765
118
py
GA_CARS_2020
GA_CARS_2020-master/Feature_Selection_Baselines.py
import os import tensorflow as tf import numpy as np import random from keras import Sequential from keras.layers import Dense from keras.callbacks import EarlyStopping, ModelCheckpoint from keras.models import load_model from sklearn.metrics import roc_auc_score from tqdm import tqdm from sklearn.model_selection import train_test_split os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" # The GPU id to use, usually either "0" or "1"; os.environ["CUDA_VISIBLE_DEVICES"] = "1" # Do other imports now... config = tf.compat.v1.ConfigProto() config.gpu_options.allow_growth = True sess = tf.compat.v1.Session(config=config) seed_value = 42 def get_mlp_model(save_initial_weights_path=None, save_fresh_mlp_path=None, input_shape=480): """ Creates A fresh-mlp model for evaluation :param save_initial_weights_path: :param save_fresh_mlp_path: :param input_shape: :return: """ classifier = Sequential() classifier.add(Dense(200, activation='relu', kernel_initializer='random_normal', input_dim=input_shape)) classifier.add(Dense(100, activation='relu', kernel_initializer='random_normal', input_dim=200)) classifier.add(Dense(50, activation='relu', kernel_initializer='random_normal', input_dim=100)) classifier.add(Dense(1, activation='sigmoid', kernel_initializer='random_normal')) classifier.compile(optimizer='ADAM', loss='binary_crossentropy', metrics=['mse', 'accuracy']) if save_initial_weights_path is not None: classifier.save_weights(save_initial_weights_path) if save_fresh_mlp_path is not None: classifier.save(save_fresh_mlp_path) return classifier def init_seeds(): os.environ['PYTHONHASHSEED'] = str(seed_value) random.seed(seed_value) np.random.seed(seed_value) tf.random.set_seed(seed_value) def reduce_by_indexes(arr, indexes): """ :param arr: Context-Feature arr :param indexes: indexes of chosen features :return: reduced arr """ res = [] for i in range(len(arr)): if i in indexes: res += [arr[i]] else: res += [0] return np.array(res) def reduce_all_by_indexes(matrix, indexes): """ reduces every vector in the arr :param matrix: matrix of contextual features :param indexes: indexes of chosen features :return: """ return np.array(list(map(lambda element: reduce_by_indexes(element, indexes), matrix))) def run_feed_forward_selection(context_train, context_val, class_weights, model, fresh_weights_path, y_train, y_val, alive_path): """ :param context_train: training data :param context_val: validation data :param class_weights: class-weights for training :param model: Model for evaluation :param fresh_weights_path: Fresh Weights for initialization :param y_train: training labels :param y_val: validation labels :param alive_path: Path to write results to :return: Selected indexes """ selected_features = [] sample = context_train[0] best_score = -float('inf') while True: print(f"Starting iteration : Best AUC {best_score} , len(featurs) = {len(selected_features)}") best_feature = -1 current_size = len(selected_features) for i in tqdm(range(len(sample))): np.save(alive_path, np.array(selected_features)) model.load_weights(fresh_weights_path) assert len(selected_features) == current_size if i in selected_features: continue current_features = selected_features.copy() current_features += [i] assert len(current_features) == current_size + 1 x_train = reduce_all_by_indexes(context_train, current_features) x_val = reduce_all_by_indexes(context_val, current_features) es = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='min') mcp = ModelCheckpoint('../data_cars_ds/FF_best_mlp_feature_selection.h5', save_best_only=True, mode='min') model.fit(x_train, y_train, validation_data=(x_val, y_val), epochs=100, batch_size=256, verbose=0, callbacks=[es, mcp], class_weights=class_weights) model = load_model('../data_cars_ds/FF_best_mlp_feature_selection.h5') val_predict = model.predict(x_val, batch_size=256) current_auc = roc_auc_score(y_val, val_predict) if current_auc > best_score: print(f"{current_auc} took over {best_score} , with index {i}") best_feature = i best_score = current_auc del es, mcp, current_features # no improvement if best_feature == -1: break selected_features.append(best_feature) return selected_features def run_backwards_elimination_selection(context_train, context_val, class_weights, model, fresh_weights_path, y_train, y_val, alive_path): """ :param context_train: training data :param context_val: validation data :param class_weights: class-weights for training :param model: Model for evaluation :param fresh_weights_path: Fresh Weights for initialization :param y_train: training labels :param y_val: validation labels :param alive_path: Path to write results to :return: Selected indexes """ sample = context_train[0] dims = len(sample) selected_features = list(range(dims)) best_score = -float('inf') while True: print(f"Starting iteration : Best AUC {best_score} , len(featurs) = {len(selected_features)}") worst_feature = -1 current_size = len(selected_features) for i in tqdm(range(len(sample))): np.save(alive_path, np.array(selected_features)) model.load_weights(fresh_weights_path) assert len(selected_features) == current_size if i not in selected_features: continue current_features = selected_features.copy() current_features.remove(i) assert len(current_features) == current_size - 1 x_train = reduce_all_by_indexes(context_train, current_features) x_val = reduce_all_by_indexes(context_val, current_features) es = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='min') mcp = ModelCheckpoint('../data_cars_ds/BE_best_mlp_feature_selection.h5', save_best_only=True, mode='min') model.fit(x_train, y_train, validation_data=(x_val, y_val), epochs=100, batch_size=256, verbose=0, callbacks=[es, mcp], class_weights=class_weights) model = load_model('../data_cars_ds/BE_best_mlp_feature_selection.h5') val_predict = model.predict(x_val, batch_size=256) current_auc = roc_auc_score(y_val, val_predict) if current_auc > best_score: print(f"{current_auc} took over {best_score} , with index {i}") worst_feature = i best_score = current_auc del es, mcp, current_features # no improvement if worst_feature == -1: break selected_features.remove(worst_feature) return selected_features def get_train_test_val_split(arr): x_train, x_test = train_test_split(arr, test_size=0.3, shuffle=False) x_val, x_test = train_test_split(x_test, test_size=0.66, shuffle=False) return x_train, x_test, x_val if __name__ == '__main__': init_seeds() Y_train, _, Y_val = get_train_test_val_split(np.load('../data_hero_ds/y.npy')) X_train, _, X_val = get_train_test_val_split(np.load('../data_hero_ds/X.npy')) density = sum(Y_train) / len(Y_train) ones_weight = (1 - density) / density zero_weight = 1 print(ones_weight, density) class_Weights = {0: zero_weight, 1: ones_weight} fresh_weights_Path = '../feature_selections/FF_mlp_initial.h5' model_ = get_mlp_model(fresh_weights_Path, None, len(X_train[0])) save_path = '../feature_selections/FF_selection.npy' features = run_feed_forward_selection(X_train, X_val, class_Weights, model_, fresh_weights_Path, Y_train, Y_val, save_path) np.save(save_path, features)
8,391
33.821577
118
py
GA_CARS_2020
GA_CARS_2020-master/base_solution.py
import gc import json import os import random import time from operator import attrgetter from typing import List import keras.backend as K import numpy as np import tensorflow as tf from deap import base from deap import creator from deap import tools from keras.callbacks import EarlyStopping, ModelCheckpoint from keras.models import load_model from sklearn.metrics import roc_auc_score from sklearn.model_selection import cross_val_score, train_test_split from tqdm import tqdm tqdm.pandas() os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" # The GPU id to use, usually either "0" or "1"; os.environ["CUDA_VISIBLE_DEVICES"] = "1" # Do other imports now... config = tf.compat.v1.ConfigProto() config.gpu_options.allow_growth = True sess = tf.compat.v1.Session(config=config) def create_output_folder(path: str) -> None: """ Creates an output folder in the given path :param path: :return: """ try: os.mkdir(path) except FileExistsError: pass def read_dict(name: str): """ :param name: path to a json dict :return: Dict """ with open(name, 'r') as f: return json.load(f) def opt_tournament(individuals, k, auc_func, tournsize, fit_attr="fitness"): """ :param individuals: Population :param k: How many individuals to select :param auc_func: function to calculate the AUC of every individual :param tournsize: How many individuals per tournament :param fit_attr: key for selection :return: k selected individuals """ chosen = [] max_ones = max(list(map(sum, individuals))) for i in range(k): comps = tools.selRandom(individuals, tournsize) set_fits_for_inds(comps, max_ones, auc_func) chosen.append(max(comps, key=attrgetter(fit_attr))) return chosen def single_point_mutation(individual): """ Currently not used Flips a random bit in the given individual :param individual: GA Individual :return: Mutated Ind """ size = len(individual) rand_index = random.randint(0, size - 1) # 0 if value was 1 , and 1 if value was 0 rand_value = abs(individual[rand_index] - 1) individual[rand_index] = rand_value return individual, def generate_child(ind1, ind2): """ N point Crossover between two individuals :param ind1: GA Parent 1 :param ind2: GA Parent 2 :return: 2 new children """ child = [] # I assume that len(ind1)=len(ind2) for i in range(len(ind1)): if random.random() < 0.5: child += [ind1[i]] else: child += [ind2[i]] return child # Generating two childern based on n_point_crossover def n_point_crossover(ind1, ind2): return generate_child(ind1, ind2), generate_child(ind1, ind2) def five_point_crossover(ind1, ind2): """ 5 Point crossover between two individuals :param ind1: Parent-1 :param ind2: Parent-2 :return: Two generated children """ points = random.sample(range(1, len(ind1)), 4) points.sort() i1, i2, i3, i4 = points[0], points[1], points[2], points[3] ind1[i1:i2], ind1[i3:i4], ind2[i1:i2], ind2[i3:i4] = ind2[i1:i2], ind2[i3:i4], ind1[i1:i2], ind1[i3:i4] return ind1, ind2 def mixedCrossovers(ind1, ind2): """ Crossover used in this Expr. 50% chance for a 5-point crossover and 50% chance for n-point crossover. Notice that family-vector is not used in the crossover (stays as it was). :param ind1: Parent-1 :param ind2: Parent-2 :return: """ crossoverSelectionThres = 0.5 if random.random() < crossoverSelectionThres: # 5 point crossover return five_point_crossover(ind1, ind2) else: # n point crossover return n_point_crossover(ind1, ind2) def save_list(lst, path): """ Saves a list in text-form, so results can be read mid-expr. :param lst: Some list :param path: path to save :return: None """ with open(path, "w") as f: for s in lst: f.write(str(s) + "\n") def set_fits_for_inds(inds, max_ones, auc_func): """ Sets the fitness for the given generation of individuals :param inds: GA Individuals :param max_ones: Max number of 1's in the given generation :param auc_func: Function that maps inds to their AUC score. :return: None """ AUC_WEIGHT = 0.8 auc_mapper = lambda indv: 1 - auc_func(indv) METRIC_INDS = list(map(auc_mapper, inds)) max_metric_inds = max(METRIC_INDS) for ind, mse_ind in zip(inds, METRIC_INDS): fitness_score = AUC_WEIGHT * (mse_ind / max_metric_inds) + (1 - AUC_WEIGHT) * (sum(ind) / max_ones) ind.fitness.values = fitness_score, class FeatureSelectionGA: def __init__(self, n_generations: int, population_size: int, mu_for_sampling: float, sigma_for_sampling: float, crossover_prob: float, mutation_prob: float, ind_length, random_state=42): """ :param n_generations: Number of generations to run :param population_size: Population Size :param mu_for_sampling: Used to initialize a normal distribution :param sigma_for_sampling: Used to initialize a normal distribution :param crossover_prob: Crossover probability :param mutation_prob: Mutation probability :param ind_length: Individual length :param random_state: Initial random seed """ assert 0 <= crossover_prob <= 1, "ILLEGAL CROSSOVER PROBABILITY" assert 0 <= mutation_prob <= 1, "ILLEGAL MUTATION PROBABILITY" assert population_size > 0, "Population size must be a positive integer" assert n_generations > 0, "Number of generations must be a positive integer" assert mu_for_sampling > 0 and sigma_for_sampling > 0, 'Illegal selection params' # params self.n_generations = n_generations self.population_size = population_size self.mu_for_sampling = mu_for_sampling self.sigma_for_sampling = sigma_for_sampling self.crossover_prob = crossover_prob self.mutation_prob = mutation_prob self.random_state = random_state # params # GA toolbox self.toolbox = base.Toolbox() # Trained mlp, used in 'only_pred' method self.fully_trained_mlp = None # solving for minimum fitness creator.create("FitnessMin", base.Fitness, weights=(-1.0,)) creator.create("Individual", list, fitness=creator.FitnessMin) # probability to generate '1' in a random selection self.one_prob = None self.fitness_dict = {} self.pop_dict = {} self.std_gen = [] self.mean_gen = [] self.median_gen = [] self.max_gen = [] self.min_gen = [] self.time_gen = [] self.__set_ind_generator(max_ind_size=ind_length) self.toolbox.register("individual", tools.initIterate, creator.Individual, self.toolbox.gen_ind) self.toolbox.register("population", tools.initRepeat, list, self.toolbox.individual) self.toolbox.register("evaluate", self.evalInd) self.toolbox.register("mate", mixedCrossovers) self.toolbox.register("mutate", tools.mutUniformInt, low=0, up=1, indpb=0.005) # between 0 and 1 , with probability of 0.005 self.toolbox.register("select", opt_tournament, tournsize=5) def reset_stats(self): """ Resets saved-metrics. :return: None """ self.fitness_dict = {} self.pop_dict = {} self.std_gen = [] self.mean_gen = [] self.median_gen = [] self.max_gen = [] self.min_gen = [] self.time_gen = [] def save_all_data(self, curr_generation, output_folder): """ Saves metrics for the given generations in a given path. :param curr_generation: Current generation (int) :param output_folder: folder to save data in. :return: None """ written_dict = {str(k): v for k, v in self.fitness_dict.items()} with open(output_folder + "fitness_dict.json", 'w') as file: json.dump(written_dict, file) with open(output_folder + "gens_dict.json", 'w') as file: json.dump(self.pop_dict, file) with open(output_folder + "am_alive", "w") as f: f.write("Still running at generation :" + str(curr_generation) + "\n") save_list(self.std_gen, output_folder + "std.txt") save_list(self.mean_gen, output_folder + "mean.txt") save_list(self.median_gen, output_folder + "median.txt") save_list(self.max_gen, output_folder + "max.txt") save_list(self.min_gen, output_folder + "min.txt") save_list(self.time_gen, output_folder + "time.txt") def __init_normal_dist(self, length_to_gen: int) -> List[int]: """ Creates a random individual with a given length :param length_to_gen: How many '1/0' to generate :return: A random binary array of the given size """ ret_arr = [] while len(ret_arr) < length_to_gen: ret_arr += [int(random.random() <= self.one_prob)] return ret_arr def __set_ind_generator(self, max_ind_size) -> None: """ Sets the individual generator. This is done by sampling from a normal distribution (from the given __init__ params) the probability of '1' to generate. For example: max_ind_size = 480 mu = 120 sigma = 40 Assuming 130 was randomly chosen , then one_probability = 0.2708 = (130 / 480) = (normal_dist_sample / max_ind_size) :param max_ind_size: Size of the individual to generate :return: None """ normal_dist_sample = np.random.normal(self.mu_for_sampling, self.sigma_for_sampling, 1000) self.one_prob = random.choice(normal_dist_sample) / max_ind_size self.toolbox.register("gen_ind", self.__init_normal_dist, max_ind_size) def __init_seeds(self) -> None: """ Resets seeds back to the initial seed :return: None """ tf.random.set_seed(self.random_state) np.random.seed(self.random_state) random.seed(self.random_state) os.environ['PYTHONHASHSEED'] = str(self.random_state) def learn_only_pred(self, output_folder, context_val, y_val, classifier_path) -> None: """ :param output_folder: Where to write the output to :param context_val: Validation Context [matrix] :param y_val: Target values for validation :param classifier_path: Path to a keras classifier that will be used used to load fresh weights for every fitness calculation (for retraining the network from scratch) :return: """ self.reset_stats() self.__init_seeds() create_output_folder(output_folder) assert len(context_val) == len(y_val) classifier = load_model(classifier_path) get_auc_func = lambda ind: self.only_predict_get_auc(ind, context_val, y_val, classifier) self.__run_gens(output_folder, get_auc_func) def learn_only_pred_full_arch(self, output_folder: str, val_input, y_val, classifier_path: str) -> None: self.reset_stats() self.__init_seeds() create_output_folder(output_folder) input_size = 3 # item, user and context (3) assert len(val_input) == input_size classifier = load_model(classifier_path) get_auc_func = lambda ind: self.only_predict_get_auc_full_arch(ind, val_input, y_val, classifier) self.__run_gens(output_folder, get_auc_func) def learn_full_train(self, output_folder, context_train, y_train, context_val, y_val, classifier_path, fresh_weights_path): """ Starts the learning process for the full-train algorithm :param output_folder: Folder to save results in :param context_train: Training Data :param y_train: Training labels :param context_val: Validation Data :param y_val: Validation Labels :param classifier_path: Path to the keras MLP classifier :param fresh_weights_path: Path to fresh-weights (non-trained) for the classifier. :return: None """ self.reset_stats() self.__init_seeds() create_output_folder(output_folder) assert len(context_val) == len(y_val) and len(context_train) == len(y_train) density = sum(y_train) / len(y_train) ones_weight = (1 - density) / density zero_weight = 1 class_weights = {0: zero_weight, 1: ones_weight} get_auc_func = lambda ind: self.train_mlp_net_get_auc(ind, context_train, y_train, context_val, y_val, classifier_path, fresh_weights_path, class_weights) self.__run_gens(output_folder, get_auc_func) def learn_full_train_sklearn_model(self, output_folder, X, y, model_class, params, cv_split=3): self.reset_stats() self.__init_seeds() create_output_folder(output_folder) assert len(X) == len(y) get_auc_func = lambda ind: self.train_model_get_auc(ind, X, y, model_class, params, cv_split) self.__run_gens(output_folder, get_auc_func) def train_mlp_net_get_auc(self, individual, context_train, y_train, context_val, y_val, classifier_path, fresh_weights, class_weights): """ Trains an MLP network and returns the resulting AUC :param individual: Ind to evaluate :param context_train: Training data :param y_train: Training labels :param context_val: Validation Data :param y_val: Validation Labels :param classifier_path: Path to a MLP classifier :param fresh_weights: Path to fresh weights :param class_weights: Class weights for training :return: AUC Score """ if self.is_in_dict(individual): return self.get_fit(individual) best_model_path = '../data/expr_running_best.h5' es = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='min') mcp = ModelCheckpoint(best_model_path, save_best_only=True, mode='min') npIndividual = np.array(individual) make_zero = lambda element: np.multiply(npIndividual, element) x_val = np.array(list(map(make_zero, context_val))) x_train = np.array(list(map(make_zero, context_train))) classifier = load_model(classifier_path) classifier.load_weights(fresh_weights) classifier.fit(x_train, y_train, validation_data=(x_val, y_val), epochs=100, batch_size=256, verbose=0, callbacks=[es, mcp], class_weight=class_weights) best_classifier = load_model(best_model_path) val_predict = best_classifier.predict(x_val, batch_size=256) res_auc = roc_auc_score(y_val, val_predict) self.save_fitness(individual, res_auc) del best_classifier, es, mcp, x_val, x_train, classifier, val_predict gc.collect() K.clear_session() tf.compat.v1.reset_default_graph() return res_auc def train_model_get_auc(self, individual, X, y, model_class, model_params, cv_split=3): """ Trains a given model class and returns the resulting AUC :param model_class: Class of model to create :param model_params: Params for the constructed model :param cv_split: Cross validation folds :param individual: Ind to evaluate :param X: Training data :param y: Training labels :return: AUC Score """ model = model_class(**model_params) if self.is_in_dict(individual): return self.get_fit(individual) npIndividual = np.array(individual) make_zero = lambda element: np.multiply(npIndividual, element) x_selection = np.array(list(map(make_zero, X))) scores = cross_val_score(model, x_selection, y, cv=cv_split, scoring='roc_auc') auc_score = np.mean(scores) self.save_fitness(individual, auc_score) del model, npIndividual, x_selection return auc_score def __run_gens(self, outputFolder, get_auc_func): """ :param outputFolder: Folder to save output data :param get_auc_func: Function that receives an individual and returns his AUC score :return: None """ CXPB, MUTPB = self.crossover_prob, self.mutation_prob pop = self.toolbox.population(n=self.population_size) start_time = time.time() for g in tqdm(range(0, self.n_generations)): self.pop_dict[g] = [list(x).copy() for x in pop] # A new generation print("-- Generation %i --" % g) # Note : The fitness is calculated inside of the select function, for better performance. # Meaning that the fitness is calculated only for the selected individuals in the tournament. # Select the next generation individuals offspring = self.toolbox.select(pop, len(pop), get_auc_func) # Clone the selected individuals offspring = list(map(self.toolbox.clone, offspring)) fits = [ind.fitness.values[0] for ind in offspring] self.update_progress_arrays(offspring, fits, g, start_time, outputFolder) # Apply crossover and mutation on the offspring for child1, child2 in zip(offspring[::2], offspring[1::2]): if random.random() < CXPB: self.toolbox.mate(child1, child2) del child1.fitness.values del child2.fitness.values for mutant in offspring: if random.random() < MUTPB: self.toolbox.mutate(mutant) del mutant.fitness.values # Changed to offsprings because fitness changes for the entire generation pop[:] = offspring def only_predict_get_auc_full_arch(self, individual, input_val, y_val, fully_trained_mlp) -> float: """ Calculates the fitness of a given individual by predicting on the fully trained arch :param individual: Ind to predict on :param input_val: a tuple of 3 elements - [item_val, user_val, context_val] :param y_val: Validation Target :param fully_trained_mlp: Fully trained arch classifier :return: AUC of the prediction """ if self.is_in_dict(individual): return self.get_fit(individual) item_val, user_val, con_val = input_val npIndividual = np.array(individual) make_zero = lambda element: np.multiply(npIndividual, element) x_cv = np.array(list(map(make_zero, con_val))) input_val_ind = [item_val, user_val, x_cv] val_predict = fully_trained_mlp.predict(input_val_ind) res_auc = roc_auc_score(y_val, val_predict) self.save_fitness(individual, res_auc) return res_auc def only_predict_get_auc(self, individual, context_val, y_val, fully_trained_mlp) -> float: """ Calculates the fitness of a given individual by predicting on the fully trained mlp :param individual: Ind to predict on :param context_val: Validation context :param y_val: Validation Target :param fully_trained_mlp: Fully trained classifier :return: AUC of the prediction """ if self.is_in_dict(individual): return self.get_fit(individual) npIndividual = np.array(individual) make_zero = lambda element: np.multiply(npIndividual, element) x_cv = np.array(list(map(make_zero, context_val))) val_predict = fully_trained_mlp.predict(x_cv) res_auc = roc_auc_score(y_val, val_predict) self.save_fitness(individual, res_auc) return res_auc def save_fitness(self, ind, val): """ :param ind: GA individual to save :param val: metric to save :return: None """ code = tuple(ind) self.fitness_dict[code] = val def is_in_dict(self, ind): """ :param ind: GA individual :return: True iff the given individual has a value store in the cache (past metric) """ code = tuple(ind) return code in self.fitness_dict def get_fit(self, ind): """ :param ind: GA individual :return: Saved metric """ code = tuple(ind) if code in self.fitness_dict: return self.fitness_dict[code] """ Not used on purpose, evaluation of individuals is done inside the selection function (for performance reasons). Instead of evaluating the entire generation, we evaluate only the individuals that were randomly selected in the tournament selection, thus saving unnecessary training of the MLP network . """ def evalInd(self, individual, METRIC): pass def update_progress_arrays(self, pop, fits, g, gen_start, outputpath): """ :param pop: current generation of GA individuals :param fits: list of fitness scores :param g: current generation number :param gen_start: time that this generation started :param outputpath: output to save the data :return: None """ length = len(pop) mean = sum(fits) / length sum2 = sum(x * x for x in fits) std = abs(sum2 / length - mean ** 2) ** 0.5 min_feat = min(fits) max_feat = max(fits) median_feat = np.median(fits) self.min_gen += [min_feat] self.max_gen += [max_feat] self.mean_gen += [mean] self.std_gen += [std] self.median_gen += [median_feat] self.time_gen += [time.time() - gen_start] self.save_all_data(g, outputpath) def get_train_test_val_split(arr): x_train, x_test = train_test_split(arr, test_size=0.3, shuffle=False) x_val, x_test = train_test_split(x_test, test_size=0.66, shuffle=False) return x_train, x_test, x_val if __name__ == '__main__': n_generations_ = 300 n_population_size = 100 sigma_for_init = 40 CXPB_ = 0.65 MUTPB_ = 1 ind_length_ = 661 mu_for_init = 0.25 * ind_length_ fs = FeatureSelectionGA(n_generations_, n_population_size, mu_for_init, sigma_for_init, CXPB_, MUTPB_, ind_length_) context_train_, _, context_val_ = get_train_test_val_split(np.load('../data_hero_ds/X.npy')) y_train_, _, y_val_ = get_train_test_val_split(np.load('../data_hero_ds/y.npy')) assert len(context_train_) == len(y_train_) and len(context_val_) == len(y_val_) mlp_path = '../fresh_mlp_data/HEARO_fresh_mlp_200_100_50_1.h5' fresh_weights_ = '../fresh_mlp_data/HEARO_fresh_weights_200_100_50_1.h5' output_folder_ = '../RESULTS_HERO_DS/GA_FAMILY/GA_RES_20_5_FULL_TRAIN/' fs.learn_full_train(output_folder_, context_train_, y_train_, context_val_, y_val_, mlp_path, fresh_weights_)
23,227
34.462595
115
py
GRAND-plus
GRAND-plus-main/model_mag.py
from __future__ import division from __future__ import print_function import sys import time import argparse import numpy as np import scipy.sparse as sp from precompute import propagation import torch import torch.nn.functional as F import torch.nn as nn import torch.optim as optim from torch_scatter import scatter from utils.data_loader import load_data, accuracy class MLP(nn.Module): def __init__(self, num_features, num_classes, hidden_size, nlayers, use_bn, input_dropout, hidden_dropout, node_norm): super().__init__() if nlayers == 1: self.embeds = torch.nn.Embedding(num_features, num_classes) self.fcs = nn.ModuleList([]) self.bns = nn.ModuleList([]) else: fcs = [] bns = [] self.embeds = torch.nn.Embedding(num_features, hidden_size) for i in range(nlayers - 2): fcs.append(nn.Linear(hidden_size, hidden_size, bias=True)) bns.append(nn.BatchNorm1d(hidden_size)) bns.append(nn.BatchNorm1d(hidden_size)) fcs.append(nn.Linear(hidden_size, num_classes, bias=True)) self.fcs = nn.ModuleList(fcs).cuda() self.bns = nn.ModuleList(bns).cuda() self.input_droprate = input_dropout self.hidden_droprate = hidden_dropout self.use_bn = use_bn self.node_norm = node_norm self.reset_param() def reset_param(self): for lin in self.fcs: lin.reset_parameters() def normalize(self, embedding): return embedding / (1e-12 + torch.norm(embedding, p=2, dim=-1, keepdim=True)) def emb(self, attr_idx, node_idx, attr_data): feat_embeds = self.embeds(attr_idx).cuda() feat_embeds = F.dropout(feat_embeds, self.input_droprate, training=self.training) dim_size = node_idx[-1] + 1 node_embeds = scatter(feat_embeds * attr_data[:, None], node_idx[:,None], dim=0, dim_size=dim_size, reduce='sum') node_s_sum = scatter(attr_data[:, None], node_idx[:, None], dim=0, dim_size=dim_size, reduce='sum') embs = node_embeds / (node_s_sum + 1e-10) return embs def forward(self, X): embs = X for fc, bn in zip(self.fcs, self.bns): embs = F.relu(embs) if self.node_norm: embs = self.normalize(embs) if self.use_bn: embs = bn(embs) embs = F.dropout(embs, self.hidden_droprate, training=self.training) embs = fc(embs) return embs class Grand_Plus(nn.Module): def __init__(self, num_features, num_classes, hidden_size, nlayers, use_bn, input_dropout, hidden_dropout, dropnode_rate=0.5, node_norm=False): super().__init__() self.mlp = MLP(num_features, num_classes, hidden_size, nlayers, use_bn, input_dropout, hidden_dropout, node_norm = node_norm) self.dropnode_rate = dropnode_rate def forward(self, X): logits = self.mlp(X) return logits def random_prop(self, feats, mat_scores, mat_idx, dropnode_rate): mat_scores = F.dropout(mat_scores, p=dropnode_rate, training=self.training) propagated_logits = scatter(feats * mat_scores[:, None], mat_idx[:, None], dim=0, dim_size=mat_idx[-1] + 1, reduce='sum') mat_sum_s = scatter(mat_scores[:,None], mat_idx[:,None], dim=0, dim_size=mat_idx[-1] + 1, reduce='sum') return propagated_logits / (mat_sum_s + 1e-12) def emb(self, attr_idx, node_idx, attr_data, cuda= True): feat_embeds = self.mlp.emb(attr_idx, node_idx, attr_data) return feat_embeds def iterate_minibatches_listinputs(index, batch_size, shuffle=False): numSamples = len(index) if shuffle: indices = np.arange(numSamples) np.random.shuffle(indices) for start_idx in range(0, numSamples, batch_size): if start_idx + batch_size > numSamples: end_idx = numSamples else: end_idx = start_idx + batch_size if shuffle: excerpt = indices[start_idx:end_idx] else: excerpt = slice(start_idx, end_idx) yield index[excerpt] def sample_unlabel(idx_unlabel, unlabel_batch_size, shuffle=False): unlabel_numSamples = idx_unlabel.shape[0] indices = np.arange(unlabel_numSamples) if shuffle: np.random.shuffle(indices) excerpt = indices[:unlabel_batch_size] return idx_unlabel[excerpt] def clip_grad_norm(params, max_norm): if max_norm > 0: return torch.nn.utils.clip_grad_norm_(params, max_norm) else: return torch.sqrt(sum(p.grad.data.norm() ** 2 for p in params if p.grad is not None)) def consis_loss(args, logps, tem, conf): ps = [torch.exp(p) for p in logps] sum_p = 0. for p in ps: sum_p = sum_p + p avg_p = sum_p/len(ps) sharp_p = (torch.pow(avg_p, 1./tem) / torch.sum(torch.pow(avg_p, 1./tem), dim=1, keepdim=True)).detach() loss = 0. for p in ps: if args.loss == 'kl': loss += torch.mean((-sharp_p * torch.log(p)).sum(1)[avg_p.max(1)[0] > conf]) elif args.loss == 'l2': loss += torch.mean((p-sharp_p).pow(2).sum(1)[avg_p.max(1)[0] > conf]) else: raise ValueError(f"Unknown loss type: {args.loss}") loss = loss/len(ps) return loss def valid(args, model, topk_adj, features, idx_val, labels, batch_size=100): model.eval() outputs = [] for idx in iterate_minibatches_listinputs(idx_val, batch_size): val_topk_adj = topk_adj[idx] source_idx, neighbor_idx = val_topk_adj.nonzero() mat_scores = val_topk_adj.data val_feat = features[neighbor_idx] mat_scores = torch.tensor(mat_scores, dtype=torch.float32) source_idx = torch.tensor(source_idx, dtype=torch.long) y_val = labels[idx] node_idx, attr_idx = val_feat.nonzero() attr_data = val_feat.data attr_data = torch.tensor(attr_data, dtype=torch.float32) node_idx = torch.tensor(node_idx, dtype=torch.long) attr_idx = torch.tensor(attr_idx, dtype=torch.long) if args.cuda: mat_scores = mat_scores.cuda() source_idx = source_idx.cuda() node_idx = node_idx.cuda() attr_data = attr_data.cuda() with torch.no_grad(): batch_emb = model.emb(attr_idx, node_idx, attr_data) batch_feat_aug = model.random_prop(batch_emb, mat_scores, source_idx, args.dropnode_rate)#.detach() output = model(batch_feat_aug) output = torch.log_softmax(output, dim=-1) outputs.append(output) outputs = torch.cat(outputs, dim=0) loss_test = F.nll_loss(outputs, labels[idx_val]) acc_test = accuracy(outputs, labels[idx_val]) return loss_test.item(), acc_test.item() def get_local_logits(model, attr_mat, batch_size=10000): device = next(model.parameters()).device nnodes = attr_mat.shape[0] logits = [] with torch.set_grad_enabled(False): for i in range(0, nnodes, batch_size): batch_attr = torch.FloatTensor(attr_mat[i:i + batch_size]).cuda() logits.append(model(batch_attr).to('cpu').numpy()) logits = np.row_stack(logits) return logits def predict(args, adj, features_np, model, idx_test, labels_org, mode='ppr', batch_size_logits=10000): model.eval() nprop = args.order embs, feats = [], [] nnodes = features_np.shape[0] with torch.set_grad_enabled(False): for i in range(0, nnodes, batch_size_logits): batch_feat = features_np[i: i + batch_size_logits] node_idx, attr_idx = batch_feat.nonzero() attr_data = batch_feat.data attr_data = torch.tensor(attr_data, dtype=torch.float32).cuda() node_idx = torch.tensor(node_idx, dtype=torch.long).cuda() attr_idx = torch.tensor(attr_idx, dtype=torch.long) batch_embs = model.emb(attr_idx, node_idx, attr_data).to('cpu').numpy() embs.append(batch_embs) embs = np.row_stack(embs) if mode == 'ppr': embs = args.alpha * embs embs_prop = embs.copy() deg_row = adj.sum(1).A1 deg_row_inv_alpha = np.asarray((1 - args.alpha) / np.maximum(deg_row, 1e-12)) for _ in range(nprop): embs = np.multiply(deg_row_inv_alpha[:, None], (adj.dot(embs))) embs_prop += embs feats.append(embs_prop.copy()) elif mode == 'avg': embs_prop = embs.copy() deg_row = adj.sum(1).A1 deg_row_inv = 1 / np.maximum(deg_row, 1e-12) for _ in range(nprop): embs = np.multiply(deg_row_inv[:,None], (adj.dot(embs))) embs_prop += embs embs_prop = embs_prop/(nprop + 1) feats.append(embs_prop) elif mode == 'single': deg_row = adj.sum(1).A1 deg_row_inv = 1 / np.maximum(deg_row, 1e-12) for _ in range(nprop): embs = np.multiply(deg_row_inv[:,None], (adj.dot(embs))) feats.append(embs) else: raise ValueError(f"Unknown propagation mode: {mode}") for feat in feats: logits = get_local_logits( model.mlp, feat, batch_size_logits) preds = logits.argmax(1) correct = np.equal(preds[idx_test], labels_org.cpu().numpy()[idx_test]).astype(float) correct = correct.sum() acc_test = correct / len(idx_test) print(acc_test) return acc_test def main_mag(args): args.cuda = not args.no_cuda and torch.cuda.is_available() torch.cuda.set_device(args.cuda_device) device = args.cuda_device torch.manual_seed(args.seed2) if args.cuda: torch.cuda.manual_seed(args.seed2) np.random.seed(args.seed2) dataset = args.dataset adj, features, labels, idx_train, idx_val, idx_test, _ = load_data(dataset_str=dataset, split_seed=args.seed1) unlabel_num = args.unlabel_num time_s1 = time.time() adj = adj + sp.eye(features.shape[0]) idx_sample = np.random.permutation( idx_test)[:unlabel_num] idx_unlabel = np.concatenate([idx_val, idx_sample]) idx_train_unlabel = np.concatenate( [idx_train, idx_unlabel]) indptr = np.array(adj.indptr, dtype=np.int32) indices = np.array(adj.indices, dtype=np.int32) graph = propagation.Graph(indptr, indices, args.seed2) row_idx = np.zeros((idx_train_unlabel.shape[0] * args.top_k), dtype=np.int32) col_idx = np.zeros((idx_train_unlabel.shape[0] * args.top_k), dtype=np.int32) mat_value = np.zeros((idx_train_unlabel.shape[0] * args.top_k), dtype=np.float64) if args.prop_mode == 'avg': coef = list(np.ones(args.order + 1, dtype=np.float64)) elif args.prop_mode == 'ppr': coef = [args.alpha] for i in range(args.order): coef.append(coef[-1] * (1-args.alpha)) elif args.prop_mode == 'single': coef = list(np.zeros(args.order + 1, dtype=np.float64)) coef[-1] = 1. else: raise ValueError(f"Unknown propagation mode: {args.prop_mode}") print(f"propagation matrix: {args.prop_mode}") coef = np.asarray(coef) / np.sum(coef) graph.gfpush_omp(idx_train_unlabel, row_idx, col_idx, mat_value, coef, args.rmax, args.top_k) #print(row_idx.astype(np.int32).max(), col_idx.astype(np.int32).max(), features.shape[0]) topk_adj = sp.coo_matrix((mat_value, (row_idx, col_idx)), ( features.shape[0], features.shape[0])) topk_adj = topk_adj.tocsr() time_preprocessing = time.time() - time_s1 print(f"preprocessing done, time: {time_preprocessing}") features_np = features n_class = labels.shape[1] labels = torch.LongTensor(np.argmax(labels, -1)) model = Grand_Plus(num_features=features.shape[1], num_classes=labels.max().item() + 1, hidden_size=args.hidden, nlayers=args.nlayers, use_bn = args.use_bn, input_dropout=args.input_droprate, hidden_dropout=args.hidden_droprate, dropnode_rate=args.dropnode_rate, node_norm = args.node_norm) optimizer = optim.Adam(model.parameters(), lr=args.lr, weight_decay=args.weight_decay) if args.cuda: labels = labels.cuda() t_begin = time.time() loss_values = [] acc_values = [] batch_time = [] bad_counter = 0 loss_best = np.inf acc_best = 0.0 loss_mn = np.inf acc_mx = 0.0 best_epoch = 0 num_batch = 0 for epoch in range(args.epochs): for train_index in iterate_minibatches_listinputs(idx_train, batch_size=args.batch_size, shuffle=True): batch_t_s = time.time() model.train() optimizer.zero_grad() unlabel_index_batch = sample_unlabel( idx_sample, args.unlabel_batch_size, shuffle=True) batch_index = np.concatenate((train_index, unlabel_index_batch)) batch_topk_adj = topk_adj[batch_index] source_idx, neighbor_idx = batch_topk_adj.nonzero() mat_scores = batch_topk_adj.data batch_feat = features[neighbor_idx]#.to(device) mat_scores = torch.tensor(mat_scores, dtype=torch.float32).to(device) source_idx = torch.tensor(source_idx, dtype=torch.long).to(device) y_train_batch = labels[train_index] node_idx, attr_idx = batch_feat.nonzero() attr_data = batch_feat.data attr_data = torch.tensor(attr_data, dtype=torch.float32).to(device) node_idx = torch.tensor(node_idx, dtype=torch.long).to(device) attr_idx = torch.tensor(attr_idx, dtype=torch.long) output_list = [] K = args.sample loss_train = 0. for i in range(K): batch_emb = model.emb(attr_idx, node_idx, attr_data) batch_feat_aug = model.random_prop(batch_emb, mat_scores, source_idx, args.dropnode_rate)#.detach() output_aug = model(batch_feat_aug) output_aug = torch.log_softmax(output_aug, dim=-1) output_list.append(output_aug[len(train_index):]) loss_train += F.nll_loss(output_aug[:len(train_index)], y_train_batch) loss_train = loss_train/K args.conf = 2./n_class loss_train += min(1.0 , float(num_batch)/args.warmup) * args.lam * consis_loss(args, output_list, args.tem, args.conf) acc_train = accuracy(output_aug[:len(train_index)], y_train_batch) loss_train.backward() grad_norm = clip_grad_norm(model.parameters(), args.clip_norm) optimizer.step() batch_time.append(time.time() - batch_t_s) if num_batch % args.eval_batch == 0: loss_val, acc_val = valid( args, model, topk_adj, features, idx_val, labels) loss_values.append(loss_val) acc_values.append(acc_val) if args.visible: print( f'epoch {epoch}, batch {num_batch}, validation loss {loss_val}, validation acc {acc_val}') if acc_values[-1] >= acc_mx: if args.stop_mode == 'acc' or (args.stop_mode == 'both' and loss_values[-1]<= loss_mn): loss_mn = loss_values[-1] acc_mx = acc_values[-1] best_epoch = epoch best_batch = num_batch torch.save(model.state_dict(), f"{args.model}_{dataset}.pkl") bad_counter = 0 else: bad_counter += 1 if bad_counter >= args.patience: if args.visible: print( f'Early stop! Min loss: {loss_mn}, Max accuracy: {acc_mx}, num batch: {num_batch} num epoch: {epoch}') break num_batch += 1 if bad_counter >= args.patience: break if args.visible: print( f'Optimization Finished! Min loss: {loss_mn}, Max accuracy: {acc_mx}, num batch: {num_batch} num epoch: {epoch}') if args.visible: print('Loading {}th epoch'.format(best_epoch)) model.load_state_dict(torch.load(f"{args.model}_{dataset}.pkl")) test_acc = predict(args, adj, features_np, model, idx_test, labels, mode = args.prop_mode) t_total = time.time() - time_s1 print("Total time elapsed: {:.4f}s".format(t_total)) print(f"Test Accuracy {test_acc}") return t_total, test_acc, np.mean(batch_time), num_batch
17,036
39.954327
147
py
GRAND-plus
GRAND-plus-main/model.py
from __future__ import division from __future__ import print_function import sys sys.path.append("..") import time import argparse import numpy as np import scipy.sparse as sp from precompute import propagation import torch import torch.nn.functional as F import torch.nn as nn import torch.optim as optim from torch_scatter import scatter from utils.data_loader import load_data, accuracy, totensor class MLP(nn.Module): def __init__(self, num_features, num_classes, hidden_size, nlayers, use_bn, input_dropout, hidden_dropout, node_norm): super().__init__() if nlayers == 1: fcs = [nn.Linear(num_features, num_classes, bias=True)] bns = [nn.BatchNorm1d(num_features)] else: fcs = [nn.Linear(num_features, hidden_size, bias=True)] bns = [nn.BatchNorm1d(num_features)] for i in range(nlayers - 2): fcs.append(nn.Linear(hidden_size, hidden_size, bias=True)) bns.append(nn.BatchNorm1d(hidden_size)) bns.append(nn.BatchNorm1d(hidden_size)) fcs.append(nn.Linear(hidden_size, num_classes, bias=True)) self.fcs = nn.ModuleList(fcs) self.bns = nn.ModuleList(bns) self.input_droprate = input_dropout self.hidden_droprate = hidden_dropout self.use_bn = use_bn self.node_norm = node_norm self.reset_param() def reset_param(self): for lin in self.fcs: lin.reset_parameters() def normalize(self, embedding): return embedding / (1e-12 + torch.norm(embedding, p=2, dim=-1, keepdim=True)) def forward(self, X): if self.node_norm: X = self.normalize(X).detach() if self.use_bn: X = self.bns[0](X) embs = F.dropout(X, self.input_droprate, training=self.training)#.detach() embs = self.fcs[0](embs) for fc, bn in zip(self.fcs[1:], self.bns[1:]): embs = F.relu(embs) if self.node_norm: embs = self.normalize(embs) if self.use_bn: embs = bn(embs) embs = F.dropout(embs, self.hidden_droprate, training=self.training) embs = fc(embs) return embs class Grand_Plus(nn.Module): def __init__(self, num_features, num_classes, hidden_size, nlayers, use_bn, input_dropout, hidden_dropout, dropnode_rate=0.5, node_norm=False): super().__init__() self.mlp = MLP(num_features, num_classes, hidden_size, nlayers, use_bn, input_dropout, hidden_dropout, node_norm = node_norm) self.dropnode_rate = dropnode_rate def forward(self, X): logits = self.mlp(X) return logits def random_prop(self, feats, mat_scores, mat_idx, dropnode_rate): mat_scores = F.dropout(mat_scores, p=dropnode_rate, training=self.training) propagated_logits = scatter(feats * mat_scores[:, None], mat_idx[:, None], dim=0, dim_size=mat_idx[-1] + 1, reduce='sum') mat_sum_s = scatter(mat_scores[:,None], mat_idx[:,None], dim=0, dim_size=mat_idx[-1] + 1, reduce='sum') return propagated_logits / (mat_sum_s + 1e-12) def iterate_minibatches_listinputs(index, batch_size, shuffle=False): numSamples = len(index) if shuffle: indices = np.arange(numSamples) np.random.shuffle(indices) for start_idx in range(0, numSamples, batch_size): if start_idx + batch_size > numSamples: end_idx = numSamples else: end_idx = start_idx + batch_size if shuffle: excerpt = indices[start_idx:end_idx] else: excerpt = slice(start_idx, end_idx) yield index[excerpt] def sample_unlabel(idx_unlabel, unlabel_batch_size, shuffle=False): unlabel_numSamples = idx_unlabel.shape[0] indices = np.arange(unlabel_numSamples) if shuffle: np.random.shuffle(indices) excerpt = indices[:unlabel_batch_size] return idx_unlabel[excerpt] def clip_grad_norm(params, max_norm): if max_norm > 0: return torch.nn.utils.clip_grad_norm_(params, max_norm) else: return torch.sqrt(sum(p.grad.data.norm() ** 2 for p in params if p.grad is not None)) def consis_loss(args, logps, tem, conf): ps = [torch.exp(p) for p in logps] sum_p = 0. for p in ps: sum_p = sum_p + p avg_p = sum_p/len(ps) sharp_p = (torch.pow(avg_p, 1./tem) / torch.sum(torch.pow(avg_p, 1./tem), dim=1, keepdim=True)).detach() loss = 0. for p in ps: if args.loss == 'kl': loss += torch.mean((-sharp_p * torch.log(p)).sum(1)[avg_p.max(1)[0] > conf]) elif args.loss == 'l2': loss += torch.mean((p-sharp_p).pow(2).sum(1)[avg_p.max(1)[0] > conf]) else: raise ValueError(f"Unknown loss type: {args.loss}") loss = loss/len(ps) return loss def valid(args, model, topk_adj, features, idx_val, labels, batch_size=10000): model.eval() outputs = [] for idx in iterate_minibatches_listinputs(idx_val, batch_size): val_topk_adj = topk_adj[idx] source_idx, neighbor_idx = val_topk_adj.nonzero() mat_scores = val_topk_adj.data val_feat = features[neighbor_idx] mat_scores = torch.tensor(mat_scores, dtype=torch.float32) source_idx = torch.tensor(source_idx, dtype=torch.long) y_val = labels[idx] if args.cuda: val_feat = val_feat.cuda() mat_scores = mat_scores.cuda() source_idx = source_idx.cuda() with torch.no_grad(): val_feat = model.random_prop(val_feat, mat_scores, source_idx, args.dropnode_rate).detach() output = model(val_feat) output = torch.log_softmax(output, dim=-1) outputs.append(output) outputs = torch.cat(outputs, dim=0) loss_test = F.nll_loss(outputs, labels[idx_val]) acc_test = accuracy(outputs, labels[idx_val]) return loss_test.item(), acc_test.item() def get_local_logits(model, attr_mat, batch_size=10000): device = next(model.parameters()).device nnodes = attr_mat.shape[0] logits = [] with torch.set_grad_enabled(False): for i in range(0, nnodes, batch_size): batch_attr = torch.FloatTensor(attr_mat[i:i + batch_size]).to(device) logits.append(model(batch_attr).to('cpu').numpy()) logits = np.row_stack(logits) return logits def predict(args, adj, features_np, model, idx_test, labels_org, mode='ppr', batch_size_logits=10000): model.eval() nprop = args.order feats = [] if mode == 'ppr': features_np = args.alpha * features_np features_np_prop = features_np.copy() deg_row = adj.sum(1).A1 deg_row_inv_alpha = np.asarray((1 - args.alpha) / np.maximum(deg_row, 1e-12)) for _ in range(nprop): features_np = np.multiply(deg_row_inv_alpha[:, None], (adj.dot(features_np))) features_np_prop += features_np feats.append(features_np_prop.copy()) elif mode == 'avg': features_np_prop = features_np.copy() deg_row = adj.sum(1).A1 deg_row_inv = 1 / np.maximum(deg_row, 1e-12) for _ in range(nprop): features_np = np.multiply(deg_row_inv[:,None], (adj.dot(features_np))) features_np_prop += features_np features_np_prop = features_np_prop/(nprop + 1) feats.append(features_np_prop) elif mode == 'single': deg_row = adj.sum(1).A1 deg_row_inv = 1 / np.maximum(deg_row, 1e-12) for _ in range(nprop): features_np = np.multiply(deg_row_inv[:,None], (adj.dot(features_np))) features_np_prop = features_np feats = [features_np_prop] else: raise ValueError(f"Unknown propagation mode: {mode}") for feat in feats: # print(feat[0]) logits = get_local_logits( model.mlp, feat, batch_size_logits) preds = logits.argmax(1) correct = np.equal(preds[idx_test], labels_org.cpu().numpy()[idx_test]).astype(float) correct = correct.sum() acc_test = correct / len(idx_test) print(acc_test) return acc_test def main(args): args.cuda = not args.no_cuda and torch.cuda.is_available() torch.cuda.set_device(args.cuda_device) device = args.cuda_device torch.manual_seed(args.seed2) if args.cuda: torch.cuda.manual_seed(args.seed2) np.random.seed(args.seed2) dataset = args.dataset adj, features, labels, idx_train, idx_val, idx_test, _ = load_data( dataset_str=dataset, split_seed=args.seed1) unlabel_num = args.unlabel_num time_s1 = time.time() adj = adj + sp.eye(features.shape[0]) idx_sample = np.random.permutation( idx_test)[:unlabel_num] idx_unlabel = np.concatenate([idx_val, idx_sample]) idx_train_unlabel = np.concatenate( [idx_train, idx_unlabel]) indptr = np.array(adj.indptr, dtype=np.int32) indices = np.array(adj.indices, dtype=np.int32) graph = propagation.Graph(indptr, indices, args.seed2) row_idx = np.zeros((idx_train_unlabel.shape[0] * args.top_k), dtype=np.int32) col_idx = np.zeros((idx_train_unlabel.shape[0] * args.top_k), dtype=np.int32) mat_value = np.zeros((idx_train_unlabel.shape[0] * args.top_k), dtype=np.float64) if args.prop_mode == 'avg': coef = list(np.ones(args.order + 1, dtype=np.float64)) elif args.prop_mode == 'ppr': coef = [args.alpha] for i in range(args.order): coef.append(coef[-1] * (1-args.alpha)) elif args.prop_mode == 'single': coef = list(np.zeros(args.order + 1, dtype=np.float64)) coef[-1] = 1.0 else: raise ValueError(f"Unknown propagation mode: {args.prop_mode}") print(f"propagation matrix: {args.prop_mode}") coef = np.asarray(coef) / np.sum(coef) graph.gfpush_omp(idx_train_unlabel, row_idx, col_idx, mat_value, coef, args.rmax, args.top_k) topk_adj = sp.coo_matrix((mat_value, (row_idx, col_idx)), ( features.shape[0], features.shape[0])) topk_adj = topk_adj.tocsr() time_preprocessing = time.time() - time_s1 print(f"preprocessing done, time: {time_preprocessing}") features_np = features features, labels = totensor(features, labels) n_class = labels.max().item() + 1 model = Grand_Plus(num_features=features.shape[1], num_classes=n_class, hidden_size=args.hidden, nlayers=args.nlayers, use_bn = args.use_bn, input_dropout=args.input_droprate, hidden_dropout=args.hidden_droprate, dropnode_rate=args.dropnode_rate, node_norm = args.node_norm) optimizer = optim.Adam(model.parameters(), lr=args.lr, weight_decay=args.weight_decay) if args.cuda: model.cuda() labels = labels.cuda() loss_values = [] acc_values = [] batch_time = [] bad_counter = 0 loss_mn = np.inf acc_mx = 0.0 best_epoch = 0 num_batch = 0 for epoch in range(args.epochs): for train_index in iterate_minibatches_listinputs(idx_train, batch_size=args.batch_size, shuffle=True): batch_t_s = time.time() model.train() optimizer.zero_grad() unlabel_index_batch = sample_unlabel( idx_sample, args.unlabel_batch_size, shuffle=True) batch_index = np.concatenate((train_index, unlabel_index_batch)) batch_topk_adj = topk_adj[batch_index] source_idx, neighbor_idx = batch_topk_adj.nonzero() mat_scores = batch_topk_adj.data batch_feat = features[neighbor_idx].to(device) mat_scores = torch.tensor(mat_scores, dtype=torch.float32).to(device) source_idx = torch.tensor(source_idx, dtype=torch.long).to(device) y_train_batch = labels[train_index] output_list = [] K = args.sample loss_train = 0. for i in range(K): batch_feat_aug = model.random_prop(batch_feat, mat_scores, source_idx, args.dropnode_rate).detach() output_aug = model(batch_feat_aug) output_aug = torch.log_softmax(output_aug, dim=-1) output_list.append(output_aug[len(train_index):]) loss_train += F.nll_loss(output_aug[:len(train_index)], y_train_batch) loss_train = loss_train/K args.conf = 2./n_class loss_train += min(args.lam, (args.lam * float(num_batch)/args.warmup)) * consis_loss(args, output_list, args.tem, args.conf) acc_train = accuracy(output_aug[:len(train_index)], y_train_batch) loss_train.backward() grad_norm = clip_grad_norm(model.parameters(), args.clip_norm) optimizer.step() batch_time.append(time.time() - batch_t_s) if num_batch % args.eval_batch == 0: loss_val, acc_val = valid( args, model, topk_adj, features, idx_val, labels, args.batch_size) loss_values.append(loss_val) acc_values.append(acc_val) if args.visible: print( f'epoch {epoch}, batch {num_batch}, validation loss {loss_val}, validation acc {acc_val}') if acc_values[-1] >= acc_mx: if args.stop_mode == 'acc' or (args.stop_mode == 'both' and loss_values[-1]<= loss_mn): loss_mn = loss_values[-1] acc_mx = acc_values[-1] best_epoch = epoch best_batch = num_batch torch.save(model.state_dict(), f"{args.model}_{dataset}.pkl") bad_counter = 0 else: bad_counter += 1 if bad_counter >= args.patience: if args.visible: print( f'Early stop! Min loss: {loss_mn}, Max accuracy: {acc_mx}, num batch: {num_batch} num epoch: {epoch}') break num_batch += 1 if bad_counter >= args.patience: break if args.visible: print( f'Optimization Finished! Min loss: {loss_mn}, Max accuracy: {acc_mx}, num batch: {num_batch} num epoch: {epoch}') if args.visible: print('Loading {}th epoch'.format(best_epoch)) model.load_state_dict(torch.load(f"{args.model}_{dataset}.pkl")) test_acc = predict(args, adj, features_np, model, idx_test, labels, mode = args.prop_mode) t_total = time.time() - time_s1 print("Total time elapsed: {:.4f}s".format(t_total)) print(f"Test Accuracy {test_acc}") return t_total, test_acc, np.mean(batch_time), num_batch
15,190
39.401596
147
py
GRAND-plus
GRAND-plus-main/utils/data_loader.py
import sys import pickle as pkl import networkx as nx import numpy as np import scipy.sparse as sp from sklearn.preprocessing import StandardScaler import torch from utils.make_dataset import get_dataset, get_train_val_test_split import os from sklearn import metrics def load_data(dataset_str='cora', split_seed=0, renormalize=False): """Load data.""" if os.path.exists("dataset/{}".format(dataset_str)): path = "dataset/{}".format(dataset_str) else: path = "dataset/" if dataset_str == 'aminer': adj = pkl.load(open(os.path.join(path, "{}.adj.sp.pkl".format(dataset_str)), "rb")) features = pkl.load( open(os.path.join(path, "{}.features.pkl".format(dataset_str)), "rb")) labels = pkl.load( open(os.path.join(path, "{}.labels.pkl".format(dataset_str)), "rb")) random_state = np.random.RandomState(split_seed) idx_train, idx_val, idx_test = get_train_val_test_split( random_state, labels, train_examples_per_class=20, val_examples_per_class=30) idx_unlabel = np.concatenate((idx_val, idx_test)) features = col_normalize(features) elif dataset_str in ['ms_academic_cs', 'ms_academic_phy', 'amazon_electronics_photo', 'amazon_electronics_computers', 'cora_full']: datapath = os.path.join(path, dataset_str + '.npz') adj, features, labels = get_dataset( dataset_str, datapath, True, train_examples_per_class=20, val_examples_per_class=30) random_state = np.random.RandomState(split_seed) idx_train, idx_val, idx_test = get_train_val_test_split( random_state, labels, train_examples_per_class=20, val_examples_per_class=30) idx_unlabel = np.concatenate((idx_val, idx_test)) features = features.todense() elif dataset_str in ['reddit']: adj = sp.load_npz(os.path.join(path, '{}_adj.npz'.format(dataset_str))) features = np.load(os.path.join(path, '{}_feat.npy'.format(dataset_str))) labels = np.load(os.path.join(path, '{}_labels.npy'.format(dataset_str))) print(labels.shape, list(np.sum(labels, axis=0))) random_state = np.random.RandomState(split_seed) idx_train, idx_val, idx_test = get_train_val_test_split( random_state, labels, train_examples_per_class=20, val_examples_per_class=30) idx_unlabel = np.concatenate((idx_val, idx_test)) print(dataset_str, features.shape) elif dataset_str in ['Amazon2M']: adj = sp.load_npz(os.path.join(path, '{}_adj.npz'.format(dataset_str))) features = np.load(os.path.join(path, '{}_feat.npy'.format(dataset_str))) labels = np.load(os.path.join(path, '{}_labels.npy'.format(dataset_str))) print(labels.shape, list(np.sum(labels, axis=0))) random_state = np.random.RandomState(split_seed) class_num = labels.shape[1] idx_train, idx_val, idx_test = get_train_val_test_split(random_state, labels, train_size=20* class_num, val_size=30 * class_num) idx_unlabel = np.concatenate((idx_val, idx_test)) elif dataset_str in ['mag_scholar_c', 'mag_scholar_f']: data_set = np.load(path + dataset_str + '.npz') adj_data = data_set['adj_matrix.data'] adj_indices = data_set['adj_matrix.indices'] adj_indptr = data_set['adj_matrix.indptr'] adj_shape = data_set['adj_matrix.shape'] feat_data = data_set['attr_matrix.data'] feat_indices = data_set['attr_matrix.indices'] feat_indptr = data_set['attr_matrix.indptr'] feat_shape = data_set['attr_matrix.shape'] labels_num = data_set['labels'] features = sp.csr_matrix((feat_data, feat_indices, feat_indptr), shape=feat_shape) adj = sp.csr_matrix((adj_data, adj_indices, adj_indptr), shape=adj_shape) random_state = np.random.RandomState(split_seed) label_count = labels_num.max() + 1 labels = np.eye(label_count)[labels_num] idx_train, idx_val, idx_test = get_train_val_test_split(random_state, labels, train_examples_per_class=20, val_examples_per_class=30) idx_unlabel = np.concatenate((idx_val, idx_test)) elif dataset_str in ['cora', 'citeseer', 'pubmed']: if os.path.exists("dataset/citation"): path = 'dataset/citation' names = ['x', 'y', 'tx', 'ty', 'allx', 'ally', 'graph'] objects = [] for i in range(len(names)): with open(os.path.join(path,"ind.{}.{}".format(dataset_str, names[i])), 'rb') as f: if sys.version_info > (3, 0): objects.append(pkl.load(f, encoding='latin1')) else: objects.append(pkl.load(f)) x, y, tx, ty, allx, ally, graph = tuple(objects) test_idx_reorder = parse_index_file( os.path.join(path, "ind.{}.test.index".format(dataset_str))) test_idx_range = np.sort(test_idx_reorder) if dataset_str == 'citeseer': test_idx_range_full = range( min(test_idx_reorder), max(test_idx_reorder)+1) tx_extended = sp.lil_matrix((len(test_idx_range_full), x.shape[1])) tx_extended[test_idx_range-min(test_idx_range), :] = tx tx = tx_extended ty_extended = np.zeros((len(test_idx_range_full), y.shape[1])) ty_extended[test_idx_range-min(test_idx_range), :] = ty ty = ty_extended features = sp.vstack((allx, tx)).tolil() features[test_idx_reorder, :] = features[test_idx_range, :] # normalize features = normalize(features) features = features.todense() adj = nx.adjacency_matrix(nx.from_dict_of_lists(graph)) # build symmetric adjacency matrix adj = adj + adj.T.multiply(adj.T > adj) - adj.multiply(adj.T > adj) labels = np.vstack((ally, ty)) labels[test_idx_reorder, :] = labels[test_idx_range, :] idx_train = np.arange(len(y)) idx_val = np.arange(len(y), len(y)+500) idx_test = np.asarray(test_idx_range.tolist()) idx_unlabel = np.arange(len(y), labels.shape[0]) else: raise NotImplementedError if renormalize: adj = adj + sp.eye(adj.shape[0]) D1 = np.array(adj.sum(axis=1))**(-0.5) D2 = np.array(adj.sum(axis=0))**(-0.5) D1 = sp.diags(D1[:, 0], format='csr') D2 = sp.diags(D2[0, :], format='csr') A = adj.dot(D1) A = D2.dot(A) adj = A return adj, features, labels, idx_train, idx_val, idx_test, idx_unlabel def totensor(features, labels): features = torch.FloatTensor(features) labels = torch.LongTensor(np.argmax(labels, -1)) return features, labels def parse_index_file(filename): """Parse index file.""" index = [] for line in open(filename): index.append(int(line.strip())) return index def accuracy(output, labels): preds = output.max(1)[1].type_as(labels) correct = preds.eq(labels).double() correct = correct.sum() return correct / len(labels) def normalize(mx): """Row-normalize sparse matrix""" rowsum = np.array(mx.sum(1)) r_inv = np.power(rowsum, -1).flatten() r_inv[np.isinf(r_inv)] = 0. r_mat_inv = sp.diags(r_inv) mx = r_mat_inv.dot(mx) return mx def col_normalize(mx): """Column-normalize sparse matrix""" scaler = StandardScaler() mx = scaler.fit_transform(mx) return mx
7,494
39.295699
141
py
BiGCN
BiGCN-master/tools/earlystopping2class.py
import numpy as np import torch class EarlyStopping: """Early stops the training if validation loss doesn't improve after a given patience.""" def __init__(self, patience=7, verbose=False): """ Args: patience (int): How long to wait after last time validation loss improved. Default: 7 verbose (bool): If True, prints a message for each validation loss improvement. Default: False """ self.patience = patience self.verbose = verbose self.counter = 0 self.best_score = None self.early_stop = False self.accs=0 self.F1=0 self.F2 = 0 self.F3 = 0 self.F4 = 0 self.val_loss_min = np.Inf def __call__(self, val_loss, accs,acc1,acc2,pre1,pre2,rec1,rec2,F1,F2,model,modelname,str): score = -val_loss if self.best_score is None: self.best_score = score self.accs = accs self.acc1=acc1 self.acc2=acc2 self.pre1=pre1 self.pre2=pre2 self.rec1=rec1 self.rec2=rec2 self.F1 = F1 self.F2 = F2 self.save_checkpoint(val_loss, model,modelname,str) elif score < self.best_score: self.counter += 1 # print('EarlyStopping counter: {} out of {}'.format(self.counter,self.patience)) if self.counter >= self.patience: self.early_stop = True print("BEST LOSS:{:.4f}| Accuracy: {:.4f}|acc1: {:.4f}|acc2: {:.4f}|pre1: {:.4f}|pre2: {:.4f}" "|rec1: {:.4f}|rec2: {:.4f}|F1: {:.4f}|F2: {:.4f}" .format(-self.best_score,self.accs,self.acc1,self.acc2,self.pre1,self.pre2,self.rec1,self.rec2,self.F1,self.F2)) else: self.best_score = score self.accs = accs self.acc1=acc1 self.acc2=acc2 self.pre1=pre1 self.pre2=pre2 self.rec1=rec1 self.rec2=rec2 self.F1 = F1 self.F2 = F2 self.save_checkpoint(val_loss, model,modelname,str) self.counter = 0 def save_checkpoint(self, val_loss, model,modelname,str): '''Saves model when validation loss decrease.''' # if self.verbose: # print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(self.val_loss_min,val_loss)) torch.save(model.state_dict(),modelname+str+'.m') self.val_loss_min = val_loss
2,599
36.681159
134
py
BiGCN
BiGCN-master/tools/earlystopping.py
import numpy as np import torch class EarlyStopping: """Early stops the training if validation loss doesn't improve after a given patience.""" def __init__(self, patience=7, verbose=False): """ Args: patience (int): How long to wait after last time validation loss improved. Default: 7 verbose (bool): If True, prints a message for each validation loss improvement. Default: False """ self.patience = patience self.verbose = verbose self.counter = 0 self.best_score = None self.early_stop = False self.accs=0 self.F1=0 self.F2 = 0 self.F3 = 0 self.F4 = 0 self.val_loss_min = np.Inf def __call__(self, val_loss, accs,F1,F2,F3,F4,model,modelname,str): score = -val_loss if self.best_score is None: self.best_score = score self.accs = accs self.F1 = F1 self.F2 = F2 self.F3 = F3 self.F4 = F4 self.save_checkpoint(val_loss, model,modelname,str) elif score < self.best_score: self.counter += 1 # print('EarlyStopping counter: {} out of {}'.format(self.counter,self.patience)) if self.counter >= self.patience: self.early_stop = True print("BEST Accuracy: {:.4f}|NR F1: {:.4f}|FR F1: {:.4f}|TR F1: {:.4f}|UR F1: {:.4f}" .format(self.accs,self.F1,self.F2,self.F3,self.F4)) else: self.best_score = score self.accs = accs self.F1 = F1 self.F2 = F2 self.F3 = F3 self.F4 = F4 self.save_checkpoint(val_loss, model,modelname,str) self.counter = 0 def save_checkpoint(self, val_loss, model,modelname,str): '''Saves model when validation loss decrease.''' # if self.verbose: # print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(self.val_loss_min,val_loss)) torch.save(model.state_dict(),modelname+str+'.m') self.val_loss_min = val_loss
2,208
35.816667
122
py
BiGCN
BiGCN-master/Process/dataset.py
import os import numpy as np import torch import random from torch.utils.data import Dataset from torch_geometric.data import Data class GraphDataset(Dataset): def __init__(self, fold_x, treeDic,lower=2, upper=100000, droprate=0, data_path=os.path.join('..','..', 'data', 'Weibograph')): self.fold_x = list(filter(lambda id: id in treeDic and len(treeDic[id]) >= lower and len(treeDic[id]) <= upper, fold_x)) self.treeDic = treeDic self.data_path = data_path self.droprate = droprate def __len__(self): return len(self.fold_x) def __getitem__(self, index): id =self.fold_x[index] data=np.load(os.path.join(self.data_path, id + ".npz"), allow_pickle=True) edgeindex = data['edgeindex'] if self.droprate > 0: row = list(edgeindex[0]) col = list(edgeindex[1]) length = len(row) poslist = random.sample(range(length), int(length * (1 - self.droprate))) poslist = sorted(poslist) row = list(np.array(row)[poslist]) col = list(np.array(col)[poslist]) new_edgeindex = [row, col] else: new_edgeindex = edgeindex return Data(x=torch.tensor(data['x'],dtype=torch.float32), edge_index=torch.LongTensor(new_edgeindex), y=torch.LongTensor([int(data['y'])]), root=torch.LongTensor(data['root']), rootindex=torch.LongTensor([int(data['rootindex'])])) def collate_fn(data): return data class BiGraphDataset(Dataset): def __init__(self, fold_x, treeDic,lower=2, upper=100000, tddroprate=0,budroprate=0, data_path=os.path.join('..','..', 'data', 'Weibograph')): self.fold_x = list(filter(lambda id: id in treeDic and len(treeDic[id]) >= lower and len(treeDic[id]) <= upper, fold_x)) self.treeDic = treeDic self.data_path = data_path self.tddroprate = tddroprate self.budroprate = budroprate def __len__(self): return len(self.fold_x) def __getitem__(self, index): id =self.fold_x[index] data=np.load(os.path.join(self.data_path, id + ".npz"), allow_pickle=True) edgeindex = data['edgeindex'] if self.tddroprate > 0: row = list(edgeindex[0]) col = list(edgeindex[1]) length = len(row) poslist = random.sample(range(length), int(length * (1 - self.tddroprate))) poslist = sorted(poslist) row = list(np.array(row)[poslist]) col = list(np.array(col)[poslist]) new_edgeindex = [row, col] else: new_edgeindex = edgeindex burow = list(edgeindex[1]) bucol = list(edgeindex[0]) if self.budroprate > 0: length = len(burow) poslist = random.sample(range(length), int(length * (1 - self.budroprate))) poslist = sorted(poslist) row = list(np.array(burow)[poslist]) col = list(np.array(bucol)[poslist]) bunew_edgeindex = [row, col] else: bunew_edgeindex = [burow,bucol] return Data(x=torch.tensor(data['x'],dtype=torch.float32), edge_index=torch.LongTensor(new_edgeindex),BU_edge_index=torch.LongTensor(bunew_edgeindex), y=torch.LongTensor([int(data['y'])]), root=torch.LongTensor(data['root']), rootindex=torch.LongTensor([int(data['rootindex'])])) class UdGraphDataset(Dataset): def __init__(self, fold_x, treeDic,lower=2, upper=100000, droprate=0, data_path=os.path.join('..','..','data', 'Weibograph')): self.fold_x = list(filter(lambda id: id in treeDic and len(treeDic[id]) >= lower and len(treeDic[id]) <= upper, fold_x)) self.treeDic = treeDic self.data_path = data_path self.droprate = droprate def __len__(self): return len(self.fold_x) def __getitem__(self, index): id =self.fold_x[index] data=np.load(os.path.join(self.data_path, id + ".npz"), allow_pickle=True) edgeindex = data['edgeindex'] row = list(edgeindex[0]) col = list(edgeindex[1]) burow = list(edgeindex[1]) bucol = list(edgeindex[0]) row.extend(burow) col.extend(bucol) if self.droprate > 0: length = len(row) poslist = random.sample(range(length), int(length * (1 - self.droprate))) poslist = sorted(poslist) row = list(np.array(row)[poslist]) col = list(np.array(col)[poslist]) new_edgeindex = [row, col] return Data(x=torch.tensor(data['x'],dtype=torch.float32), edge_index=torch.LongTensor(new_edgeindex), y=torch.LongTensor([int(data['y'])]), root=torch.LongTensor(data['root']), rootindex=torch.LongTensor([int(data['rootindex'])]))
4,936
40.141667
128
py
BiGCN
BiGCN-master/model/Weibo/BiGCN_Weibo.py
import sys,os sys.path.append(os.getcwd()) from Process.process import * import torch as th from torch_scatter import scatter_mean import torch.nn.functional as F import numpy as np from tools.earlystopping2class import EarlyStopping from torch_geometric.data import DataLoader from tqdm import tqdm from Process.rand5fold import * from tools.evaluate import * from torch_geometric.nn import GCNConv import copy class TDrumorGCN(th.nn.Module): def __init__(self,in_feats,hid_feats,out_feats): super(TDrumorGCN, self).__init__() self.conv1 = GCNConv(in_feats, hid_feats) self.conv2 = GCNConv(hid_feats+in_feats, out_feats) def forward(self, data): x, edge_index = data.x, data.edge_index x1=copy.copy(x.float()) x = self.conv1(x, edge_index) x2=copy.copy(x) rootindex = data.rootindex root_extend = th.zeros(len(data.batch), x1.size(1)).to(device) batch_size = max(data.batch) + 1 for num_batch in range(batch_size): index = (th.eq(data.batch, num_batch)) root_extend[index] = x1[rootindex[num_batch]] x = th.cat((x,root_extend), 1) x = F.relu(x) x = F.dropout(x, training=self.training) x = self.conv2(x, edge_index) x=F.relu(x) root_extend = th.zeros(len(data.batch), x2.size(1)).to(device) for num_batch in range(batch_size): index = (th.eq(data.batch, num_batch)) root_extend[index] = x2[rootindex[num_batch]] x = th.cat((x,root_extend), 1) x= scatter_mean(x, data.batch, dim=0) return x class BUrumorGCN(th.nn.Module): def __init__(self,in_feats,hid_feats,out_feats): super(BUrumorGCN, self).__init__() self.conv1 = GCNConv(in_feats, hid_feats) self.conv2 = GCNConv(hid_feats+in_feats, out_feats) def forward(self, data): x, edge_index = data.x, data.BU_edge_index x1 = copy.copy(x.float()) x = self.conv1(x, edge_index) x2 = copy.copy(x) rootindex = data.rootindex root_extend = th.zeros(len(data.batch), x1.size(1)).to(device) batch_size = max(data.batch) + 1 for num_batch in range(batch_size): index = (th.eq(data.batch, num_batch)) root_extend[index] = x1[rootindex[num_batch]] x = th.cat((x,root_extend), 1) x = F.relu(x) x = F.dropout(x, training=self.training) x = self.conv2(x, edge_index) x = F.relu(x) root_extend = th.zeros(len(data.batch), x2.size(1)).to(device) for num_batch in range(batch_size): index = (th.eq(data.batch, num_batch)) root_extend[index] = x2[rootindex[num_batch]] x = th.cat((x,root_extend), 1) x= scatter_mean(x, data.batch, dim=0) return x class Net(th.nn.Module): def __init__(self,in_feats,hid_feats,out_feats): super(Net, self).__init__() self.TDrumorGCN = TDrumorGCN(in_feats, hid_feats, out_feats) self.BUrumorGCN = BUrumorGCN(in_feats, hid_feats, out_feats) self.fc=th.nn.Linear((out_feats+hid_feats)*2,2) def forward(self, data): TD_x = self.TDrumorGCN(data) BU_x = self.BUrumorGCN(data) x = th.cat((BU_x,TD_x), 1) x=self.fc(x) x = F.log_softmax(x, dim=1) return x def train_GCN(treeDic, x_test, x_train,TDdroprate,BUdroprate,lr, weight_decay,patience,n_epochs,batchsize,dataname,iter): model = Net(5000,64,64).to(device) BU_params=list(map(id,model.BUrumorGCN.conv1.parameters())) BU_params += list(map(id, model.BUrumorGCN.conv2.parameters())) base_params=filter(lambda p:id(p) not in BU_params,model.parameters()) optimizer = th.optim.Adam([ {'params':base_params}, {'params':model.BUrumorGCN.conv1.parameters(),'lr':lr/5}, {'params': model.BUrumorGCN.conv2.parameters(), 'lr': lr/5} ], lr=lr, weight_decay=weight_decay) model.train() train_losses,val_losses,train_accs,val_accs = [],[],[],[] early_stopping = EarlyStopping(patience=patience, verbose=True) for epoch in range(n_epochs): traindata_list, testdata_list = loadBiData(dataname, treeDic, x_train, x_test, TDdroprate,BUdroprate) train_loader = DataLoader(traindata_list, batch_size=batchsize, shuffle=False, num_workers=10) test_loader = DataLoader(testdata_list, batch_size=batchsize, shuffle=True, num_workers=10) avg_loss,avg_acc = [],[] batch_idx = 0 tqdm_train_loader = tqdm(train_loader) for Batch_data in tqdm_train_loader: Batch_data.to(device) out_labels = model(Batch_data) loss = F.nll_loss(out_labels, Batch_data.y) optimizer.zero_grad() loss.backward() avg_loss.append(loss.item()) optimizer.step() _, pred = out_labels.max(dim=-1) correct = pred.eq(Batch_data.y).sum().item() train_acc = correct / len(Batch_data.y) avg_acc.append(train_acc) postfix = "Iter {:03d} | Epoch {:05d} | Batch{:02d} | Train_Loss {:.4f}| Train_Accuracy {:.4f}".format(iter, epoch, batch_idx, loss.item(), train_acc) tqdm_train_loader.set_postfix_str(postfix) batch_idx = batch_idx + 1 train_losses.append(np.mean(avg_loss)) train_accs.append(np.mean(avg_acc)) temp_val_losses,temp_val_accs,temp_val_Acc_all, temp_val_Acc1, temp_val_Prec1, temp_val_Recll1, temp_val_F1, \ temp_val_Acc2, temp_val_Prec2, temp_val_Recll2, temp_val_F2 = [],[],[], [], [], [], [], [], [], [], [] model.eval() tqdm_test_loader = tqdm(test_loader) for Batch_data in tqdm_test_loader: Batch_data.to(device) val_out = model(Batch_data) val_loss = F.nll_loss(val_out, Batch_data.y) temp_val_losses.append(val_loss.item()) _, val_pred = val_out.max(dim=1) correct = val_pred.eq(Batch_data.y).sum().item() val_acc = correct / len(Batch_data.y) Acc_all, Acc1, Prec1, Recll1, F1, Acc2, Prec2, Recll2, F2 = evaluationclass( val_pred, Batch_data.y) temp_val_Acc_all.append(Acc_all), temp_val_Acc1.append(Acc1), temp_val_Prec1.append( Prec1), temp_val_Recll1.append(Recll1), temp_val_F1.append(F1), \ temp_val_Acc2.append(Acc2), temp_val_Prec2.append(Prec2), temp_val_Recll2.append( Recll2), temp_val_F2.append(F2) temp_val_accs.append(val_acc) val_losses.append(np.mean(temp_val_losses)) val_accs.append(np.mean(temp_val_accs)) print("Epoch {:05d} | Val_Loss {:.4f}| Val_Accuracy {:.4f}".format(epoch, np.mean(temp_val_losses), np.mean(temp_val_accs))) res = ['acc:{:.4f}'.format(np.mean(temp_val_Acc_all)), 'C1:{:.4f},{:.4f},{:.4f},{:.4f}'.format(np.mean(temp_val_Acc1), np.mean(temp_val_Prec1), np.mean(temp_val_Recll1), np.mean(temp_val_F1)), 'C2:{:.4f},{:.4f},{:.4f},{:.4f}'.format(np.mean(temp_val_Acc2), np.mean(temp_val_Prec2), np.mean(temp_val_Recll2), np.mean(temp_val_F2))] print('results:', res) early_stopping(np.mean(temp_val_losses), np.mean(temp_val_Acc_all), np.mean(temp_val_Acc1), np.mean(temp_val_Acc2), np.mean(temp_val_Prec1), np.mean(temp_val_Prec2), np.mean(temp_val_Recll1), np.mean(temp_val_Recll2), np.mean(temp_val_F1), np.mean(temp_val_F2), model, 'BiGCN', "weibo") accs = np.mean(temp_val_Acc_all) acc1 = np.mean(temp_val_Acc1) acc2 = np.mean(temp_val_Acc2) pre1 = np.mean(temp_val_Prec1) pre2 = np.mean(temp_val_Prec2) rec1 = np.mean(temp_val_Recll1) rec2 = np.mean(temp_val_Recll2) F1 = np.mean(temp_val_F1) F2 = np.mean(temp_val_F2) if early_stopping.early_stop: print("Early stopping") accs = early_stopping.accs acc1 = early_stopping.acc1 acc2 = early_stopping.acc2 pre1 = early_stopping.pre1 pre2 = early_stopping.pre2 rec1 = early_stopping.rec1 rec2 = early_stopping.rec2 F1 = early_stopping.F1 F2 = early_stopping.F2 break return train_losses, val_losses, train_accs, val_accs, accs, acc1, pre1, rec1, F1, acc2, pre2, rec2, F2 lr=0.0005 weight_decay=1e-4 patience=10 n_epochs=200 batchsize=16 tddroprate=0 budroprate=0 datasetname="Weibo" iterations=int(sys.argv[1]) model="BiGCN" device = th.device('cuda:1' if th.cuda.is_available() else 'cpu') test_accs,ACC1,ACC2,PRE1,PRE2,REC1,REC2,F1,F2 = [],[],[],[],[],[],[],[],[] for iter in range(iterations): fold0_x_test, fold0_x_train,\ fold1_x_test, fold1_x_train, \ fold2_x_test, fold2_x_train, \ fold3_x_test, fold3_x_train, \ fold4_x_test, fold4_x_train = load5foldData(datasetname) treeDic=loadTree(datasetname) train_losses, val_losses, train_accs, val_accs, accs_0, acc1_0, pre1_0, rec1_0, F1_0, acc2_0, pre2_0, rec2_0, F2_0 = train_GCN(treeDic, fold0_x_test, fold0_x_train, tddroprate,budroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) train_losses, val_losses, train_accs, val_accs,accs_1, acc1_1, pre1_1, rec1_1, F1_1, acc2_1, pre2_1, rec2_1, F2_1 = train_GCN(treeDic, fold1_x_test, fold1_x_train, tddroprate,budroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) train_losses, val_losses, train_accs, val_accs, accs_2, acc1_2, pre1_2, rec1_2, F1_2, acc2_2, pre2_2, rec2_2, F2_2 = train_GCN(treeDic, fold2_x_test, fold2_x_train, tddroprate,budroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) train_losses, val_losses, train_accs, val_accs, accs_3, acc1_3, pre1_3, rec1_3, F1_3, acc2_3, pre2_3, rec2_3, F2_3 = train_GCN(treeDic, fold3_x_test, fold3_x_train, tddroprate,budroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) train_losses, val_losses, train_accs, val_accs, accs_4, acc1_4, pre1_4, rec1_4, F1_4, acc2_4, pre2_4, rec2_4, F2_4 = train_GCN(treeDic, fold4_x_test, fold4_x_train, tddroprate,budroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) test_accs.append((accs_0+accs_1+accs_2+accs_3+accs_4)/5) ACC1.append((acc1_0 + acc1_1 + acc1_2 + acc1_3 + acc1_4) / 5) ACC2.append((acc2_0 + acc2_1 +acc2_2 + acc2_3 +acc2_4) / 5) PRE1.append((pre1_0 + pre1_1 + pre1_2 + pre1_3 + pre1_4) / 5) PRE2.append((pre2_0 + pre2_1 + pre2_2 + pre2_3 + pre2_4) / 5) REC1.append((rec1_0 + rec1_1 + rec1_2 + rec1_3 + rec1_4) / 5) REC2.append((rec2_0 + rec2_1 + rec2_2 + rec2_3 + rec2_4) / 5) F1.append((F1_0+F1_1+F1_2+F1_3+F1_4)/5) F2.append((F2_0 + F2_1 + F2_2 + F2_3 + F2_4) / 5) print("weibo:|Total_Test_ Accuracy: {:.4f}|acc1: {:.4f}|acc2: {:.4f}|pre1: {:.4f}|pre2: {:.4f}" "|rec1: {:.4f}|rec2: {:.4f}|F1: {:.4f}|F2: {:.4f}".format(sum(test_accs) / iterations, sum(ACC1) / iterations, sum(ACC2) / iterations, sum(PRE1) / iterations, sum(PRE2) /iterations, sum(REC1) / iterations, sum(REC2) / iterations, sum(F1) / iterations, sum(F2) / iterations))
16,447
58.379061
168
py
BiGCN
BiGCN-master/model/Twitter/BiGCN_Twitter.py
import sys,os sys.path.append(os.getcwd()) from Process.process import * import torch as th from torch_scatter import scatter_mean import torch.nn.functional as F import numpy as np from tools.earlystopping import EarlyStopping from torch_geometric.data import DataLoader from tqdm import tqdm from Process.rand5fold import * from tools.evaluate import * from torch_geometric.nn import GCNConv import copy class TDrumorGCN(th.nn.Module): def __init__(self,in_feats,hid_feats,out_feats): super(TDrumorGCN, self).__init__() self.conv1 = GCNConv(in_feats, hid_feats) self.conv2 = GCNConv(hid_feats+in_feats, out_feats) def forward(self, data): x, edge_index = data.x, data.edge_index x1=copy.copy(x.float()) x = self.conv1(x, edge_index) x2=copy.copy(x) rootindex = data.rootindex root_extend = th.zeros(len(data.batch), x1.size(1)).to(device) batch_size = max(data.batch) + 1 for num_batch in range(batch_size): index = (th.eq(data.batch, num_batch)) root_extend[index] = x1[rootindex[num_batch]] x = th.cat((x,root_extend), 1) x = F.relu(x) x = F.dropout(x, training=self.training) x = self.conv2(x, edge_index) x = F.relu(x) root_extend = th.zeros(len(data.batch), x2.size(1)).to(device) for num_batch in range(batch_size): index = (th.eq(data.batch, num_batch)) root_extend[index] = x2[rootindex[num_batch]] x = th.cat((x,root_extend), 1) x= scatter_mean(x, data.batch, dim=0) return x class BUrumorGCN(th.nn.Module): def __init__(self,in_feats,hid_feats,out_feats): super(BUrumorGCN, self).__init__() self.conv1 = GCNConv(in_feats, hid_feats) self.conv2 = GCNConv(hid_feats+in_feats, out_feats) def forward(self, data): x, edge_index = data.x, data.BU_edge_index x1 = copy.copy(x.float()) x = self.conv1(x, edge_index) x2 = copy.copy(x) rootindex = data.rootindex root_extend = th.zeros(len(data.batch), x1.size(1)).to(device) batch_size = max(data.batch) + 1 for num_batch in range(batch_size): index = (th.eq(data.batch, num_batch)) root_extend[index] = x1[rootindex[num_batch]] x = th.cat((x,root_extend), 1) x = F.relu(x) x = F.dropout(x, training=self.training) x = self.conv2(x, edge_index) x = F.relu(x) root_extend = th.zeros(len(data.batch), x2.size(1)).to(device) for num_batch in range(batch_size): index = (th.eq(data.batch, num_batch)) root_extend[index] = x2[rootindex[num_batch]] x = th.cat((x,root_extend), 1) x= scatter_mean(x, data.batch, dim=0) return x class Net(th.nn.Module): def __init__(self,in_feats,hid_feats,out_feats): super(Net, self).__init__() self.TDrumorGCN = TDrumorGCN(in_feats, hid_feats, out_feats) self.BUrumorGCN = BUrumorGCN(in_feats, hid_feats, out_feats) self.fc=th.nn.Linear((out_feats+hid_feats)*2,4) def forward(self, data): TD_x = self.TDrumorGCN(data) BU_x = self.BUrumorGCN(data) x = th.cat((BU_x,TD_x), 1) x=self.fc(x) x = F.log_softmax(x, dim=1) return x def train_GCN(treeDic, x_test, x_train,TDdroprate,BUdroprate,lr, weight_decay,patience,n_epochs,batchsize,dataname,iter): model = Net(5000,64,64).to(device) BU_params=list(map(id,model.BUrumorGCN.conv1.parameters())) BU_params += list(map(id, model.BUrumorGCN.conv2.parameters())) base_params=filter(lambda p:id(p) not in BU_params,model.parameters()) optimizer = th.optim.Adam([ {'params':base_params}, {'params':model.BUrumorGCN.conv1.parameters(),'lr':lr/5}, {'params': model.BUrumorGCN.conv2.parameters(), 'lr': lr/5} ], lr=lr, weight_decay=weight_decay) model.train() train_losses = [] val_losses = [] train_accs = [] val_accs = [] early_stopping = EarlyStopping(patience=patience, verbose=True) for epoch in range(n_epochs): traindata_list, testdata_list = loadBiData(dataname, treeDic, x_train, x_test, TDdroprate,BUdroprate) train_loader = DataLoader(traindata_list, batch_size=batchsize, shuffle=True, num_workers=5) test_loader = DataLoader(testdata_list, batch_size=batchsize, shuffle=True, num_workers=5) avg_loss = [] avg_acc = [] batch_idx = 0 tqdm_train_loader = tqdm(train_loader) for Batch_data in tqdm_train_loader: Batch_data.to(device) out_labels= model(Batch_data) finalloss=F.nll_loss(out_labels,Batch_data.y) loss=finalloss optimizer.zero_grad() loss.backward() avg_loss.append(loss.item()) optimizer.step() _, pred = out_labels.max(dim=-1) correct = pred.eq(Batch_data.y).sum().item() train_acc = correct / len(Batch_data.y) avg_acc.append(train_acc) print("Iter {:03d} | Epoch {:05d} | Batch{:02d} | Train_Loss {:.4f}| Train_Accuracy {:.4f}".format(iter,epoch, batch_idx, loss.item(), train_acc)) batch_idx = batch_idx + 1 train_losses.append(np.mean(avg_loss)) train_accs.append(np.mean(avg_acc)) temp_val_losses = [] temp_val_accs = [] temp_val_Acc_all, temp_val_Acc1, temp_val_Prec1, temp_val_Recll1, temp_val_F1, \ temp_val_Acc2, temp_val_Prec2, temp_val_Recll2, temp_val_F2, \ temp_val_Acc3, temp_val_Prec3, temp_val_Recll3, temp_val_F3, \ temp_val_Acc4, temp_val_Prec4, temp_val_Recll4, temp_val_F4 = [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [] model.eval() tqdm_test_loader = tqdm(test_loader) for Batch_data in tqdm_test_loader: Batch_data.to(device) val_out = model(Batch_data) val_loss = F.nll_loss(val_out, Batch_data.y) temp_val_losses.append(val_loss.item()) _, val_pred = val_out.max(dim=1) correct = val_pred.eq(Batch_data.y).sum().item() val_acc = correct / len(Batch_data.y) Acc_all, Acc1, Prec1, Recll1, F1, Acc2, Prec2, Recll2, F2, Acc3, Prec3, Recll3, F3, Acc4, Prec4, Recll4, F4 = evaluation4class( val_pred, Batch_data.y) temp_val_Acc_all.append(Acc_all), temp_val_Acc1.append(Acc1), temp_val_Prec1.append( Prec1), temp_val_Recll1.append(Recll1), temp_val_F1.append(F1), \ temp_val_Acc2.append(Acc2), temp_val_Prec2.append(Prec2), temp_val_Recll2.append( Recll2), temp_val_F2.append(F2), \ temp_val_Acc3.append(Acc3), temp_val_Prec3.append(Prec3), temp_val_Recll3.append( Recll3), temp_val_F3.append(F3), \ temp_val_Acc4.append(Acc4), temp_val_Prec4.append(Prec4), temp_val_Recll4.append( Recll4), temp_val_F4.append(F4) temp_val_accs.append(val_acc) val_losses.append(np.mean(temp_val_losses)) val_accs.append(np.mean(temp_val_accs)) print("Epoch {:05d} | Val_Loss {:.4f}| Val_Accuracy {:.4f}".format(epoch, np.mean(temp_val_losses), np.mean(temp_val_accs))) res = ['acc:{:.4f}'.format(np.mean(temp_val_Acc_all)), 'C1:{:.4f},{:.4f},{:.4f},{:.4f}'.format(np.mean(temp_val_Acc1), np.mean(temp_val_Prec1), np.mean(temp_val_Recll1), np.mean(temp_val_F1)), 'C2:{:.4f},{:.4f},{:.4f},{:.4f}'.format(np.mean(temp_val_Acc2), np.mean(temp_val_Prec2), np.mean(temp_val_Recll2), np.mean(temp_val_F2)), 'C3:{:.4f},{:.4f},{:.4f},{:.4f}'.format(np.mean(temp_val_Acc3), np.mean(temp_val_Prec3), np.mean(temp_val_Recll3), np.mean(temp_val_F3)), 'C4:{:.4f},{:.4f},{:.4f},{:.4f}'.format(np.mean(temp_val_Acc4), np.mean(temp_val_Prec4), np.mean(temp_val_Recll4), np.mean(temp_val_F4))] print('results:', res) early_stopping(np.mean(temp_val_losses), np.mean(temp_val_accs), np.mean(temp_val_F1), np.mean(temp_val_F2), np.mean(temp_val_F3), np.mean(temp_val_F4), model, 'BiGCN', dataname) accs =np.mean(temp_val_accs) F1 = np.mean(temp_val_F1) F2 = np.mean(temp_val_F2) F3 = np.mean(temp_val_F3) F4 = np.mean(temp_val_F4) if early_stopping.early_stop: print("Early stopping") accs=early_stopping.accs F1=early_stopping.F1 F2 = early_stopping.F2 F3 = early_stopping.F3 F4 = early_stopping.F4 break return train_losses , val_losses ,train_accs, val_accs,accs,F1,F2,F3,F4 lr=0.0005 weight_decay=1e-4 patience=10 n_epochs=200 batchsize=128 TDdroprate=0.2 BUdroprate=0.2 datasetname=sys.argv[1] #"Twitter15"、"Twitter16" iterations=int(sys.argv[2]) model="GCN" device = th.device('cuda:0' if th.cuda.is_available() else 'cpu') test_accs = [] NR_F1 = [] FR_F1 = [] TR_F1 = [] UR_F1 = [] for iter in range(iterations): fold0_x_test, fold0_x_train, \ fold1_x_test, fold1_x_train, \ fold2_x_test, fold2_x_train, \ fold3_x_test, fold3_x_train, \ fold4_x_test,fold4_x_train = load5foldData(datasetname) treeDic=loadTree(datasetname) train_losses, val_losses, train_accs, val_accs0, accs0, F1_0, F2_0, F3_0, F4_0 = train_GCN(treeDic, fold0_x_test, fold0_x_train, TDdroprate,BUdroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) train_losses, val_losses, train_accs, val_accs1, accs1, F1_1, F2_1, F3_1, F4_1 = train_GCN(treeDic, fold1_x_test, fold1_x_train, TDdroprate,BUdroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) train_losses, val_losses, train_accs, val_accs2, accs2, F1_2, F2_2, F3_2, F4_2 = train_GCN(treeDic, fold2_x_test, fold2_x_train, TDdroprate,BUdroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) train_losses, val_losses, train_accs, val_accs3, accs3, F1_3, F2_3, F3_3, F4_3 = train_GCN(treeDic, fold3_x_test, fold3_x_train, TDdroprate,BUdroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) train_losses, val_losses, train_accs, val_accs4, accs4, F1_4, F2_4, F3_4, F4_4 = train_GCN(treeDic, fold4_x_test, fold4_x_train, TDdroprate,BUdroprate, lr, weight_decay, patience, n_epochs, batchsize, datasetname, iter) test_accs.append((accs0+accs1+accs2+accs3+accs4)/5) NR_F1.append((F1_0+F1_1+F1_2+F1_3+F1_4)/5) FR_F1.append((F2_0 + F2_1 + F2_2 + F2_3 + F2_4) / 5) TR_F1.append((F3_0 + F3_1 + F3_2 + F3_3 + F3_4) / 5) UR_F1.append((F4_0 + F4_1 + F4_2 + F4_3 + F4_4) / 5) print("Total_Test_Accuracy: {:.4f}|NR F1: {:.4f}|FR F1: {:.4f}|TR F1: {:.4f}|UR F1: {:.4f}".format( sum(test_accs) / iterations, sum(NR_F1) /iterations, sum(FR_F1) /iterations, sum(TR_F1) / iterations, sum(UR_F1) / iterations))
15,749
54.65371
139
py
segodsidb
segodsidb-main/setup.py
#!/usr/bin/env python # -*- coding: utf-8 -*- import setuptools import unittest # Read the contents of the README file from os import path this_directory = path.abspath(path.dirname(__file__)) with open(path.join(this_directory, 'README.md'), encoding='utf-8') as f: long_description = f.read() setuptools.setup(name='torchseg', version='0.1.0', description='Python module to segment hyperspectral dental images.', author='Luis C. Garcia-Peraza Herrera', author_email='luiscarlos.gph@gmail.com', license='MIT License', url='https://github.com/luiscarlosgph/dentalseg', packages=[ 'torchseg', 'torchseg.base', 'torchseg.config', 'torchseg.data_loader', 'torchseg.logger', 'torchseg.model', 'torchseg.machine', 'torchseg.utils', 'torchseg.visualization', ], package_dir={ 'torchseg': 'src', 'torchseg.base': 'src/base', 'torchseg.config': 'src/config', 'torchseg.data_loader': 'src/data_loader', 'torchseg.logger': 'src/logger', 'torchseg.model': 'src/model', 'torchseg.machine': 'src/machine', 'torchseg.utils': 'src/utils', 'torchseg.visualization': 'src/visualization', }, install_requires = [ 'numpy', 'opencv-python', 'torch', 'torchvision', 'tensorboard', 'tqdm', 'tifffile', 'monai', 'pandas', 'colour-science', 'matplotlib', ], long_description=long_description, long_description_content_type='text/markdown', test_suite = 'test', )
1,699
28.310345
73
py
segodsidb
segodsidb-main/src/generate_odsi_db_split.py
""" @brief Script that generates a convenient split of training and testing images for the ODSI-DB dataset. Then you can split the training fold into training and validation however you want, but this is expected to be done in your dataloader. @details A convenient dataset must have at least pixels of all the classes in all the splits. An additional desirable feature is to have an even balance of pixels per class in all the splits. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 1 Jun 2021. """ import argparse import numpy as np import os import shutil import sys import tqdm import json import random # My imports import torchseg.data_loader import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' classnames = torchseg.data_loader.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-o': 'Path to the output folder where the splits will be saved \ (required: True)', '-t': 'Portion of images to be used for training (default: 0.9)', '-f': 'Number of cross-validation folds (default: 1).', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-o', '--output', required=True, type=str, help=help('-o')) parser.add_argument('-t', '--train', default=0.9, type=float, help=help('-t')) parser.add_argument('-f', '--folds', default=1, type=int, help=help('-f')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): # Input directory must exist if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') # Output directory should not exist, we will create it if os.path.exists(args.output): raise RuntimeError('[ERROR] Output directory already exists.') return args def read_image(self, path, bands=np.linspace(450, 950, 51)): """ @param[in] path String with the path to the hyperspectral image. @returns an image with shape (C, H, W) containing values in the range [0, 1]. """ # Read raw TIFF file raw_tiff = torchseg.data_loader.read_stiff(path, silent=True, rgb_only=False) # In the ODSI-DB dataset there are images with 51 bands # (450-950nm, both inclusive) and images with 204 bands # (400-1000nm) # Create empty image h = raw_tiff[0].shape[0] w = raw_tiff[0].shape[1] chan = bands.shape[0] im = np.empty((chan, h, w), dtype=np.float32) # Populate image with the wanted bands (bands parameter of this # function) raw_im = raw_tiff[0].transpose((2, 0, 1)).astype(np.float32) wavelengths = raw_tiff[1] for idx, wl in enumerate(bands.tolist()): matched_idx = np.abs(wavelengths - wl).argmin() im[idx, ...] = raw_im[matched_idx, ...].copy() return im def read_label(path): """ @param[in] path String with the path to the annotation file. @returns a numpy.ndarray with shape (C, H, W) containing values in the range [0, 1]. """ raw_tiff = torchseg.data_loader.read_mtiff(path) # Get the shape of the labels (which should be identical to the # height and width of the image) shape = raw_tiff[list(raw_tiff.keys())[0]].shape # Create tensor of labels n_classes = len(classnames) label = np.zeros((n_classes, *shape), dtype=np.float32) # Build dictionaries to convert quickly from index to class name # and vice versa idx2class = classnames class2idx = {y: x for x, y in idx2class.items()} # Populate the binary array for each class for k, gt in raw_tiff.items(): if k not in class2idx: raise ValueError('[ERROR] Unknown <' + k + '> label.') # Find the index of the class with name 'k' idx = class2idx[k] # Set the label for the class with name 'k' label[idx] = gt.astype(np.float32) return label def read_files(path): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def generate_random_split(images, labels, train_ratio, force_train=['Fibroma', 'Makeup', 'Malignant lesion']): """ @brief Generates a split of the data for training and testing. @details Every element in the list 'images' must correspond to a label in the same index of the list 'labels'. @param[in] images List of paths to the images. @param[in] labels List of paths to the labels. @param[in] train_ratio Portion of the images that will go for training. @param[in] force_train List of classes. Images containing pixels of these classes will be automatically put in the training set. The reason for this is that there is only one image in the dataset containing pixels of these classes. @returns a dictionary of dictionaries. Keys 'train' and 'test' in the upper level. Keys 'images' and 'labels' in the lower level. """ np.random.seed() class2idx = {y: x for x, y in classnames.items()} split_dic = { 'train': { 'images': [], 'labels': [], }, 'test': { 'images': [], 'labels': [], } } # Make a list of the indices of all the images remaining = list(range(len(images))) # Make a dictionary from each class to the indices of the images that # contain pixels of such class class2img = {c: [] for c in range(len(classnames))} for im_idx in remaining: label_path = labels[im_idx] label = read_label(label_path).sum(axis=(1, 2)).astype(np.int64) for c in label.nonzero()[0].tolist(): class2img[c].append(im_idx) # Put an image of each class in the training set for c in class2img: random.shuffle(class2img[c]) im_idx = class2img[c][0] if im_idx in remaining: split_dic['train']['images'].append(images[im_idx]) split_dic['train']['labels'].append(labels[im_idx]) remaining.remove(im_idx) # Split the rest of the images according to the given probability while remaining: im_idx = remaining.pop() dst = 'train' if np.random.binomial(1, train_ratio) else 'test' split_dic[dst]['images'].append(images[im_idx]) split_dic[dst]['labels'].append(labels[im_idx]) return split_dic def validate_random_split(split_dic, force_train=['Fibroma', 'Makeup', 'Malignant lesion']): """ @brief Checks that all the classes are represented in both training and testing splits. @param[in] split_dic Dictionary of dictionaries. Keys 'train' and 'test' in the upper level. Keys 'images' and 'labels' in the lower level. @returns True if the split is good. Otherwise, returns False. """ classes = len(classnames) train_classes = np.zeros((classes,), dtype=np.int64) test_classes = np.zeros((classes,), dtype=np.int64) # Find the classes present in the training set for label_path in split_dic['train']['labels']: train_classes += read_label(label_path).sum(axis=(1, 2)).astype(np.int64) # Find the classes present in the testing set for label_path in split_dic['test']['labels']: test_classes += read_label(label_path).sum(axis=(1, 2)).astype(np.int64) # All the classes in the testing set must be present in the training set for idx, count in enumerate(test_classes): if count > 0 and train_classes[idx] == 0: return False return True def copy_into_dir(file_list, dst_dir): """ @brief Copies the files in the list into the provided destination directory. @param[in] file_list List of paths to files. @param[in] dst_dir Path to the destination directory. @returns nothing. """ for src_path in file_list: filename = os.path.basename(src_path) dst_path = os.path.join(dst_dir, filename) shutil.copyfile(src_path, dst_path) def copy_data(split_dic, train_path, test_path): """ @brief Gets a split of training/testing files and copies them into separate folders. @param[in] split_dic Dictionary of dictionaries. Keys 'train' and 'test' in the upper level. Keys 'images' and 'labels' in the lower level. @returns nothing. """ copy_into_dir(split_dic['train']['images'], train_path) copy_into_dir(split_dic['train']['labels'], train_path) copy_into_dir(split_dic['test']['images'], test_path) copy_into_dir(split_dic['test']['labels'], test_path) def compute_dataset_stats(images, labels): # Gather data for report no_classes = len(classnames) no_pixels = np.zeros((no_classes,), dtype=np.int64) no_images = np.zeros((no_classes,), dtype=np.int64) # Find the number of pixels and images per class for label_path in labels: label = read_label(label_path).sum(axis=(1, 2)).astype(np.int64) no_pixels += label label[label > 0] = 1 no_images += label return no_pixels, no_images def print_dataset_stats(no_pixels, no_images, labels): """ @brief Prints the number of pixels and images per class. @param[in] no_pixels Number of pixels of each class, np.ndarray, shape (C,). @param[in] no_images Number of images of each class, np.ndarray, shape (C,). @returns nothing. """ no_classes = len(classnames) print('Number of classes:', no_classes) print('Number of annotated images in the dataset:', len(labels)) print('Number of pixels per class:') no_classes = len(classnames) margin = ' ' for c in range(no_classes): print(' ' + str(classnames[c]) + ' [' + str(c) + ']:' \ + margin[:len(margin) - len(classnames[c]) - len(str(c))] \ + str(no_pixels[c])) print('Number of images per class:') for c in range(no_classes): print(' ' + str(classnames[c]) + ' [' + str(c) + ']:' \ + margin[:len(margin) - len(classnames[c]) - len(str(c))] \ + str(no_images[c])) def singular_classes(no_images): """ @brief Get a list of classes that are present in only one image of the dataset. @param[in] no_images Number of images per class, shape (C,). @returns a list of the class indices. """ return [c for c in range(no_images.shape[0]) if no_images[c] < 2] def save_dataset_metadata(images, labels, path): """ @brief Save JSON file with the information of the dataset. @param[in] images List of image paths. @param[in] labels List of label image paths. @param[in] path String with the path of the dataset folder. @returns nothing. """ data = {} # Append class indices idx2class = classnames class2idx = {y: x for x, y in idx2class.items()} data['idx2class'] = idx2class data['class2idx'] = class2idx # Append a list of the images that should be found in the dataset data['images'] = images # Append a list of the labels that should be found in the dataset data['labels'] = labels # Append the number of images data['number_of_images'] = len(images) # Append the list of images per class and the number of pixels per class data['images_per_class'] = {c: [] for c in class2idx} data['number_pixels_per_class'] = {c: 0 for c in class2idx} for image_path, label_path in zip(images, labels): label = read_label(label_path).sum(axis=(1, 2)).astype(np.int64) for class_name in data['images_per_class']: class_id = class2idx[class_name] if label[class_id] > 0: data['images_per_class'][class_name].append(image_path) data['number_pixels_per_class'][class_name] += int(label[class_id]) # Append the number of images per class data['number_images_per_class'] = {c: len(data['images_per_class'][c]) \ for c in class2idx} # Append ignore labels, that is, those classes that are not present # in the dataset data['ignore_labels'] = [c for c in data['number_images_per_class'] \ if data['number_images_per_class'][c] == 0] # Create metadata file with open(path, 'w', encoding='utf-8') as f: json.dump(data, f, ensure_ascii=False, indent=4) def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Create output folder os.mkdir(args.output) # Get list of image and segmentation files images, labels = read_files(args.input) no_pixels, no_images = compute_dataset_stats(images, labels) #print_dataset_stats(no_pixels, no_images, labels) # Get the list of singular classes, i.e. those that are only present in one # image, we will put these images in the training set #sc = [classnames[c] for c in singular_classes(no_images)] #print('Classes that are only in one image:', sc) # TODO: Save dataset metadata dataset_metadata_path = os.path.join(args.output, 'metadata.txt') save_dataset_metadata(images, labels, dataset_metadata_path) # Generate the split for each fold for i in tqdm.tqdm(range(args.folds)): # Get a valid split of the data valid_split = False split_dic = None attempt = 0 while not valid_split: attempt += 1 sys.stdout.write('Attempt to generate a valid split... ') sys.stdout.flush() split_dic = generate_random_split(images, labels, args.train) valid_split = validate_random_split(split_dic) if valid_split: sys.stdout.write("[OK]\n") else: sys.stdout.write("[FAIL]\n") # Create output folders sys.stdout.write('Creating output folders... ') sys.stdout.flush() fold_path = os.path.join(args.output, 'fold_' + str(i)) fold_train_path = os.path.join(fold_path, 'train') fold_test_path = os.path.join(fold_path, 'test') os.mkdir(fold_path) os.mkdir(fold_train_path) os.mkdir(fold_test_path) sys.stdout.write("[OK]\n") # Save fold metadata train_metadata_path = os.path.join(fold_path, 'train_metadata.txt') test_metadata_path = os.path.join(fold_path, 'test_metadata.txt') save_dataset_metadata(split_dic['train']['images'], split_dic['train']['labels'], train_metadata_path) save_dataset_metadata(split_dic['test']['images'], split_dic['test']['labels'], test_metadata_path) # Copy data to the fold sys.stdout.write('Copying data to the fold... ') sys.stdout.flush() copy_data(split_dic, fold_train_path, fold_test_path) sys.stdout.write("[OK]\n") if __name__ == '__main__': main()
16,274
34.927152
82
py
segodsidb
segodsidb-main/src/test.py
""" @brief Main script to launch the testing process. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 5 Jun 2021. """ import argparse import collections import torch import numpy as np import tqdm import json import cv2 import skimage.color # My imports import torchseg.config.parser import torchseg.data_loader as dl import torchseg.model import torchseg.machine # There are 35 classes plus background in ODSI-DB palette = np.round(np.array([ [0.0 , 0.0, 0.0 ], [0.73910129, 0.54796227, 0.70659469], [0.07401779, 0.48485457, 0.2241555 ], [0.35201324, 0.9025658 , 0.81062183], [0.08126211, 0.23986311, 0.54880697], [0.33267484, 0.6119932 , 0.30272535], [0.45419585, 0.1818727 , 0.5877175 ], [0.1239585 , 0.17862775, 0.6892662 ], [0.20556493, 0.44462774, 0.38364081], [0.18754881, 0.1831789 , 0.00863592], [0.37702173, 0.075744 , 0.07170247], [0.9487006 , 0.90159635, 0.26639963], [0.8954375 , 0.58731839, 0.87918311], [0.83980577, 0.77131811, 0.02192928], [0.47681103, 0.72962211, 0.96439166], [0.44293943, 0.60166042, 0.5879358 ], [0.52419707, 0.18690438, 0.69027514], [0.34720014, 0.57450984, 0.96570434], [0.78380941, 0.2237716 , 0.52199938], [0.98170786, 0.61735585, 0.73834123], [0.44000012, 0.06259595, 0.76726459], [0.47754739, 0.13137904, 0.04615173], [0.65486219, 0.24028978, 0.75424866], [0.79301129, 0.75970907, 0.06562084], [0.14864707, 0.55623561, 0.80328385], [0.54439947, 0.234355 , 0.81248573], [0.24443958, 0.00697174, 0.59921356], [0.76808718, 0.56387681, 0.52199431], [0.69855907, 0.73646473, 0.8320837 ], [0.85436454, 0.86456808, 0.61494475], [0.34944949, 0.79188401, 0.8251793 ], [0.43554137, 0.18054355, 0.80210866], [0.76501493, 0.38795293, 0.49637574], [0.31552006, 0.3704537 , 0.90083695], [0.26176471, 0.66781917, 0.65375891], [0.21141543, 0.16505171, 0.53799316], ]) * 255.).astype(np.uint8) def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-c': 'config file path (default: None)', '-l': 'logger config file path (default: None)', '-r': 'path to latest checkpoint (default: None)', '-d': 'indices of GPUs to enable (default: all)', #'-o': 'path to the output JSON file (default: None)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" args = argparse.ArgumentParser(description='PyTorch segmenter.') args.add_argument('-c', '--conf', default=None, type=str, help=help('-c')) args.add_argument('-l', '--logconf', default=None, type=str, help=help('-l')) args.add_argument('-r', '--resume', default=None, type=str, help=help('-r')) args.add_argument('-d', '--device', default=None, type=str, help=help('-d')) #args.add_argument('-o', '--output', default=None, type=str, help=help('-o')) return args def parse_config(args): """ @brief Combines parameters from both JSON and command line into a @param[in] args Argparse args object. @returns A torchseg.config.parser.ConfigParser object. """ # Custom CLI options to modify the values provided in the JSON configuration CustomArgs = collections.namedtuple('CustomArgs', 'flags type target') options = [ CustomArgs(['-o', '--output'], type=str, target='output'), ] config = torchseg.config.parser.ConfigParser.from_args(args, options) return config def image_based_accuracy(log_pred, gt, nclasses=35): # Convert log-probabilities into probabilities pred = torch.exp(log_pred) # Remove the batch dimension pred = pred[0, :] gt = gt[0, :] # Flatten the images pred = torch.reshape(pred, (nclasses, -1)) gt = torch.reshape(gt, (nclasses, -1)) # List of classes considered idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames class2idx = {y: x for x, y in idx2class.items()} relevant_class = ["Skin", "Oral mucosa", "Enamel", "Tongue", "Lip", "Hard palate", "Attached gingiva", "Soft palate", "Hair"] relevant_class_idx = torch.tensor([class2idx[x] for x in relevant_class]) # Get only the annotated pixels ann_idx = torch.sum(gt, dim=0) == 1 y_pred = pred[:, ann_idx] y_true = gt[:, ann_idx] # Discard non-relevant classes y_pred = y_pred[relevant_class_idx, :] y_true = y_true[relevant_class_idx, :] # Get class index predictions y_pred = torch.argmax(y_pred, dim=0) y_true = torch.argmax(y_true, dim=0) # Compute accuracy correct_predictions = (y_pred == y_true).float().sum() total_predictions = y_true.shape[0] accuracy = (correct_predictions / total_predictions).item() return accuracy def patch_based_prediction(data, gt, model, device, patch_size=512): # Create full prediction output = torch.zeros_like(gt).to(device) im_height = data.shape[2] im_width = data.shape[3] i = 0 j = 0 while j < im_height: # Compute patch height j_end = j + patch_size if j_end > im_height: j_end = im_height while i < im_width: # Compute patch width i_end = i + patch_size if i_end > im_width: i_end = im_width # Get patch into GPU patch = data[:, :, j:j_end, i:i_end] patch = patch.to(device) # Perform patch inference output[:, :, j:j_end, i:i_end] = model(patch) # Move to the right i += patch_size # Move to the bottom j += patch_size return output def label2bgr(im, pred, gt): """ @brief Function to convert a segmentation label into an RGB image. Label is expected to be a Torch tensor with shape C, H, W. """ # Convert prediction and ground truth to single channel pred_sinchan = torch.argmax(pred, dim=0).numpy() gt_sinchan = torch.argmax(gt, dim=0).numpy() # Add one to all the classes because the non-annotated pixels will be black pred_sinchan += 1 gt_sinchan += 1 # Black out the non-annotated pixels nan_idx = torch.sum(gt, dim=0) != 1 pred_sinchan[nan_idx, ...] = 0 gt_sinchan[nan_idx, ...] = 0 # List of classes considered idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames class2idx = {y: x for x, y in idx2class.items()} relevant_class = ["Skin", "Oral mucosa", "Enamel", "Tongue", "Lip", "Hard palate", "Attached gingiva", "Soft palate", "Hair"] relevant_class_idx = [class2idx[x] for x in relevant_class] # Black out the non-relevant classes for idx in relevant_class_idx: pixels_of_this_class = gt_sinchan == (idx + 1) pred_sinchan[pixels_of_this_class] = 0 gt_sinchan[pixels_of_this_class] = 0 # Convert single-channel label prediction and ground truth to BGR pred_bgr = skimage.color.label2rgb(pred_sinchan, colors=palette) gt_bgr = skimage.color.label2rgb(gt_sinchan, colors=palette) return pred_bgr, gt_bgr def main(): # Read command line arguments args = parse_cmdline_params() # Read configuration file config = parse_config(args) logger = config.get_logger('test') # Hack config file to add 'odsi_db_conf_mat' as a metric config['metrics'].append('odsi_db_conf_mat') # Get testing logger logger = config.get_logger('test') # Build network architecture model = config.init_obj('model', torchseg.model) logger.info(model) # Load the weights of a trained architecture (typically model_best.pth) logger.info('Loading checkpoint: {} ...'.format(config.resume)) device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') checkpoint = torch.load(config.resume, map_location=device) state_dict = checkpoint['state_dict'] if torch.cuda.is_available() and config['n_gpu'] > 1: model = torch.nn.DataParallel(model) model.load_state_dict(state_dict) model = model.to(device) # Put model in test mode model.eval() # Get function handles of loss and metrics loss_fn = getattr(torchseg.model.loss, config['loss']) metric_fns = [getattr(torchseg.model.metric, met) \ for met in config['metrics']] # Loop over all testing datasets for dataset in config['testing']['datasets']: # Create dataloader for this particular testing dataset data_loader = getattr(torchseg.data_loader, dataset['type'])(**dataset['args']) data_loader.training = False # Initialise image-based accuracy accumulator im_data = [] im_pred = [] im_gt = [] im_acc = [] im_path = [] # Run inference on all the samples in the dataset total_loss = 0.0 #total_metrics = torch.zeros(len(metric_fns)) total_metrics = [None] * len(metric_fns) counter = 0 with torch.no_grad(): for i, raw_data in enumerate(tqdm.tqdm(data_loader)): counter += 1 # Load input data data = raw_data['image'] gt = raw_data['label'] path = raw_data['path'] # Run inference on the GPU output = None if data.shape[3] > 512 and data_loader.mode in ['rgbpixel_test', 'spixel_170_test']: output = patch_based_prediction(data, gt, model, device) gt = gt.to(device) else: # Send image and label to GPU data, gt = data.to(device), gt.to(device) # Perform inference on the input output = model(data) # Update image-based accuracy im_data.append(data.detach().cpu()[0, ...]) im_pred.append(output.detach().cpu()[0, ...]) im_gt.append(gt.detach().cpu()[0, ...]) im_acc.append(image_based_accuracy(output.detach().cpu(), gt.detach().cpu())) im_path.append(path) # Update accumulated loss loss = loss_fn(output, gt) batch_size = data.shape[0] total_loss += loss.item() * batch_size # Update accumulated metrics for i, metric in enumerate(metric_fns): metric_value = metric(output, gt) if total_metrics[i] is None: total_metrics[i] = metric_value * batch_size else: total_metrics[i] += metric_value * batch_size # Get best, median, worst image index best_idx = np.argmax(im_acc) median_idx = np.argsort(im_acc)[len(im_acc) // 2] worst_idx = np.argmin(im_acc) # FIXME: debugging pred_bgr, gt_bgr = label2bgr(im_data[best_idx], im_pred[best_idx], im_gt[best_idx]) cv2.imwrite('/tmp/pred_bgr.png', pred_bgr) cv2.imwrite('/tmp/gt_bgr.png', gt_bgr) # Calculate the final image-based accuracy print('Image-based minimum accuracy:', im_acc[worst_idx]) print('Image-based median accuracy:', im_acc[median_idx]) print('Image-based maximum accuracy:', im_acc[best_idx]) print('Image-based average accuracy:', sum(im_acc) / len(im_acc)) #print('Batch size:', batch_size) #print('Metric functions:', metric_fns) #print('Total metrics:', [x / counter for x in total_metrics]) # Find the index of the 'confusion matrix' metric conf_mat_idx = None for i, met in enumerate(metric_fns): if met.__name__ == 'odsi_db_conf_mat': conf_mat_idx = i if not conf_mat_idx: raise RuntimeError('Confusion matrix fake metric not found.') # Get confusion matrix (accumulated over all the testing pixels) conf_mat = total_metrics[conf_mat_idx][0].detach().cpu().numpy() # Compute the balance accuracy per class bal_acc = {} sensitivity = {} specificity = {} accuracy = {} i2c = torchseg.data_loader.OdsiDbDataLoader.OdsiDbDataset.classnames for i in range(conf_mat.shape[0]): tp, fp, tn, fn = conf_mat[i] sensitivity[i2c[i]] = 1. * tp / (tp + fn) specificity[i2c[i]] = 1. * tn / (tn + fp) accuracy[i2c[i]] = 1. * (tp + tn) / (tp + tn + fp + fn) bal_acc[i2c[i]] = .5 * sensitivity[i2c[i]] + .5 * specificity[i2c[i]] # Save JSON file with the results per class with open(config['output'] + '_sensitivity.json', 'w') as f: json.dump(sensitivity, f) with open(config['output'] + '_specificity.json', 'w') as f: json.dump(specificity, f) with open(config['output'] + '_accuracy.json', 'w') as f: json.dump(accuracy, f) with open(config['output'] + '_balanced_accuracy.json', 'w') as f: json.dump(bal_acc, f) # Create a dictionary with the resuls on this dataset #n_samples = len(data_loader.sampler) #log = {'loss': total_loss / n_samples} #log.update({ # met.__name__: total_metrics[i].item() / n_samples \ # for i, met in enumerate(metric_fns) #}) # Show or save the results for this dataset in the log #logger.info({dataset['type']: log}) if __name__ == '__main__': main()
13,865
34.92228
100
py
segodsidb
segodsidb-main/src/generate_rgb_recon.py
""" @brief Script that reconstructs the RGB images of the ODSI-DB dataset and saves them in output folders, one per camera. - Nuance EX (1392x1040 pixels, 450-950nm, 10nm steps). - Specim IQ (512x512 pixels, 400-1000nm, 3nm steps). @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 16 Feb 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-o': 'Path to the output directory (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-o', '--output', required=True, type=str, help=help('-o')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): # Input directory must exist if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') # Output directory should not exist, we will create it if os.path.exists(args.output): raise RuntimeError('[ERROR] Output directory already exists.') return args def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Create output folders nuance_ex_path = os.path.join(args.output, 'nuance_ex') specim_iq_path = os.path.join(args.output, 'specim_iq') os.mkdir(args.output) os.mkdir(nuance_ex_path) os.mkdir(specim_iq_path) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) for im_path, label_path in tqdm.tqdm(zip(image_paths, label_paths)): im_fname = os.path.splitext(ntpath.basename(im_path))[0] label_fname = ntpath.basename(label_path) # Read image im_hyper, wl, im_rgb_orig, metadata = dl.read_stiff(im_path, silent=True, rgb_only=False) nbands = im_hyper.shape[2] # Copy image to the right folder depending on the type of camera, this is to observe # the diferences between the images reconstructed from each of the different cameras if nbands == 51: # Nuance EX im_dpath = os.path.join(nuance_ex_path, im_fname + '.jpg') im_dpath_orig = os.path.join(nuance_ex_path, im_fname + '_orig.jpg') #label_dpath = os.path.join(nuance_ex_path, label_fname) elif nbands == 204: # Specim IQ im_dpath = os.path.join(specim_iq_path, im_fname) + '.jpg' im_dpath_orig = os.path.join(specim_iq_path, im_fname + '_orig.jpg') #label_dpath = os.path.join(specim_iq_path, label_fname) else: raise ValueError('[ERROR] The image {} has {} bands.'.format(im_fname, nbands)) # Convert hyperspectral image to RGB im_rgb = dl.OdsiDbDataLoader.LoadImage.hyper2rgb(im_hyper, wl) # Save reconstructed RGB image into the output folder im_bgr = im_rgb[...,::-1].copy() cv2.imwrite(im_dpath, im_bgr, [int(cv2.IMWRITE_JPEG_QUALITY), 100]) # Save RGB provided in ODSI-DB into the output folder im_bgr_orig = im_rgb_orig[...,::-1].copy() cv2.imwrite(im_dpath_orig, im_bgr_orig, [int(cv2.IMWRITE_JPEG_QUALITY), 100]) if __name__ == '__main__': main()
4,659
32.52518
93
py
segodsidb
segodsidb-main/src/compute_specim_iq_mean.py
""" @brief Compute the channel mean of the Specim IQ images. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 16 Feb 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # RGB average nbands = 204 hyper_sums = np.zeros((nbands,), dtype=np.float64) pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, im_rgb_orig, metadata = dl.read_stiff(im_path, silent=True, rgb_only=False) # We are only interested in Specim IQ images if im_hyper.shape[2] != nbands: continue # Update average computation for i in range(nbands): hyper_sums[i] += im_hyper[:, :, i].sum() pixel_count += im_hyper.shape[0] * im_hyper.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('Specim IQ average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('Specim IQ average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
3,733
26.659259
80
py
segodsidb
segodsidb-main/src/validate_dataset.py
""" @brief Script to validate the hyperspectral data of ODSI-DB along with the generation of RGB images from hyperspectral data. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 27 Sep 2021. """ import argparse import numpy as np import os import tqdm import random import pandas as pd import seaborn as sns import matplotlib.pyplot as plt import sklearn.manifold import multiprocessing import colour import sys import cv2 import scipy # My imports import torchseg.data_loader import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = torchseg.data_loader.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-o': 'Path to the output directory (required: True)', '-n': 'Max. no. of randomly selected pixels for t-SNE (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-o', '--output', required=True, type=str, help=help('-o')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): # Input directory must exist if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') # Output directory should not exist, we will create it if os.path.exists(args.output): raise RuntimeError('[ERROR] Output directory already exists.') return args def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Create output folder os.mkdir(args.output) # Get list of image and segmentation files sys.stdout.write('[INFO] Reading list of input files... ') sys.stdout.flush() image_paths, label_paths = read_files(args.input) print('OK') # Loop over the images for im_path, label_path in zip(image_paths, label_paths): # Read image im_hyper, wl, im_rgb, metadata = torchseg.data_loader.read_stiff(im_path, silent=True, rgb_only=False) # Convert hyperspectral image to RGB fname = os.path.basename(os.path.splitext(im_path)[0]) sys.stdout.write("[INFO] Converting {} to RGB... ".format(fname)) image_loader = torchseg.data_loader.OdsiDbDataLoader.LoadImage im_recon_rgb = image_loader.hyper2rgb(im_hyper, wl) sys.stdout.flush() print('OK') # Convert RGB image to BGR so that we can save it with OpenCV im_recon_bgr = im_recon_rgb[...,::-1] # Save reconstructed RGB image to the output folder im_recon_path = os.path.join(args.output, fname + '_reconstructed.png') cv2.imwrite(im_recon_path, im_recon_bgr) # Save RGB image provided by the dataset to the output folder im_orig_bgr = im_rgb[...,::-1] im_orig_path = os.path.join(args.output, fname + '_original.png') cv2.imwrite(im_orig_path, im_orig_bgr) if __name__ == '__main__': main()
4,126
29.57037
82
py
segodsidb
segodsidb-main/src/compute_all_hyper_450_950_51_mean.py
""" @brief Compute the channel mean of the 51 interpolated bands from 450-950nm. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 21 Jun 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # Hyperspectral mean computation stuff nbands = 51 new_wl = np.linspace(450, 950, nbands) hyper_sums = np.zeros((nbands,), dtype=np.float64) pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, _, _ = dl.read_stiff(im_path, silent=True, rgb_only=False) # Interpolate hyperspectral image to the 450-950nm range (51 bands) interp_func = dl.OdsiDbDataLoader.LoadImage.interp_spectra im_hyper = interp_func(im_hyper, wl, new_wl) # Update average computation for i in range(nbands): hyper_sums[i] += im_hyper[:, :, i].sum() pixel_count += im_hyper.shape[0] * im_hyper.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('All hyper 450-950nm (51 bands) average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('All hyper 450-950nm (51 bands) average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
3,960
27.912409
80
py
segodsidb
segodsidb-main/src/compute_all_hyper_450_950_170_mean.py
""" @brief Compute the channel mean of the 170 interpolated bands from 450-950nm. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 21 Jun 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # Hyperspectral mean computation stuff nbands = 170 new_wl = np.linspace(450, 950, nbands) hyper_sums = np.zeros((nbands,), dtype=np.float64) pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, _, _ = dl.read_stiff(im_path, silent=True, rgb_only=False) # Interpolate hyperspectral image to the 450-950nm range (170 bands) interp_func = dl.OdsiDbDataLoader.LoadImage.interp_spectra im_hyper = interp_func(im_hyper, wl, new_wl) # Update average computation for i in range(nbands): hyper_sums[i] += im_hyper[:, :, i].sum() pixel_count += im_hyper.shape[0] * im_hyper.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('All hyper 450-950nm (170 bands) average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('All hyper 450-950nm (170 bands) average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
3,964
27.941606
80
py
segodsidb
segodsidb-main/src/generate_tsne.py
""" @brief Script that generates a t-SNE of all the pixels in the ODSI-DB dataset, comparing RGB pixels with hyperspectral pixels. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 27 Sep 2021. """ import argparse import numpy as np import os import tqdm import random import pandas as pd import seaborn as sns import matplotlib import matplotlib.pyplot as plt import sklearn.manifold import multiprocessing import scipy.spatial #import colour # My imports import torchseg.data_loader import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = torchseg.data_loader.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-o': 'Path to the output directory (required: True)', '-n': 'Max. no. of selected pixels for t-SNE (required: True)', '-r': 'Use reconstructed RGB images (required: True)', '-e': 'Convert Specim IQ images to Nuance EX style of 51 bands (required: True)', '-v': 'Use only the visible spectrum (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-o', '--output', required=True, type=str, help=help('-o')) parser.add_argument('-n', '--npixels', required=True, type=int, help=help('-n')) parser.add_argument('-r', '--recon', required=True, type=bool, help=help('-r')) parser.add_argument('-v', '--visible', required=True, type=bool, help=help('-v')) parser.add_argument('-e', '--nuance', required=True, type=str, help=help('-e')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): # Input directory must exist if not os.path.isdir(args.input): raise ValueError('[ERROR] Input directory does not exist.') # Output directory should not exist, we will create it if os.path.exists(args.output): raise ValueError('[ERROR] Output directory already exists.') if args.nuance not in ['none', 'nearest', 'linear']: raise ValueError('[ERROR] Specim -> Nuance conversion mode not recognised.') return args def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def specim2nuance(im_specim, specim_wl=None, nuance_wl=None, mode='linear'): """ @brief Convert a hyperspectral image from the Specim IQ camera (204 bands) to the Nuance EX (51 bands). @param[in] im_specim Numpy ndarray, shape (h, w, c). @param[in] specim_wl Array of wavelengths of the Specim IQ bands. @param[in] nuance_wl Array of wavelengths of the Nuance EX bands. @param[in] mode Two modes are available: 'nearest' and 'linear'. nearest: the band with nearest wavelength of the Specim IQ is used. linear: linear interpolation of the two surrounding wavelengths. """ # Get an array of the wavelengths of each camera if specim_wl is None: specim_wl = np.array([ 397.32000732, 400.20001221, 403.08999634, 405.97000122, 408.8500061 , 411.73999023, 414.63000488, 417.51998901, 420.3999939 , 423.29000854, 426.19000244, 429.07998657, 431.97000122, 434.86999512, 437.76000977, 440.66000366, 443.55999756, 446.45001221, 449.3500061 , 452.25 , 455.16000366, 458.05999756, 460.95999146, 463.86999512, 466.76998901, 469.67999268, 472.58999634, 475.5 , 478.41000366, 481.32000732, 484.23001099, 487.14001465, 490.05999756, 492.97000122, 495.89001465, 498.79998779, 501.72000122, 504.64001465, 507.55999756, 510.48001099, 513.40002441, 516.33001709, 519.25 , 522.17999268, 525.09997559, 528.0300293 , 530.96002197, 533.89001465, 536.82000732, 539.75 , 542.67999268, 545.61999512, 548.54998779, 551.48999023, 554.42999268, 557.35998535, 560.29998779, 563.23999023, 566.17999268, 569.11999512, 572.07000732, 575.01000977, 577.96002197, 580.90002441, 583.84997559, 586.79998779, 589.75 , 592.70001221, 595.65002441, 598.59997559, 601.54998779, 604.51000977, 607.46002197, 610.41998291, 613.38000488, 616.34002686, 619.29998779, 622.26000977, 625.2199707 , 628.17999268, 631.15002441, 634.10998535, 637.08001709, 640.03997803, 643.01000977, 645.97998047, 648.95001221, 651.91998291, 654.89001465, 657.86999512, 660.84002686, 663.80999756, 666.78997803, 669.77001953, 672.75 , 675.72998047, 678.71002197, 681.69000244, 684.66998291, 687.65002441, 690.64001465, 693.61999512, 696.60998535, 699.59997559, 702.58001709, 705.57000732, 708.57000732, 711.55999756, 714.54998779, 717.53997803, 720.53997803, 723.5300293 , 726.5300293 , 729.5300293 , 732.5300293 , 735.5300293 , 738.5300293 , 741.5300293 , 744.5300293 , 747.53997803, 750.53997803, 753.54998779, 756.55999756, 759.55999756, 762.57000732, 765.58001709, 768.59997559, 771.60998535, 774.61999512, 777.64001465, 780.65002441, 783.66998291, 786.67999268, 789.70001221, 792.7199707 , 795.73999023, 798.77001953, 801.78997803, 804.80999756, 807.84002686, 810.85998535, 813.89001465, 816.91998291, 819.95001221, 822.97998047, 826.01000977, 829.03997803, 832.07000732, 835.10998535, 838.14001465, 841.17999268, 844.2199707 , 847.25 , 850.28997803, 853.33001709, 856.36999512, 859.41998291, 862.46002197, 865.5 , 868.54998779, 871.59997559, 874.64001465, 877.69000244, 880.73999023, 883.78997803, 886.84002686, 889.90002441, 892.95001221, 896.01000977, 899.05999756, 902.11999512, 905.17999268, 908.23999023, 911.29998779, 914.35998535, 917.41998291, 920.47998047, 923.54998779, 926.60998535, 929.67999268, 932.73999023, 935.80999756, 938.88000488, 941.95001221, 945.02001953, 948.09997559, 951.16998291, 954.23999023, 957.32000732, 960.40002441, 963.4699707 , 966.54998779, 969.63000488, 972.71002197, 975.78997803, 978.88000488, 981.96002197, 985.04998779, 988.13000488, 991.2199707 , 994.30999756, 997.40002441, 1000.48999023, 1003.58001709, ]) if nuance_wl is None: nuance_wl = np.linspace(450, 950, 51) # If we get a Nuance EX image, we don't need to convert anything if im_specim.shape[2] == 51: return im_specim, nuance_wl assert(im_specim.shape[2] == 204) # Compute the distance between wavelengths of the two cameras dist = scipy.spatial.distance.cdist(specim_wl.reshape((204, 1)), nuance_wl.reshape((51, 1)), metric='euclidean') # Compute the nearest nuance wavelength for each specim band nearest = np.argmin(dist, axis=0) # Synthesize Nuance-like image using the nearest bands h, w, _ = im_specim.shape im_nuance = np.empty((h, w, nuance_wl.shape[0]), dtype=np.float32) if mode == 'nearest': for i in range(nuance_wl.shape[0]): im_nuance[:, :, i] = im_specim[:, :, nearest[i]].copy() elif mode == 'linear': for i in range(nuance_wl.shape[0]): # Find the bands before and after if specim_wl[nearest[i]] > nuance_wl[i]: j = nearest[i] - 1 k = nearest[i] else: j = nearest[i] k = nearest[i] + 1 # Compute the weights of the bands before and after d1 = nuance_wl[i] - specim_wl[j] d2 = specim_wl[k] - nuance_wl[i] norm = d1 + d2 w1 = d2 / norm w2 = d1 / norm # Add interpolated channel to the synthetic Nuance EX image im_nuance[:, :, i] = w1 * im_specim[:, :, j] + w2 * im_specim[:, :, k] else: raise ValueError('[ERROR] Unknown interpolation mode.') assert(im_nuance.shape == (h, w, 51)) return im_nuance, nuance_wl def process_image(im_path: str, label_path: str, no_pixels: int, rgb_recon: bool, visible: bool, nuance_recon: str): """ @brief Get the annotated pixels from an image of the ODSI-DB dataset. @param[in] im_path Path to the image file. @param[in] label_path Path to the annotation file. @param[in] no_pixels Number of annotated pixels to be sampled from the image. @param[in] rgb_recon Reconstruct RGB image from hyperspectral. @param[in] visible Only use visible wavelengths. @param[in] nuance_recon Convert Specim IQ images to the Nuance EX style of 51 bands. """ # Read raw hyperspectral image im_hyper, wl, im_rgb, metadata = torchseg.data_loader.read_stiff(im_path, silent=True, rgb_only=False) # Convert image to Nuance EX (in case it is Specim IQ) if nuance_recon is not None: im_hyper, wl = specim2nuance(im_hyper, mode=nuance_recon) # Remove the non-visible part of the spectrum if requested if visible: im_hyper, wl = torchseg.data_loader.OdsiDbDataLoader.LoadImage.filter_bands(im_hyper, wl, 380., 740.) # Reconstruct RGB image if requested if rgb_recon: im_rgb = torchseg.data_loader.OdsiDbDataLoader.LoadImage.hyper2rgb(im_hyper, wl) # Read label label = torchseg.data_loader.OdsiDbDataLoader.LoadImage.read_label( label_path).transpose((1, 2, 0)) # Make sure that image and label have the same height and width assert(label.shape[2] == 35) # Labels are supposed to have 35 classes assert(im_hyper.shape[0] == label.shape[0]) assert(im_hyper.shape[1] == label.shape[1]) assert(im_rgb.shape[0] == label.shape[0]) assert(im_rgb.shape[1] == label.shape[1]) # Collect the labelled pixels ann = np.where(np.sum(label, axis=2) == 1) ann = list(zip(ann[0], ann[1])) # Downsample the number of pixels random.shuffle(ann) ann = ann[:no_pixels] # Save the randomly selected pixels hyper_pixels = [] rgb_pixels = [] labels = [] for i, j in ann: hyper_pixels.append(im_hyper[i, j, :].tolist()) rgb_pixels.append(im_rgb[i, j, :].tolist()) labels.append(np.argmax(label[i, j, :])) return hyper_pixels, rgb_pixels, labels def get_pixels(image_paths, label_paths, no_pixels=1000000, rgb_recon=False, visible=False, nuance_recon=False): """ @brief Get all the labelled pixels of the ODSI-DB dataset. @param[in] image_paths List of image paths. @param[in] label_paths List of label paths in sync with the list of images. @param[in] rgb_recon True if you want to reconstruct the RGB images from the hyperspectral images and not use the ones already provided in ODSI-DB. @param[in] visible True if you want to use just the visible part of the spectrum. @param[in] nuance_recon Convert Specim IQ image to Nuance EX style. """ assert(len(image_paths) == len(label_paths)) nimages = len(image_paths) ppi = no_pixels // nimages retval = None with multiprocessing.Pool() as pool: data_input = list(zip(image_paths, label_paths, [ppi] * nimages, [rgb_recon] * nimages, [visible] * nimages, [nuance_recon] * nimages)) retval = pool.starmap(process_image, data_input)[0] #hyper_pixels, rgb_pixels = retval[0] #labels = retval[1] return retval def plot(rgb_embed, hyper_embed, labels): """ @brief Plots the t-SNE of RGB and hyperspectral pixels. @param[in] rgb_embed Embedding of RGB pixels (N, 2). @param[in] hyper_embed Embedding of hyperspectral pixels (N, 2). @param[in] labels List of strings (N,). @returns a tuple (fig, axes, lgd) where fig and ax are the usual matplotlib objects and lgd is the fig.legend. """ # Prepare data to plot with seaborn rgb_data = {'y': rgb_embed[:, 0], 'x': rgb_embed[:, 1], 'label': labels} hyper_data = {'y': hyper_embed[:, 0], 'x': hyper_embed[:, 1], 'label': labels} # Create figure sns.set() fig, axes = plt.subplots(1, 2, figsize=(16, 8)) # Generate subplots palette = dict(zip(idx2class.values(), sns.color_palette('husl', n_colors=len(idx2class.values())))) # sns.color_palette(n_colors=len(idx2class.values())))) rgb_plot = sns.scatterplot(ax=axes[0], data=rgb_data, x='x', y='y', hue='label', s=5) hyper_plot = sns.scatterplot(ax=axes[1], data=hyper_data, x='x', y='y', hue='label', s=5) # Make plots beautiful ymax = max(rgb_embed[:, 0].max(), hyper_embed[:, 0].max()) ymin = min(rgb_embed[:, 0].min(), hyper_embed[:, 0].min()) xmax = max(rgb_embed[:, 1].max(), hyper_embed[:, 1].max()) xmin = min(rgb_embed[:, 1].min(), hyper_embed[:, 1].min()) handles, figlabels = fig.axes[-1].get_legend_handles_labels() lgd = fig.legend(handles, figlabels, loc='lower center', bbox_to_anchor=(0.5, -0.15), ncol=10) #lgd = fig.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.) axes[0].set_title('RGB', pad=20) #rgb_plot.set(title='RGB') rgb_plot.set(xlabel=None) rgb_plot.set(ylabel=None) rgb_plot.set(ylim=[ymin, ymax]) rgb_plot.set(xlim=[xmin, xmax]) rgb_plot.set(xticklabels=[]) rgb_plot.set(yticklabels=[]) rgb_plot.set(facecolor='white') rgb_plot.get_legend().remove() axes[1].set_title('Hyperspectral', pad=20) #hyper_plot.set(title='Hyperspectral') hyper_plot.set(xlabel=None) hyper_plot.set(ylabel=None) hyper_plot.set(ylim=[ymin, ymax]) hyper_plot.set(xlim=[xmin, xmax]) hyper_plot.set(xticklabels=[]) hyper_plot.set(yticklabels=[]) hyper_plot.set(facecolor='white') hyper_plot.get_legend().remove() return fig, axes, lgd def savefig(fig: matplotlib.figure.Figure, lgd: matplotlib.legend.Legend, path: str): """ @brief Save figure to file. @param[in] fig Seaborn figure. @param[in] lgd Seaborn legend. This is needed because we want to place the legend outside of the plotting area. To do this, we need to pass the legend to the 'bbox_extra_artists' argument of fig.savefig(). @param[in] path Path to the output file where the figure will be saved. """ plt.tight_layout() fig.savefig(path, bbox_extra_artists=(lgd,), bbox_inches='tight') def specimen_iq_visble_rgb_metric(pixa, pixb): wl = np.array([ 397.32000732, 400.20001221, 403.08999634, 405.97000122, 408.8500061, 411.73999023, 414.63000488, 417.51998901, 420.3999939 , 423.29000854, 426.19000244, 429.07998657, 431.97000122, 434.86999512, 437.76000977, 440.66000366, 443.55999756, 446.45001221, 449.3500061, 452.25, 455.16000366, 458.05999756, 460.95999146, 463.86999512, 466.76998901, 469.67999268, 472.58999634, 475.5, 478.41000366, 481.32000732, 484.23001099, 487.14001465, 490.05999756, 492.97000122, 495.89001465, 498.79998779, 501.72000122, 504.64001465, 507.55999756, 510.48001099, 513.40002441, 516.33001709, 519.25, 522.17999268, 525.09997559, 528.0300293, 530.96002197, 533.89001465, 536.82000732, 539.75, 542.67999268, 545.61999512, 548.54998779, 551.48999023, 554.42999268, 557.35998535, 560.29998779, 563.23999023, 566.17999268, 569.11999512, 572.07000732, 575.01000977, 577.96002197, 580.90002441, 583.84997559, 586.79998779, 589.75, 592.70001221, 595.65002441, 598.59997559, 601.54998779, 604.51000977, 607.46002197, 610.41998291, 613.38000488, 616.34002686, 619.29998779, 622.26000977, 625.2199707 , 628.17999268, 631.15002441, 634.10998535, 637.08001709, 640.03997803, 643.01000977, 645.97998047, 648.95001221, 651.91998291, 654.89001465, 657.86999512, 660.84002686, 663.80999756, 666.78997803, 669.77001953, 672.75, 675.72998047, 678.71002197, 681.69000244, 684.66998291, 687.65002441, 690.64001465, 693.61999512, 696.60998535, 699.59997559, 702.58001709, 705.57000732, 708.57000732, 711.55999756, 714.54998779, 717.53997803, 720.53997803, 723.5300293, 726.5300293, 729.5300293, 732.5300293, 735.5300293, 738.5300293, ]) # Create image with the two pixels im_hyper = np.dstack((pixa, pixb)).transpose((0, 2, 1)) # Convert image to RGB image_loader = torchseg.data_loader.OdsiDbDataLoader.LoadImage im = image_loader.hyper2rgb(im_hyper, wl) # Compute Euclidean distance in RGB dist = np.linalg.norm(im[0, 0] - im[0, 1]) return dist def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Create output folder os.mkdir(args.output) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # FIXME: this is for debugging #image_paths = image_paths[:1] #label_paths = label_paths[:1] # Read pixels from the dataset args.nuance = None if args.nuance == 'none' else args.nuance hyper_pixels, rgb_pixels, labels = get_pixels(image_paths, label_paths, args.npixels, rgb_recon=args.recon, visible=args.visible, nuance_recon=args.nuance) #rgb_pixels = np.array(rgb_pixels) #hyper_pixels = np.array(hyper_pixels) #assert(rgb_pixels.shape[0] == hyper_pixels.shape[0]) assert(len(hyper_pixels) == len(rgb_pixels)) assert(len(rgb_pixels) == len(labels)) # Convert labels from class indices to class names labels = [idx2class[l] for l in labels] # t-SNE for perp in list(range(10, 210, 10)): print("Running t-SNE with perplexity = {}...".format(perp)) rgb_tsne = sklearn.manifold.TSNE(n_components=2, verbose=0, perplexity=perp, n_iter=1000, learning_rate=200, init='pca') hyper_tsne = sklearn.manifold.TSNE(n_components=2, verbose=0, perplexity=perp, n_iter=1000, learning_rate=200, init='pca') rgb_embed = rgb_tsne.fit_transform(rgb_pixels) hyper_embed = hyper_tsne.fit_transform(hyper_pixels) # Save embeddings with open(os.path.join(args.output, "rgb_embed_perplexity_{}.npy".format(perp)), 'wb') as f: np.save(f, rgb_embed) with open(os.path.join(args.output, "hyper_embed_perplexity_{}.npy".format(perp)), 'wb') as f: np.save(f, hyper_embed) # Plot RGB and hyperspectral t-SNE subplots fig, axes, lgd = plot(rgb_embed, hyper_embed, labels) # Save figure inside the output folder savefig(fig, lgd, os.path.join(args.output, "figure_perplexity_{}.png".format(perp))) if __name__ == '__main__': main()
20,785
42.668067
102
py
segodsidb
segodsidb-main/src/generate_results_latex.py
""" @brief Script that generates a Latex table where every row represents a class and every column represents a cross-validation fold. The cells will contain the metric, which for ODSI-DB we have decided will be balanced accuracy. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 5 Aug 2021. """ import argparse import numpy as np import os import json import numpy as np # My imports import torchseg.data_loader as dl def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Dictionary of fold names to paths.', '-c': 'Caption of the table of results.', '-l': 'Table label.', '-s': 'List containing a subset of classes to display.', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-c', '--caption', required=True, type=str, help=help('-c')) parser.add_argument('-l', '--label', required=True, type=str, help=help('-l')) parser.add_argument('-s', '--classes', required=False, default='None', type=str, help=help('-l')) # Read parameters args = parser.parse_args() args.input = eval(args.input) args.classes = eval(args.classes) assert(type(args.input) == dict) return args def validate_cmdline_params(args): # Check that the JSON files exist #for fid in args.input: # if not os.path.isfile(args.input[fid]): # raise ValueError('[ERROR] The path `' + str(args.input[fid]) \ # + '` does not exist.') return args def read_json(path): data = None with open(path) as f: data = json.load(f) return data def print_latex_table(metrics, caption, label, class_subset=None): """ @brief Print a table with classes as rows and folds as columns. @param[in] metric_per_class Dictionary of class names -> metric value. @param[in] metric_name Name of the metric. @returns Nothing. """ # Get the list of classes that the user wants to consider list_of_classes = None if class_subset: list_of_classes = class_subset else: list_of_classes = list(metrics[metrics.keys()[0]].keys()) #for c in list_of_classes: # if class_subset is not None and c not in class_subset: # del metric_per_class[c] # Average over classes #metric_avg = np.nanmean([v for k, v in metric_per_class.items()]) # Convert metric to percentage #metric_per_class = {k: 100. * v for k, v in metric_per_class.items()} #metric_avg *= 100. # Compute average over classes average = {} for m in metrics: acc = 0. for c in metrics[m]: if c not in list_of_classes: continue acc += metrics[m][c] avg = acc / len(list_of_classes) average[m] = avg # Table header print() print("\\begin{table*}[!htb]") print(" \\centering") print(" \\caption{" + caption + "}") print(" \\vspace{0.2cm}") print(" \\begin{tabular}{l" + "c" * len(metrics) + "}") print(" \\hline") print(" \\multicolumn{1}{c}{\\bfseries Class} ") # Print column names for m in metrics: print(" & \\multicolumn{1}{c}{\\bfseries " \ + m.capitalize().replace('_', ' ') + "}") # + " (\\%)}") print(" \\\\") print(" \\hline") # Print all the metrics for each class for c in sorted(list_of_classes): line_str = ' ' + c for m in metrics: line_str += " & {:.2f}".format(metrics[m][c] * 100) line_str += " \\\\" print(line_str) print(" \\hline") # Print the average for each metric line_str = " \\multicolumn{1}{c}{\\bfseries Average} " for m in metrics: line_str += " & {:.2f}".format(average[m] * 100) line_str += " \\\\" print(line_str) # Table footer print(" \\end{tabular}") print(" \\vspace{0.2cm}") print(" \\label{" + label + "}") print("\\end{table*}") def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) classnames = dl.OdsiDbDataLoader.OdsiDbDataset.classnames # List of suffixes lsuffix = ['_sensitivity.json', '_specificity.json', '_accuracy.json', '_balanced_accuracy.json'] # Produce latex table for all the metrics metrics = {} for suffix in lsuffix: # Read JSON files with the resuls for each class fold_results = {fold_id: read_json(path + suffix) for fold_id, path in args.input.items()} # Average over folds metric_per_class = {classnames[k]: None for k in classnames} for class_name in metric_per_class: values = [fold_results[fold_id][class_name] for fold_id in fold_results] metric_per_class[class_name] = np.nanmean(values) # Store the per-class results for this metric metrics[suffix.split('.')[0][1:]] = metric_per_class # Print Latex table formated for Photonics West print_latex_table(metrics, caption=args.caption, label=args.label, class_subset=args.classes) #metric_name="Balanced accuracy (\\%)", if __name__ == '__main__': main()
5,686
30.41989
98
py
segodsidb
segodsidb-main/src/split_images_per_camera_model.py
""" @brief Script that splits ODSI-DB dataset into two parts: images generated by the Nuance EX (1392x1040 pixels, 450-950nm, 10nm steps) and the Specim IQ (512x512 pixels, 400-1000nm, 3nm steps). @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 27 Sep 2021. """ import argparse import numpy as np import os import ntpath import shutil import tqdm # My imports import torchseg.data_loader # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = torchseg.data_loader.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-o': 'Path to the output directory (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-o', '--output', required=True, type=str, help=help('-o')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): # Input directory must exist if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') # Output directory should not exist, we will create it if os.path.exists(args.output): raise RuntimeError('[ERROR] Output directory already exists.') return args def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Create output folders nuance_ex_path = os.path.join(args.output, 'nuance_ex') specim_iq_path = os.path.join(args.output, 'specim_iq') os.mkdir(args.output) os.mkdir(nuance_ex_path) os.mkdir(specim_iq_path) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) for im_path, label_path in tqdm.tqdm(zip(image_paths, label_paths)): im_fname = ntpath.basename(im_path) label_fname = ntpath.basename(label_path) # Read image im_hyper, wl, im_rgb, metadata = torchseg.data_loader.read_stiff(im_path, silent=True, rgb_only=False) nbands = im_hyper.shape[2] # Copy image to the right folder depending on the type of camera if nbands == 51: # Nuance EX im_dpath = os.path.join(nuance_ex_path, im_fname) label_dpath = os.path.join(nuance_ex_path, label_fname) elif nbands == 204: # Specim IQ im_dpath = os.path.join(specim_iq_path, im_fname) label_dpath = os.path.join(specim_iq_path, label_fname) else: raise ValueError('[ERROR] The image {} has {} bands.'.format(im_fname, nbands)) shutil.copyfile(im_path, im_dpath) shutil.copyfile(label_path, label_dpath) if __name__ == '__main__': main()
3,860
30.137097
82
py
segodsidb
segodsidb-main/src/compute_all_hyper_400_1000_204_std.py
""" @brief Compute the channel std of the 204 interpolated bands from 400-1000nm. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 8 Jul 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, # e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-m': 'Channel means normalised to the range [0, 1] (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-m', '--means', required=True, type=str, help=help('-m')) # Read parameters args = parser.parse_args() args.means = eval(args.means) return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, _ = read_files(args.input) # Read means mean_sq = np.array(args.means) * np.array(args.means) # Hyperspectral std computation stuff min_wl = 400 max_wl = 1000 nbands = 204 new_wl = np.linspace(min_wl, max_wl, nbands) hyper_sums = np.zeros((new_wl.shape[0],), dtype=np.float64) pixel_count = 0 image_count = 0 interp_func = dl.OdsiDbDataLoader.LoadImage.interp_spectra # Get the indices of the wavelengths that are shared among both cameras inside_indices = np.array([idx for idx, x in enumerate(new_wl.tolist()) \ if x > 450. and x < 950.]) inside_wl = new_wl[inside_indices] # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, _, _ = dl.read_stiff(im_path, silent=True, rgb_only=False) npixels = im_hyper.shape[0] * im_hyper.shape[1] # Handle Nuance EX images if im_hyper.shape[2] == 51: # Interpolate to bands in the range [450-950] im_hyper_inside = interp_func(im_hyper, wl, inside_wl) # Extrapolate to 204 bands im_hyper = np.empty((im_hyper_inside.shape[0], im_hyper_inside.shape[1], nbands), dtype=im_hyper_inside.dtype) j = 0 for i in range(nbands): if i < np.min(inside_indices): im_hyper[:, :, i] = im_hyper_inside[:, :, 0] elif i > np.max(inside_indices): im_hyper[:, :, i] = im_hyper_inside[:, :, -1] else: im_hyper[:, :, i] = im_hyper_inside[:, :, j] j += 1 # Handle Specim IQ images else: im_hyper = interp_func(im_hyper, wl, new_wl) # Update std computation im_hyper_sq = im_hyper * im_hyper for i in range(hyper_sums.shape[0]): hyper_sums[i] += (im_hyper_sq[:, :, i] - mean_sq[i]).sum() # Update counters pixel_count += npixels image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('All hyper 400-1000nm (204 bands) std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('All hyper 400-1000nm (204 bands) std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
5,487
29.831461
78
py
segodsidb
segodsidb-main/src/compute_specim_iq_std.py
""" @brief Compute the channel std of the Specim IQ images. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 16 Feb 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-m': 'Channel means normalised to the range [0, 1] (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-m', '--means', required=True, type=str, help=help('-m')) # Read parameters args = parser.parse_args() args.means = eval(args.means) return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # Read means mean_sq = np.array(args.means) * np.array(args.means) # Hyperspectral std computation stuff nbands = 204 hyper_sums = np.zeros((nbands,), dtype=np.float64) pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, im_rgb_orig, metadata = dl.read_stiff(im_path, silent=True, rgb_only=False) # We are only interested in Specim IQ images if im_hyper.shape[2] != nbands: continue # Update std computation im_hyper_sq = im_hyper * im_hyper for i in range(nbands): hyper_sums[i] += (im_hyper_sq[:, :, i] - mean_sq[i]).sum() pixel_count += im_hyper.shape[0] * im_hyper.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('Specim IQ std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('Specim IQ std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
4,105
27.713287
80
py
segodsidb
segodsidb-main/src/compute_all_hyper_400_1000_204_mean.py
""" @brief Compute the channel mean of the 204 interpolated bands from 400-1000nm. This script handles hyperspectral images from both cameras, Nuance EX (450-950nm) and Specim IQ (400-1000nm). @details The channels of both cameras are interpolated to a fixed range define as: np.linspace(400, 1000, 204) However, the number of pixels within the range [450-950] is higher than that in the ranges [400-450] and [950-1000]. This is because for the ranges [400-450] and [950-1000] only the pixels of the Specim IQ contribute, whereas for the range in-between, the images of both cameras contribute. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 7 Jul 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # Hyperspectral mean computation stuff min_wl = 400 max_wl = 1000 nbands = 204 new_wl = np.linspace(min_wl, max_wl, nbands) hyper_sums = np.zeros((new_wl.shape[0],), dtype=np.float64) #inside_pixel_count = 0 #outside_pixel_count = 0 pixel_count = 0 image_count = 0 interp_func = dl.OdsiDbDataLoader.LoadImage.interp_spectra # Get the indices of the wavelengths that are shared among both cameras inside_indices = np.array([idx for idx, x in enumerate(new_wl.tolist()) \ if x > 450. and x < 950.]) inside_wl = new_wl[inside_indices] # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, _, _ = dl.read_stiff(im_path, silent=True, rgb_only=False) npixels = im_hyper.shape[0] * im_hyper.shape[1] # Handle Nuance EX images if im_hyper.shape[2] == 51: # Interpolate to bands in the range [450-950] im_hyper_inside = interp_func(im_hyper, wl, inside_wl) # Extrapolate to 204 bands im_hyper = np.empty((im_hyper_inside.shape[0], im_hyper_inside.shape[1], nbands), dtype=im_hyper_inside.dtype) j = 0 for i in range(nbands): if i < np.min(inside_indices): im_hyper[:, :, i] = im_hyper_inside[:, :, 0] elif i > np.max(inside_indices): im_hyper[:, :, i] = im_hyper_inside[:, :, -1] else: im_hyper[:, :, i] = im_hyper_inside[:, :, j] j += 1 # Handle Specim IQ images else: im_hyper = interp_func(im_hyper, wl, new_wl) # Update average computation for i in range(hyper_sums.shape[0]): hyper_sums[i] += im_hyper[:, :, i].sum() # Update counters pixel_count += npixels image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('All hyper 400-1000nm (204 bands) average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('All hyper 400-1000nm (204 bands) average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') if __name__ == '__main__': main()
5,821
30.989011
80
py
segodsidb
segodsidb-main/src/odsi_db_stats.py
""" @brief Script that show some ODSI-DB stats that are relevant for the paper: - Number of images of 51 bands - Number of images of 204 bands - Number of pixels per class - Number of images in which a class is annotated @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 3 Sep 2021. """ import argparse import numpy as np import os import shutil import sys import tqdm import json import random # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' classnames = torchseg.data_loader.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): """@brief Ensure that input directory exits.""" if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') return args def read_files(path): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def num_images_per_camera(images): """@brief Compute number of images per camera.""" bands_51 = 0 bands_204 = 0 for im_path in images: im, wl, im_rgb, metadata = dl.read_stiff(im_path, silent=True, rgb_only=False) if im.shape[2] == 51: bands_51 += 1 elif im.shape[2] == 204: bands_204 += 1 else: raise ValueError('[ERROR] The image ' + im_path + ' has ' \ + str(im.shape[2]) + ' bands.') return bands_51, bands_204 def get_latex_table(label, columns, data_pix, data_im, caption): """ @brief Get a table with the stats for each class in the dataset. @param[in] label Latex label of the table. @param[in] columns List of column names. @param[in] data_pix Class -> number of pixels. @param[in] data_im Class -> number of images. @param[in] caption Caption of the figure. @returns the string with the latex code. """ latex = "\\begin{table*}[htb!]\n" latex += " \centering\n" latex += " \caption{" + caption + "}\n" latex += " \\vspace{0.2cm}\n" latex += " \\begin{tabular}{lrr}\n" latex += " \\hline\n" latex += " \multicolumn{1}{c}{\\bfseries " + columns[0] + "} &\n" latex += " \multicolumn{1}{c}{\\bfseries " + columns[1] + "} &\n" latex += " \multicolumn{1}{c}{\\bfseries " + columns[2] + "} \\\\ \n" latex += " \hline\n" # Print rows here data_pix = dict(sorted(data_pix.items(), key=lambda item: item[1], reverse=True)) for k in data_pix: latex += ' ' + k + ' & ' + str(data_pix[k]) + ' & ' + str(data_im[k]) + " \\\\ \n" latex += " \end{tabular}\n" latex += " \\vspace{0.2cm}\n" latex += " \label{tab:" + label + "}\n" latex += "\end{table*}" return latex def num_pixels_per_class(images, labels): """@brief Compute the number of pixels belonging to each class.""" class2numpix = {v: 0 for k, v \ in dl.OdsiDbDataLoader.OdsiDbDataset.classnames.items()} class2numim = {v: 0 for k, v \ in dl.OdsiDbDataLoader.OdsiDbDataset.classnames.items()} for label_path in labels: label = torchseg.data_loader.read_mtiff(label_path) for c in label: class2numim[c] += 1 class2numpix[c] += np.count_nonzero(label[c]) return class2numpix, class2numim def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files images, labels = read_files(args.input) # Compute the number of pixels per class class2numpix, class2numim = num_pixels_per_class(images, labels) print(get_latex_table('tab:class_to_numpix', ['Class', 'Number of pixels', 'Number of images'], class2numpix, class2numim, 'Number of pixels per class.')) # Compute the number of images recorded by Nuance EX (51 bannds) and # the number of images recorded by Specim IQ (204 bands) bands_51, bands_204 = num_images_per_camera(images) print('Number of images with 51 bands:', bands_51) print('Number of images with 204 bands:', bands_204) print('Total number of annotated images:', len(images)) if __name__ == '__main__': main()
5,669
31.4
97
py
segodsidb
segodsidb-main/src/compute_class_loss_weights.py
""" @brief Compute the weights for each class of the ODSI-DB dataset based on the number of pixels of each class. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 21 Jun 2022. """ # My imports import torchseg.data_loader as dl pixels_per_class = { u'Skin': 18993686, u'Out of focus area': 8454491, u'Oral mucosa': 7917543, u'Enamel': 4913805, u'Tongue': 4081689, u'Lip': 3920600, u'Hard palate': 2375998, u'Specular reflection': 1931845, u'Attached gingiva': 1922545, u'Soft palate': 1398594, u'Hair': 1383970, u'Marginal gingiva': 804393, u'Prosthetics': 755554, u'Shadow/Noise': 732209, u'Plastic': 255017, u'Metal': 196682, u'Gingivitis': 161874, u'Attrition/Erosion': 100919, u'Inflammation': 81098, u'Pigmentation': 43144, u'Calculus': 28615, u'Initial caries': 22008, u'Stain': 19428, u'Fluorosis': 17872, u'Microfracture': 14759, u'Root': 13962, u'Plaque': 10024, u'Dentine caries': 6616, u'Ulcer': 5552, u'Leukoplakia': 4623, u'Blood vessel': 3667, u'Mole': 2791, u'Malignant lesion': 1304, u'Fibroma': 593, u'Makeup': 406, } def main(): # Make sure that all the classes of ODSI-DB are present in the dic dataloader_classes = [dl.OdsiDbDataLoader.OdsiDbDataset.classnames[k] \ for k in dl.OdsiDbDataLoader.OdsiDbDataset.classnames] for k in dataloader_classes: if k not in pixels_per_class: raise ValueError('[ERROR] The number of pixels of the class ' \ + k + ' is unknown.') # Make sure that the classes present in the dic are in ODSI-DB for k in pixels_per_class: if k not in dataloader_classes: raise ValueError('[ERROR] The class ' + k + ' is not known' \ + ' to the dataloader.') # Count the total number of pixels total = 0 for k, v in pixels_per_class.items(): total += v # Compute 1 - weight of each of the classes weights = {k: 1 - (float(v) / total) for k, v in pixels_per_class.items()} # Print result in Python format print('class_weights = {') for k, v in weights.items(): print(" '" + k + "': " + "{:.9f}".format(v) + ',') print('}') # Print weights using class indices inverted = {v: k for k, v in dl.OdsiDbDataLoader.OdsiDbDataset.classnames.items()} print('class_weights = {') for k, v in weights.items(): number = inverted[k] print(" '" + "{}".format(number) + "': " + "{:.9f}".format(v) + ',') print('}') # Print weights as a vector vector = [weights[dl.OdsiDbDataLoader.OdsiDbDataset.classnames[k]] for k in range(35)] print('class_weights = np.array(', vector, ')') if __name__ == '__main__': main()
3,193
32.621053
90
py
segodsidb
segodsidb-main/src/compute_all_hyper_450_950_51_std.py
""" @brief Compute the channel std of the 51 interpolated bands from 450-950nm. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 21 Jun 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, # e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-m': 'Channel means normalised to the range [0, 1] (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-m', '--means', required=True, type=str, help=help('-m')) # Read parameters args = parser.parse_args() args.means = eval(args.means) return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, _ = read_files(args.input) # Read means mean_sq = np.array(args.means) * np.array(args.means) # Hyperspectral std computation stuff nbands = 51 new_wl = np.linspace(450, 950, nbands) hyper_sums = np.zeros((nbands,), dtype=np.float64) pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, _, _ = dl.read_stiff(im_path, silent=True, rgb_only=False) # Interpolate hyperspectral image to the 450-950nm range (51 bands) interp_func = dl.OdsiDbDataLoader.LoadImage.interp_spectra im_hyper = interp_func(im_hyper, wl, new_wl) # Update std computation im_hyper_sq = im_hyper * im_hyper for i in range(nbands): hyper_sums[i] += (im_hyper_sq[:, :, i] - mean_sq[i]).sum() pixel_count += im_hyper.shape[0] * im_hyper.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('All hyper 450-950nm (51 bands) std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('All hyper 450-950nm (51 bands) std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
4,314
28.353741
78
py
segodsidb
segodsidb-main/src/compute_rgb_mean.py
""" @brief Compute the RGB channel mean of the ODSI-DB reconstructed images. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 16 Feb 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', # '-o': 'Path to the output directory (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) #parser.add_argument('-o', '--output', required=True, type=str, # help=help('-o')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # RGB average red_sum = 0.0 green_sum = 0.0 blue_sum = 0.0 pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, im_rgb_orig, metadata = dl.read_stiff(im_path, silent=True, rgb_only=False) nbands = im_hyper.shape[2] # Convert hyperspectral image to RGB (H, W, 3) im_rgb = dl.OdsiDbDataLoader.LoadImage.hyper2rgb(im_hyper, wl) # Normalise the image in the range [0., 1.] im_rgb = im_rgb.astype(np.float64) / 255. # Update average computation red_sum += im_rgb[:, :, 0].sum() green_sum += im_rgb[:, :, 1].sum() blue_sum += im_rgb[:, :, 2].sum() pixel_count += im_rgb.shape[0] * im_rgb.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) print('Red average:', red_sum / pixel_count) print('Green average:', green_sum / pixel_count) print('Blue average:', blue_sum / pixel_count) # Print final stats print('Image count:', image_count) print('Red average:', red_sum / pixel_count) print('Green average:', green_sum / pixel_count) print('Blue average:', blue_sum / pixel_count) if __name__ == '__main__': main()
3,891
27.617647
80
py
segodsidb
segodsidb-main/src/compute_all_hyper_450_950_170_std.py
""" @brief Compute the channel std of the 170 interpolated bands from 450-950nm. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 21 Jun 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, # e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-m': 'Channel means normalised to the range [0, 1] (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-m', '--means', required=True, type=str, help=help('-m')) # Read parameters args = parser.parse_args() args.means = eval(args.means) return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, _ = read_files(args.input) # Read means mean_sq = np.array(args.means) * np.array(args.means) # Hyperspectral std computation stuff nbands = 170 new_wl = np.linspace(450, 950, nbands) hyper_sums = np.zeros((nbands,), dtype=np.float64) pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, _, _ = dl.read_stiff(im_path, silent=True, rgb_only=False) # Interpolate hyperspectral image to the 450-950nm range (170 bands) interp_func = dl.OdsiDbDataLoader.LoadImage.interp_spectra im_hyper = interp_func(im_hyper, wl, new_wl) # Update std computation im_hyper_sq = im_hyper * im_hyper for i in range(nbands): hyper_sums[i] += (im_hyper_sq[:, :, i] - mean_sq[i]).sum() pixel_count += im_hyper.shape[0] * im_hyper.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('All hyper 450-950nm (170 bands) std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('All hyper 450-950nm (170 bands) std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
4,318
28.380952
78
py
segodsidb
segodsidb-main/src/compute_nuance_ex_mean.py
""" @brief Compute the channel mean of the Nuance EX images. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 16 Feb 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', # '-o': 'Path to the output directory (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) #parser.add_argument('-o', '--output', required=True, type=str, # help=help('-o')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # RGB average nbands = 51 hyper_sums = np.zeros((nbands,), dtype=np.float64) pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, im_rgb_orig, metadata = dl.read_stiff(im_path, silent=True, rgb_only=False) # We are only interested in Nuance EX images if im_hyper.shape[2] != nbands: continue # Update average computation for i in range(nbands): hyper_sums[i] += im_hyper[:, :, i].sum() pixel_count += im_hyper.shape[0] * im_hyper.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('Nuance EX average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('Nuance EX average: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
3,875
27.086957
80
py
segodsidb
segodsidb-main/src/compute_nuance_ex_std.py
""" @brief Compute the channel std of the Nuance EX images. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 16 Feb 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 import sys # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-m': 'Channel means normalised to the range [0, 1] (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-m', '--means', required=True, type=str, help=help('-m')) # Read parameters args = parser.parse_args() args.means = eval(args.means) return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # Read means mean_sq = np.array(args.means) * np.array(args.means) # Hyperspectral std computation stuff nbands = 51 hyper_sums = np.zeros((nbands,), dtype=np.float64) pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, im_rgb_orig, metadata = dl.read_stiff(im_path, silent=True, rgb_only=False) # We are only interested in Nuance EX images if im_hyper.shape[2] != nbands: continue # Update std computation im_hyper_sq = im_hyper * im_hyper for i in range(nbands): hyper_sums[i] += (im_hyper_sq[:, :, i] - mean_sq[i]).sum() pixel_count += im_hyper.shape[0] * im_hyper.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) sys.stdout.write('Nuance EX std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']\n') # Print final stats print('Image count:', image_count) sys.stdout.write('Nuance EX std: [') for i in range(nbands): sys.stdout.write(str(hyper_sums[i] / pixel_count)) if i < nbands - 1: sys.stdout.write(', ') sys.stdout.write(']') if __name__ == '__main__': main()
4,104
27.706294
80
py
segodsidb
segodsidb-main/src/compute_rgb_std.py
""" @brief Compute the std of the RGB images in the ODSI-DB dataset, normalised to the range [0, 1]. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 16 Feb 2022. """ import argparse import numpy as np import os import ntpath import shutil import tqdm import cv2 # My imports import torchseg.data_loader as dl import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-m': 'RGB means normalised to the range [0, 1] (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-m', '--means', required=True, type=str, help=help('-m')) # Read parameters args = parser.parse_args() args.means = eval(args.means) return args def validate_cmdline_params(args): """ @brief Make sure that input directory exists. @returns nothing. """ if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # Read means red_mean_sq = args.means[0] * args.means[0] green_mean_sq = args.means[1] * args.means[1] blue_mean_sq = args.means[2] * args.means[2] # RGB average red_sum = 0.0 green_sum = 0.0 blue_sum = 0.0 pixel_count = 0 image_count = 0 # Print interval interval = 10 # Loop over the images for im_path in tqdm.tqdm(image_paths): im_fname = os.path.splitext(ntpath.basename(im_path))[0] # Read image im_hyper, wl, im_rgb_orig, metadata = dl.read_stiff(im_path, silent=True, rgb_only=False) nbands = im_hyper.shape[2] # Convert hyperspectral image to RGB (H, W, 3) im_rgb = dl.OdsiDbDataLoader.LoadImage.hyper2rgb(im_hyper, wl) # Normalise the image in the range [0., 1.] im_rgb = im_rgb.astype(np.float64) / 255. # Update std computation im_rgb_sq = im_rgb * im_rgb red_sum += (im_rgb_sq[:, :, 0] - red_mean_sq).sum() green_sum += (im_rgb_sq[:, :, 1] - green_mean_sq).sum() blue_sum += (im_rgb_sq[:, :, 2] - blue_mean_sq).sum() pixel_count += im_rgb.shape[0] * im_rgb.shape[1] image_count += 1 # Print intermediate stats if image_count % interval == 0: print('Image count:', image_count) print('Red std:', np.sqrt(red_sum / pixel_count)) print('Green std:', np.sqrt(green_sum / pixel_count)) print('Blue std:', np.sqrt(blue_sum / pixel_count)) # Print final stats print('Image count:', image_count) print('Red std:', np.sqrt(red_sum / pixel_count)) print('Green std:', np.sqrt(green_sum / pixel_count)) print('Blue std:', np.sqrt(blue_sum / pixel_count)) if __name__ == '__main__': main()
4,390
29.282759
80
py
segodsidb
segodsidb-main/src/validate_class_presence.py
""" @brief We want to use image-level annotations to learn the segmentation. Therefore, we need to make sure that there is at least one image (ideally more) where each of the classes is not present. The opposite is not a concern, as we know that all the classes are present at least in one image. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 27 Apr 2022. """ import argparse import numpy as np import os import sys # My imports import torchseg.data_loader as dl import torchseg.utils def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-o': 'Path to the output directory (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): # Input directory must exist if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') return args def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def class_in_label(class_name: str, label: np.ndarray) -> bool: # Get class index idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames class2idx = {y: x for x, y in idx2class.items()} class_id = class2idx[class_name] # Check if there is any pixel of this class return bool(np.count_nonzero(label[:, :, class_id])) def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files sys.stdout.write('[INFO] Reading list of input files... ') sys.stdout.flush() image_paths, label_paths = read_files(args.input) print('OK') # Initially we assume that all the classes are in all the images present_in_all = {v: True for v in dl.OdsiDbDataLoader.OdsiDbDataset.classnames.values()} # Loop over the images for im_path, label_path in zip(image_paths, label_paths): # Real label label = dl.OdsiDbDataLoader.LoadImage.read_label( label_path).transpose((1, 2, 0)) # Loop over the classes and find out which are not present for c in present_in_all: # If we know that this class was not present in a previous image # we do not need to check this image if not present_in_all[c]: continue # If the class is not present in the image, we write it down if not class_in_label(c, label): present_in_all[c] = False # Make sure that all the classes are missing from at least one image success = True for c in present_in_all: if present_in_all[c]: success = False break if success: print("[OK] You can use image-level presence labels with the ODSI-DB dataset.") else: print("[ERROR] You cannot use image-level presence labels with the ODSI_DB dataset.") print("Check the classes that are present in ALL the images below:") print(present_in_all) if __name__ == '__main__': main()
4,147
30.664122
93
py
segodsidb
segodsidb-main/src/produce_cmf_plot.py
#!/usr/bin/env python3 # # @brief Script to produce HSI to XYZ conversion colour matching function. # # @author Luis C. Garcia Peraza Herrera (luiscarlos.gph@gmail.com). # @date 23 Aug 2022. import numpy as np import seaborn as sns import matplotlib.pyplot as plt # My imports import torchseg.data_loader as dl def main(): plt.rc('font', size=12) # Produce CMF function cmf = 'cie_2_1931' min_wl = 380 max_wl = 780 steps = max_wl - min_wl + 1 wl = new_wl=np.linspace(min_wl, max_wl, steps) f_xbar, f_ybar, f_zbar = \ dl.OdsiDbDataLoader.LoadImage.get_corrected_cmf(cmf, wl) # Plot CMF xbar = f_xbar(wl) ybar = f_ybar(wl) zbar = f_zbar(wl) cmf_plot = sns.lineplot(x=wl, y=xbar, color='red', label='X') cmf_plot = sns.lineplot(x=wl, y=ybar, color='green', label='Y') cmf_plot = sns.lineplot(x=wl, y=zbar, color='blue', label='Z') # Compute the integral of the CMFs xbar_integral = np.trapz(xbar, wl) ybar_integral = np.trapz(ybar, wl) zbar_integral = np.trapz(zbar, wl) print('Original CMF xbar integral:', xbar_integral) print('Original CMF ybar integral:', ybar_integral) print('Original CMF zbar integral:', zbar_integral) print() # Save CMF plot to disk plt.title('CIE XYZ standard observer CMFs') plt.xlim([380, 780]) plt.ylim([0., 4.]) plt.xticks([x for x in range(380, 790, 10)], ['' if x not in [y for y in range(380, 800, 40)] else x for x in range(380, 790, 10)]) plt.legend() plt.xlabel('Wavelength (nm)') fig = cmf_plot.get_figure() fig.savefig('cie_2_1931.svg', format='svg', dpi=600) # Clear plot plt.clf() # Produce modified CMF function min_wl = 400 max_wl = 780 steps = max_wl - min_wl + 1 wl = new_wl=np.linspace(min_wl, max_wl, steps) f_xbar, f_ybar, f_zbar = \ dl.OdsiDbDataLoader.LoadImage.get_corrected_cmf(cmf, wl) # Plot modified CMF xbar = f_xbar(wl) ybar = f_ybar(wl) zbar = f_zbar(wl) cmf_plot = sns.lineplot(x=wl, y=xbar, color='red', label='X') cmf_plot = sns.lineplot(x=wl, y=ybar, color='green', label='Y') cmf_plot = sns.lineplot(x=wl, y=zbar, color='blue', label='Z') # Compute the integral of the CMFs xbar_integral = np.trapz(xbar, wl) ybar_integral = np.trapz(ybar, wl) zbar_integral = np.trapz(zbar, wl) print('Specim IQ CMF xbar integral:', xbar_integral) print('Specim IQ CMF ybar integral:', ybar_integral) print('Specim IQ CMF zbar integral:', zbar_integral) print() # Save modified CMF to disk plt.title('Modified CIE XYZ CMFs for Specim IQ') plt.xlim([380, 780]) plt.ylim([0., 4.]) plt.xticks([x for x in range(380, 790, 10)], ['' if x not in [y for y in range(380, 800, 40)] else x for x in range(380, 790, 10)]) plt.legend() plt.xlabel('Wavelength (nm)') fig = cmf_plot.get_figure() fig.savefig('cie_2_1931_specim_iq.svg', format='svg', dpi=600) # Clear plot plt.clf() # Produce modified CMF function min_wl = 450 max_wl = 780 steps = max_wl - min_wl + 1 wl = new_wl=np.linspace(min_wl, max_wl, steps) f_xbar, f_ybar, f_zbar = \ dl.OdsiDbDataLoader.LoadImage.get_corrected_cmf(cmf, wl) # Plot modified CMF xbar = f_xbar(wl) ybar = f_ybar(wl) zbar = f_zbar(wl) cmf_plot = sns.lineplot(x=wl, y=xbar, color='red', label='X') cmf_plot = sns.lineplot(x=wl, y=ybar, color='green', label='Y') cmf_plot = sns.lineplot(x=wl, y=zbar, color='blue', label='Z') # Compute the integral of the CMFs xbar_integral = np.trapz(xbar, wl) ybar_integral = np.trapz(ybar, wl) zbar_integral = np.trapz(zbar, wl) print('Nuance EX CMF xbar integral:', xbar_integral) print('Nuance EX CMF ybar integral:', ybar_integral) print('Nuance EX CMF zbar integral:', zbar_integral) print() # Save modified CMF to disk plt.title('Modified CIE XYZ CMFs for Nuance EX') plt.xlim([380, 780]) plt.ylim([0., 4.]) plt.xticks([x for x in range(380, 790, 10)], ['' if x not in [y for y in range(380, 800, 40)] else x for x in range(380, 790, 10)]) plt.legend() plt.xlabel('Wavelength (nm)') fig = cmf_plot.get_figure() fig.savefig('cie_2_1931_nuance_ex.svg', format='svg', dpi=600) if __name__ == '__main__': main()
4,371
31.626866
135
py
segodsidb
segodsidb-main/src/train.py
""" @brief Main script to kick off the training. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 1 Jun 2021. """ import argparse import collections import torch import numpy as np # My imports import torchseg.config.parser import torchseg.data_loader import torchseg.model import torchseg.machine # Fix random seeds for reproducibility SEED = 18303 torch.manual_seed(SEED) torch.backends.cudnn.deterministic = True torch.backends.cudnn.benchmark = False np.random.seed(SEED) def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-c': 'config file path (default: None)', '-l': 'logger config file path (default: None)', '-r': 'path to latest checkpoint (default: None)', '-d': 'indices of GPUs to enable (default: all)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" args = argparse.ArgumentParser(description='PyTorch segmenter.') args.add_argument('-c', '--conf', default=None, type=str, help=help('-c')) args.add_argument('-l', '--logconf', default=None, type=str, help=help('-l')) args.add_argument('-r', '--resume', default=None, type=str, help=help('-r')) args.add_argument('-d', '--device', default=None, type=str, help=help('-d')) return args def parse_config(args): """ @brief Combines parameters from both JSON and command line into a @param[in] args Argparse args object. @returns A torchseg.config.parser.ConfigParser object. """ # Custom CLI options to modify the values provided in the JSON configuration CustomArgs = collections.namedtuple('CustomArgs', 'flags type target') options = [ CustomArgs(['--lr', '--learning-rate'], type=float, target='optimizer;args;lr'), CustomArgs(['--bs', '--batch-size'], type=int, target='data_loader;args;batch_size'), CustomArgs(['--data-dir'], type=str, target='data_loader;args;data_dir'), CustomArgs(['--save-dir'], type=str, target='machine;args;save_dir'), ] config = torchseg.config.parser.ConfigParser.from_args(args, options) return config def main(): args = parse_cmdline_params() config = parse_config(args) # Note: the 'type' in config.json indicates the class, and the 'args' in # config.json will be passed as parameters to the constructor of that class # Get training logger logger = config.get_logger('train') # Setup data loader data_loader = config.init_obj('data_loader', torchseg.data_loader) valid_data_loader = data_loader.split_validation() # Create model model = config.init_obj('model', torchseg.model) logger.info(model) # Prepare for (multi-device) GPU training device, device_ids = torchseg.utils.setup_gpu_devices(config['n_gpu']) model = model.to(device) if len(device_ids) > 1: model = torch.nn.DataParallel(model, device_ids=device_ids) # Get function handles of loss and evaluation metrics criterion = getattr(torchseg.model.loss, config['loss']) metrics = [getattr(torchseg.model.metric, m) for m in config['metrics']] # Create optmizer trainable_params = filter(lambda p: p.requires_grad, model.parameters()) optimizer = config.init_obj('optimizer', torch.optim, trainable_params) # Learning rate scheduler lr_scheduler = config.init_obj('lr_scheduler', torch.optim.lr_scheduler, optimizer) # Create learning machine trainer = torchseg.machine.GenericMachine(model, criterion, metrics, optimizer, config=config, device=device, data_loader=data_loader, valid_data_loader=valid_data_loader, lr_scheduler=lr_scheduler, acc_steps=config['machine']['args']['acc_steps']) # TODO: We should be able to create the learning machine just with this type # of call #trainer = config.init_obj('machine', torchseg.machine) # Launch training trainer.train() if __name__ == '__main__': main()
4,439
34.238095
95
py
segodsidb
segodsidb-main/src/compute_pixel_stats.py
""" @brief Script that computes iteratively the mean and unbiased sample standard deviation. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 15 Dec 2021. """ import argparse import numpy as np import os import copy import tqdm #import random #import pandas as pd #import seaborn as sns #import matplotlib #import matplotlib.pyplot as plt #import sklearn.manifold #import multiprocessing #import colour # My imports import torchseg.data_loader import torchseg.utils # Get dictionary from class index to class string, e.g. 0 -> u'Attached gingiva' idx2class = torchseg.data_loader.OdsiDbDataLoader.OdsiDbDataset.classnames def help(short_option): """ @returns The string with the help information for each command line option. """ help_msg = { '-i': 'Path to the root of the ODSI-DB dataset (required: True)', '-m': 'Path to the output mean (required: True)', '-s': 'Path to the output std (required: True)', } return help_msg[short_option] def parse_cmdline_params(): """@returns The argparse args object.""" # Create command line parser parser = argparse.ArgumentParser(description='PyTorch segmenter.') parser.add_argument('-i', '--input', required=True, type=str, help=help('-i')) parser.add_argument('-m', '--mean', required=True, type=str, help=help('-m')) parser.add_argument('-s', '--std', required=True, type=str, help=help('-s')) # Read parameters args = parser.parse_args() return args def validate_cmdline_params(args): if not os.path.isdir(args.input): raise RuntimeError('[ERROR] Input directory does not exist.') if os.path.exists(args.mean): raise RuntimeError('[ERROR] Mean file already exists.') if os.path.exists(args.std): raise RuntimeError('[ERROR] Std file already exists.') return args def read_files(path: str): """ @param[in] path Path to the folder with the ODSI-DB data. @returns a tuple (imgs, segs) containing two lists of paths to images and segmentation labels. """ # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(path) if '.tif' in f] # Get list of segmentation files segs = [os.path.join(path, f) for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(path, im))) return imgs, segs def main(): # Read command line parameters args = parse_cmdline_params() validate_cmdline_params(args) # Get list of image and segmentation files image_paths, label_paths = read_files(args.input) # Get number of spectral channels im_hyper, _, _, _ = torchseg.data_loader.read_stiff(image_paths[0], silent=True, rgb_only=False) # Compute stats on the dataset nchan = im_hyper.shape[2] mean = np.zeros((nchan,)) m2 = np.zeros((nchan,)) n = 0 for im_path, label_path in tqdm.tqdm(zip(image_paths, label_paths)): # Load image (h, w, c) im_hyper, _, _, _ = torchseg.data_loader.read_stiff(image_paths[0], silent=True, rgb_only=False) # Update the iterative stats h, w, c = im_hyper.shape for i in range(h): for j in range(w): n += 1 x_n = im_hyper[i, j] old_mean = copy.deepcopy(mean) mean += (x_n - mean) / n m2 += (x_n - old_mean) * (x_n - mean) # Compute unbiased sample std std = np.sqrt(m2 / (n - 1)) # Save mean and std to file meanfile = open(args.mean, 'w') stdfile = open(args.std, 'w') meanfile.write('[') stdfile.write('[') for i in range(nchan): meanfile.write("{:.6f}".format(mean[i]) + ', ') stdfile.write("{:.6f}".format(std[i]) + ', ') meanfile.write("]\n") stdfile.write("]\n") meanfile.close() stdfile.close() if __name__ == '__main__': main()
4,349
28.794521
80
py
segodsidb
segodsidb-main/src/data_loader/data_loader.py
""" @brief Collection of data loaders for different datasets. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 2 Jun 2021. """ import os import torch import torchvision import PIL import numpy as np import random import colour import scipy import matplotlib.pyplot as plt # My imports import torchseg.base import torchseg.utils import torchseg.data_loader as dl import monai.data.dataset class MnistDataLoader(torchseg.base.BaseDataLoader): """ @class MnistDataLoader loads MNIST data. """ def __init__(self, data_dir, batch_size, shuffle=True, validation_split=0.0, num_workers=1, training=True): # Store parameters self.data_dir = data_dir # Create normalisation transform trsfm = torchvision.transforms.Compose([ torchvision.transforms.ToTensor(), torchvision.transforms.Normalize((0.1307,), (0.3081,)) ]) # Create dataset handler self.dataset = torchvision.datasets.MNIST(self.data_dir, train=training, download=True, transform=trsfm) # Call the constructor of torchseg.base.BaseDataLoader super().__init__(self.dataset, batch_size, shuffle, validation_split, num_workers) class OdsiDbDataLoader(torchseg.base.BaseDataLoader): # Input image modes modes = [ 'rgbpixel', 'rgbpixel_test', 'rgbimage', 'spixel_51', 'simage_51', 'spixel_170', 'spixel_170_test', 'simage_170', 'spixel_204', 'simage_204', 'boiko', ] # The array of wavelengths depends on the input mode mode2wl = { 'spixel_51' : np.linspace(450, 950, 51), 'simage_51' : np.linspace(450, 950, 51), 'spixel_170': np.linspace(450, 950, 170), 'spixel_170_test': np.linspace(450, 950, 170), 'simage_170': np.linspace(450, 950, 170), 'spixel_204': np.linspace(400, 1000, 204), 'simage_204': np.linspace(400, 1000, 204), 'boiko' : np.array([450., 500., 600.]), } # RGB normalisation stats rgb_mean = np.array([0.403129887605, 0.288007343691, 0.299380141092], dtype=np.float32) rgb_std = np.array([0.159463123785, 0.130188316525, 0.12346950031], dtype=np.float32) # Nuance EX normalisation stats nuance_ex_mean = np.array([ 0.0776138608947, 0.0783226294712, 0.0915784099028, 0.0851693853982, 0.0831637362668, 0.0847127943303, 0.085340519524, 0.0845109800243, 0.0790671446523, 0.0751274680049, 0.0761701337076, 0.0781188917647, 0.0782783973927, 0.0825329193882, 0.0988390369608, 0.120948266959, 0.140439376036, 0.154276410607, 0.164172320211, 0.171390068447, 0.178038853241, 0.184233131521, 0.189796964942, 0.195031949356, 0.198907463087, 0.198520651813, 0.200358926766, 0.203468102035, 0.201678233612, 0.198429438372, 0.193854994342, 0.191312533015, 0.189530537391, 0.18868570976, 0.18854169948, 0.189165124177, 0.189642954468, 0.189296877222, 0.180562774436, 0.177428725348, 0.174671366762, 0.172125912418, 0.16933901838, 0.167608797022, 0.165615862068, 0.163236092449, 0.16056347767, 0.156600176469, 0.143126733362, 0.138918268583, 0.135092694289, ], dtype=np.float32) nuance_ex_std = np.array([ 0.0038290728339, 0.00386745890456, 0.00461893938556, 0.00427318759442, 0.00421193344028, 0.00433005583596, 0.00441629367651, 0.00442865858953, 0.00416620537279, 0.00409641547929, 0.00419634961461, 0.0043631548856, 0.00440318450805, 0.00471044520012, 0.00585655629628, 0.00776347677282, 0.00967594428827, 0.0112555765706, 0.0125962817541, 0.0135586755183, 0.0144026812212, 0.015031395142, 0.0156616627971, 0.0162450380476, 0.016777610301, 0.0171646905704, 0.0174596376809, 0.0175789718117, 0.0174627609924, 0.0171899509512, 0.0168640925817, 0.0167477550425, 0.0166329796535, 0.016384616472, 0.0162204907189, 0.0160845513226, 0.015832708941, 0.0154426560894, 0.0146519935725, 0.0140359361608, 0.0134760286473, 0.0129535803584, 0.0124506272206, 0.0117099903208, 0.0112119812513, 0.0107739222483, 0.0103392189441, 0.0100008831543, 0.00827122500705, 0.00774512242113, 0.00726368979869, ], dtype=np.float32) # Specim IQ normalisation stats specim_iq_mean = np.array([ 0.112085002548, 0.102140085151, 0.0943113060797, 0.0885467882906, 0.0846466315166, 0.0820774967744, 0.0802850078561, 0.0791292167016, 0.0784498196042, 0.0780782467903, 0.0780644673204, 0.0784523791739, 0.0792390083204, 0.0804284923453, 0.0818753314213, 0.0834831640881, 0.0851511772687, 0.0867686173785, 0.0883751009176, 0.0898204818059, 0.0912809000747, 0.0926440870137, 0.0939971489817, 0.0952519009781, 0.0965328084729, 0.097690027486, 0.0987274062461, 0.0996016817623, 0.100458407892, 0.101202582001, 0.10195598233, 0.102806150098, 0.103694965839, 0.104664556992, 0.105653205393, 0.10659280337, 0.10746400651, 0.108253455173, 0.108935761923, 0.109321392809, 0.109375497886, 0.108865325572, 0.107869014925, 0.106254806345, 0.104402478702, 0.102317535445, 0.100404782722, 0.0988052551902, 0.0977005111836, 0.0969098524837, 0.0966750172086, 0.0968155206467, 0.0974678382789, 0.0984965585346, 0.0997345260659, 0.100758822558, 0.101404000532, 0.101469627798, 0.101135541391, 0.100144589499, 0.0986686515618, 0.0974109084362, 0.0973515855222, 0.0992310783409, 0.103559105299, 0.110339683424, 0.119200190181, 0.129200049368, 0.139490848038, 0.149170142667, 0.158144175815, 0.166132853615, 0.173451190929, 0.179981349582, 0.185961931089, 0.191155738785, 0.195689772047, 0.199596934812, 0.203163066338, 0.206382320236, 0.209400294235, 0.212104318164, 0.214609173942, 0.216990035595, 0.219227218176, 0.221412241784, 0.223472401843, 0.225275373864, 0.227072332864, 0.228744232706, 0.230303433887, 0.231838015095, 0.233366165061, 0.234887439646, 0.236506400265, 0.237957146659, 0.239588794172, 0.241068114, 0.242681232832, 0.244230795064, 0.245745089833, 0.24692634115, 0.248064798697, 0.248622792136, 0.249324235054, 0.249977211781, 0.250498342546, 0.250987490527, 0.251156369557, 0.251248702127, 0.25096838481, 0.250822536009, 0.250102103495, 0.249356839651, 0.248629243344, 0.247524143344, 0.246625750813, 0.245947435307, 0.245181069521, 0.243986731324, 0.242938204373, 0.242353213664, 0.241708338683, 0.241195778181, 0.241245919936, 0.241264032642, 0.241610304121, 0.242049243169, 0.2421130799, 0.242264614189, 0.242367022354, 0.2409313546, 0.241159969908, 0.242161924135, 0.242228185646, 0.242473373156, 0.242266678267, 0.24223316472, 0.241815605688, 0.241536003903, 0.240810299738, 0.239586301619, 0.238636256726, 0.238204898501, 0.237551727056, 0.236373644054, 0.235011936757, 0.233764670484, 0.232366132656, 0.231146941133, 0.23080009416, 0.230533240929, 0.230076653688, 0.229484626509, 0.228599976517, 0.227914845773, 0.227556968449, 0.22725171334, 0.226303809768, 0.225347473023, 0.224614910223, 0.224014662255, 0.22339760286, 0.222414620401, 0.221382271403, 0.220377157411, 0.219338854373, 0.218130296434, 0.217091878005, 0.215903149325, 0.214644602196, 0.213494551926, 0.212338519749, 0.211226395102, 0.210369281188, 0.209304051737, 0.207966701355, 0.206544497363, 0.204824802113, 0.203001666349, 0.201089720809, 0.199036227562, 0.196907502338, 0.194647061506, 0.191861582388, 0.188844590578, 0.18527816507, 0.181248328695, 0.1777737335, 0.175078837825, 0.1733158301, 0.172441615107, 0.171608116792, 0.171192572651, 0.171049377519, 0.170862948778, 0.171274492165, 0.171573746293, 0.172177039069, 0.172957085572, 0.173301498106, 0.174158032792, 0.174963397296, 0.161150677433 ], dtype=np.float32) specim_iq_std = np.array([ 0.0051204327844, 0.0049010894168, 0.00490408951459, 0.00501242131618, 0.00515869582467, 0.00531481044734, 0.00547126338076, 0.00563346798113, 0.00579447352576, 0.00596555732144, 0.00615566882416, 0.00635973070577, 0.00657340461617, 0.0067967285048, 0.0070286671205, 0.00727708514704, 0.00752424001689, 0.00775700912999, 0.00798220085672, 0.00818689563432, 0.00838465923951, 0.00856988323327, 0.0087392056371, 0.00889675897692, 0.00905206000264, 0.0091959914293, 0.009312072202, 0.00940073464096, 0.00947625651694, 0.00953227890046, 0.00957837186835, 0.00964017941934, 0.00969605544023, 0.00977571641123, 0.00985408362882, 0.00993917538033, 0.0100143769771, 0.0100859121458, 0.0101409295895, 0.0101669455923, 0.0101443092583, 0.0100463008317, 0.00989182845609, 0.00966438387347, 0.00942136906458, 0.00917258006915, 0.008953890893, 0.00878593637129, 0.00867668098681, 0.00860270653642, 0.00859433373635, 0.0086204846356, 0.00869749103483, 0.00882064339864, 0.00896339789551, 0.00909145584191, 0.00918067503764, 0.00920361403469, 0.00919296874221, 0.00911463416519, 0.00897490062522, 0.00886443804711, 0.00888037267064, 0.00910847772243, 0.00962114957633, 0.0104625277379, 0.0116591370095, 0.0131564475032, 0.014852059166, 0.0165929757477, 0.0183182801335, 0.019935233596, 0.0214820271417, 0.0229137210796, 0.0242616624892, 0.0254825335416, 0.0265475514249, 0.0274784619776, 0.0283352254374, 0.0291056673323, 0.0298353614964, 0.0304790896885, 0.0310573625231, 0.0316169604888, 0.0321468398482, 0.0326668677501, 0.0331730392939, 0.0335956874147, 0.0340282795814, 0.0344635042052, 0.0348561658458, 0.0352599262886, 0.0356911594505, 0.0360880062706, 0.0365148940448, 0.0369118063317, 0.0373465100767, 0.0377660923652, 0.0382188255657, 0.0386452354932, 0.0390436577487, 0.0393508233, 0.0396008200152, 0.0396906081602, 0.0398572928164, 0.0399856573532, 0.0400535027338, 0.0401183618839, 0.0400770409866, 0.0400355447514, 0.0398820349639, 0.0397914604514, 0.0395229589731, 0.0392744169267, 0.0390286206185, 0.0386696315616, 0.0383528940588, 0.0381384245943, 0.0378284388325, 0.0374353366368, 0.0370882825179, 0.0368806895126, 0.0366293825145, 0.0364590994265, 0.036409711205, 0.0363757080885, 0.0364176906163, 0.0364910524371, 0.0364356384961, 0.0364550964887, 0.0363572534936, 0.035713651417, 0.0357296075661, 0.0360257484172, 0.036304159429, 0.0367355386762, 0.0366726984998, 0.0366845383865, 0.0365487921643, 0.0364630193957, 0.0362154434343, 0.0358276964578, 0.0355990390691, 0.0355736606419, 0.0354255001813, 0.0351446246617, 0.0347914317484, 0.0344802785938, 0.0341160734367, 0.0337945665366, 0.0336832205541, 0.0336308778566, 0.0334985680856, 0.0333494362052, 0.0331237910006, 0.0329646205593, 0.0328705891889, 0.0328222972035, 0.0326090837709, 0.0324180583771, 0.0322387706372, 0.0320901724742, 0.0319185152311, 0.0316350319596, 0.0313471149088, 0.031120219116, 0.0308533330801, 0.0305213851429, 0.0302389431622, 0.029960570957, 0.0296949319738, 0.0294617564081, 0.0291807764587, 0.0288812632713, 0.0286547736342, 0.0283600323959, 0.0279933843565, 0.0276050068161, 0.027095059129, 0.0265664812623, 0.026083041933, 0.0256118408366, 0.025073052516, 0.0244892219447, 0.0238096923167, 0.0230908555588, 0.0222759083261, 0.0213622231765, 0.0205653866989, 0.0199707523122, 0.0194896643608, 0.0192172825747, 0.0189468026241, 0.0187808803217, 0.0186485068087, 0.0184880364592, 0.0184822248112, 0.0184771573814, 0.0184898120673, 0.0185603547897, 0.018496862837, 0.0185234983434, 0.0185140609897, 0.0130384043098, ], dtype=np.float32) # Normalisation stats for interpolated bands 450-950nm (51 bands) hyper_51_mean = np.array([ 0.0810067996349, 0.0829831647364, 0.093483960472, 0.0899734763065, 0.0894423085798, 0.0915193721587, 0.092661139279, 0.0915338960809, 0.0857899568086, 0.0817884475053, 0.0828441900691, 0.0852258059802, 0.0848364019569, 0.0874677529617, 0.105330525865, 0.130889929167, 0.152258716436, 0.167234953479, 0.177657939938, 0.185337380804, 0.192140167946, 0.198199267684, 0.203636422124, 0.208865833329, 0.213144227103, 0.213884648171, 0.215777113836, 0.218022347793, 0.216236726977, 0.213079549878, 0.208936060479, 0.206582950682, 0.205533508108, 0.205109621906, 0.204955697994, 0.205414174429, 0.205367417871, 0.204263258639, 0.197107520177, 0.193805749636, 0.191465798416, 0.189074327189, 0.186375384941, 0.184456663586, 0.182047172551, 0.179237997628, 0.176214905006, 0.172386557792, 0.161391944603, 0.156415102283, 0.150869385165 ], dtype=np.float32) hyper_51_std = np.array([ 0.00514014020925, 0.00539040833008, 0.00603190822433, 0.0059271066889, 0.00597948945687, 0.00616100200841, 0.00629645035981, 0.00619439587162, 0.00575553630156, 0.00557603115215, 0.00569343612698, 0.00594936225045, 0.00592971065021, 0.00609018586403, 0.00776464603105, 0.0109345712314, 0.0139807234261, 0.0163833898851, 0.0181979949318, 0.0195246053556, 0.0206441248362, 0.0215073191503, 0.0223579382643, 0.0231928340066, 0.0240259381558, 0.0246012694501, 0.0249232918579, 0.0248939753191, 0.0246073320615, 0.0241213238954, 0.0235886267345, 0.0233070117303, 0.023279264433, 0.0231155168478, 0.022900319036, 0.0229853245983, 0.022653123846, 0.0221113475025, 0.0214060499578, 0.0206702027159, 0.0202024043012, 0.0196976652234, 0.0191932768321, 0.0184724650055, 0.017832398352, 0.0172010450473, 0.0166074947096, 0.0160897372069, 0.0146111361932, 0.0136754420132, 0.0125149234869, ], dtype=np.float32) # Normalisation stats for interpolated bands 450-950nm (170 bands) hyper_170_mean = np.array([ 0.0810067996349, 0.0816043995643, 0.0821970031859, 0.0827673146001, 0.0849229440499, 0.0880378175387, 0.0911460114559, 0.0932456460856, 0.0922341744158, 0.09118836566, 0.0901252695013, 0.0898196994614, 0.0896553436568, 0.089513093682, 0.0897392633201, 0.0903613267805, 0.090979011488, 0.0915592812428, 0.0919451792569, 0.0923012143522, 0.0925915664477, 0.0925672795887, 0.0923141765019, 0.0918983693193, 0.0909842773978, 0.0893221808685, 0.0875870100037, 0.0858583116221, 0.084507488729, 0.0833029689943, 0.0822162716762, 0.0818705060695, 0.0820900963864, 0.0824551648085, 0.0829875784395, 0.0837494090013, 0.0844853113882, 0.0851180245572, 0.0852873398232, 0.0852510529314, 0.085032962395, 0.0850213767293, 0.0854903869798, 0.0862520932472, 0.0877033510983, 0.0921965406489, 0.0974439886213, 0.103359354549, 0.110504447195, 0.1181816836, 0.125735154289, 0.132916819701, 0.139431940908, 0.14572431233, 0.151778762252, 0.156579367882, 0.161062623956, 0.16533550223, 0.16897751279, 0.172123531667, 0.175163861539, 0.178042253332, 0.180372587267, 0.182632682035, 0.184851135499, 0.186934765385, 0.188969781225, 0.19097210744, 0.192868746172, 0.194688788255, 0.196478620273, 0.198231629835, 0.199843773024, 0.201451354722, 0.203056473628, 0.204639118831, 0.206169527005, 0.207723786193, 0.209189575556, 0.210462771206, 0.211735450466, 0.212993707719, 0.21341177993, 0.213682033696, 0.213833884439, 0.214179921524, 0.214760809264, 0.215310997878, 0.21587149132, 0.216600461745, 0.217276433301, 0.217878384209, 0.217710349168, 0.217203954087, 0.216615029805, 0.215917558555, 0.214973406228, 0.214003477609, 0.213096996496, 0.21194623461, 0.210718301228, 0.209438320294, 0.208491630456, 0.207786470717, 0.207085517874, 0.206505790068, 0.206151632208, 0.205824085203, 0.205570909186, 0.205465349108, 0.205318419385, 0.205186642554, 0.204999346857, 0.204648895535, 0.204734935304, 0.204967520688, 0.205123997043, 0.205306585329, 0.205378787542, 0.205464869726, 0.205440862705, 0.205448678606, 0.205270700196, 0.204834264548, 0.204479281473, 0.204276377615, 0.202394212481, 0.20025954336, 0.1980644597, 0.196538453003, 0.195484550546, 0.194468031395, 0.193685258656, 0.1930344551, 0.192345946229, 0.191614290102, 0.190845453654, 0.190099431202, 0.189436042822, 0.188795821318, 0.188020086576, 0.187164688065, 0.186352484731, 0.18579908009, 0.185262986887, 0.184676669537, 0.18398799608, 0.183275393733, 0.182564743891, 0.181809907514, 0.180977080903, 0.180169855655, 0.179324102047, 0.178414858592, 0.177525557928, 0.176636254971, 0.175626738953, 0.174556558391, 0.17342668134, 0.171943895798, 0.168758784331, 0.165489093699, 0.162187915121, 0.160305287515, 0.158842194262, 0.157354797792, 0.15586327933, 0.154299476375, 0.152647749572, 0.150869385165, ], dtype=np.float32) hyper_170_std = np.array([ 0.00514014020925, 0.00518764780204, 0.00525823383173, 0.00535012794449, 0.00547291214664, 0.00563897203488, 0.00584719693034, 0.00601840401274, 0.00597312393174, 0.00594230808148, 0.00592796769322, 0.00592149651265, 0.00593073969924, 0.00595863235588, 0.00599366047689, 0.00603598420629, 0.00609498985838, 0.00616283944647, 0.00619021429061, 0.00623129889715, 0.00628079540453, 0.00628164484446, 0.0062554232367, 0.00622091327839, 0.00614385296943, 0.00600149880839, 0.00586976669371, 0.00575943407638, 0.00566313714922, 0.00560383617056, 0.00557784065547, 0.00557070429369, 0.00558360532025, 0.00563066298374, 0.00570444278682, 0.00577308272315, 0.00585341427136, 0.00593415400159, 0.0059523230271, 0.00595304568405, 0.00594175737341, 0.0059186263843, 0.00591387748421, 0.00596412250667, 0.00610436154186, 0.00643399196802, 0.00690699378334, 0.00753532691265, 0.0083193032524, 0.0092326761507, 0.0102199661071, 0.0112077588638, 0.0121033252134, 0.0130072069277, 0.0139072203158, 0.0146611876077, 0.0153828289821, 0.0160744121867, 0.0166766603911, 0.0172164795582, 0.0177488468134, 0.0182632136293, 0.0186566346647, 0.0190425397251, 0.0194357015034, 0.0197617010843, 0.0200872805015, 0.0204344186764, 0.0207311245281, 0.0209694099278, 0.0212335403014, 0.0215115949242, 0.0217373517193, 0.0219897650935, 0.0222568826317, 0.0224981370852, 0.0227274337641, 0.0229864699088, 0.0232474138528, 0.0234793138343, 0.0237303123809, 0.0239932552179, 0.0241723466339, 0.0243493688858, 0.0245137766694, 0.024635517016, 0.0247200187774, 0.0248183733464, 0.0249209704549, 0.0248972215491, 0.0248901565673, 0.0248923538104, 0.0248399712793, 0.024752651806, 0.0246569935811, 0.0245488673966, 0.0243823754977, 0.0242277833625, 0.024122808071, 0.023967193481, 0.0237996737411, 0.0236418743447, 0.0235121271635, 0.0234115922515, 0.0233334145963, 0.0232919621708, 0.0232425509873, 0.0232313866122, 0.0232695414247, 0.0232177091961, 0.0231520231536, 0.0231285687793, 0.0230275623595, 0.0228080246666, 0.0228104850908, 0.0229009641254, 0.0229397383441, 0.0230192843686, 0.0229851797824, 0.0229261270692, 0.0228131414284, 0.0227337707128, 0.0225904626606, 0.0223573308616, 0.0221935579431, 0.0221160037899, 0.0219219949626, 0.0217030911434, 0.0214877019102, 0.021263340512, 0.021007989729, 0.0207843640743, 0.0206411914581, 0.020503339569, 0.0203641347982, 0.0202278523375, 0.0200458075605, 0.0198785898809, 0.0197536648597, 0.0196417853907, 0.0194847954329, 0.0193236536251, 0.0191825470482, 0.0189384395597, 0.0187318605473, 0.0185380521754, 0.0183197827216, 0.0181161554718, 0.0179453503841, 0.0177620880403, 0.0175395335466, 0.0173615623233, 0.0172140745547, 0.0170036939038, 0.0168258604467, 0.0166715328373, 0.0165087393902, 0.0163600570587, 0.0162155713658, 0.0160108820457, 0.0154940480883, 0.0150403349807, 0.0146808652842, 0.0143833065811, 0.0141067654461, 0.0138384200987, 0.0135452274736, 0.0131943819911, 0.0128539926089, 0.0125149234869, ], dtype=np.float32) # Normalisation stats for interpolated bands 400-1000nm (204 bands) hyper_204_mean = np.array([ 0.0853443115378, 0.082848078348, 0.0810082485533, 0.0797739291089, 0.0789765010856, 0.0784253976231, 0.0780743386258, 0.0778704430139, 0.0777675948717, 0.0777807791937, 0.0779223368812, 0.0781914794129, 0.0785791016654, 0.0790429332303, 0.0795490719257, 0.0800657360188, 0.0805705773184, 0.0810564855142, 0.0816537555639, 0.0822434301913, 0.0828129241917, 0.0851697518554, 0.0882823012806, 0.0913843392534, 0.0931698296219, 0.0921555102572, 0.0911103278881, 0.0900456833142, 0.089806860604, 0.0896448332975, 0.0895035055374, 0.0897823078737, 0.0904041013803, 0.0910202057368, 0.0915858176773, 0.0919696644263, 0.0923231893207, 0.0926073570099, 0.0925576733003, 0.0922939783025, 0.0918693242863, 0.090887923841, 0.0892229863784, 0.0874854547235, 0.0857666488026, 0.0844349513903, 0.0832400678965, 0.0821597277192, 0.0818779649125, 0.0821034052056, 0.0824762599571, 0.0830232482897, 0.0837875118399, 0.0845195548363, 0.0851460471168, 0.0852897622035, 0.0852479044979, 0.0850210136633, 0.0850393938706, 0.0855107406393, 0.086287057134, 0.0878600273907, 0.0923785274538, 0.0976499201371, 0.103583522281, 0.110778056089, 0.118450928734, 0.12599042822, 0.133139807533, 0.139638323898, 0.145918215417, 0.151959042544, 0.156715107509, 0.161186999087, 0.165450053057, 0.169061771918, 0.172201978179, 0.175236667448, 0.178097874942, 0.180423778624, 0.18268025368, 0.184895774742, 0.186974093397, 0.18900679109, 0.191006481496, 0.192898356772, 0.194716583326, 0.196504052264, 0.198253142208, 0.199863596941, 0.201469554295, 0.203073066396, 0.204653778571, 0.206182218116, 0.207735383038, 0.209197545352, 0.210469742978, 0.211741056628, 0.212998041639, 0.213412478345, 0.213682428528, 0.213833922756, 0.214179626296, 0.214759964092, 0.215309688396, 0.21586877854, 0.216597401741, 0.217272827733, 0.217874829296, 0.217713402978, 0.217208848111, 0.216620579664, 0.215926719795, 0.214984752083, 0.214015036535, 0.213108572771, 0.211962954426, 0.210738131305, 0.209458724713, 0.208503667422, 0.20779950603, 0.20709852571, 0.206512866728, 0.206159266439, 0.205829899335, 0.2055763301, 0.205469072134, 0.2053216321, 0.205190373083, 0.205011808434, 0.204647795705, 0.204727073534, 0.204963079918, 0.205117748587, 0.205304463145, 0.205374891607, 0.205465791763, 0.205440357818, 0.205452973033, 0.205285706192, 0.20485040917, 0.204493097663, 0.204284354293, 0.202475507375, 0.200349367799, 0.198158744982, 0.196583004897, 0.195532072024, 0.194514200173, 0.193716582944, 0.193065301919, 0.192380201088, 0.191651233395, 0.190885340272, 0.190137014281, 0.189469187682, 0.188831440083, 0.18806680562, 0.187212345013, 0.186388536669, 0.185829852653, 0.185294547243, 0.18471523209, 0.184031349326, 0.18331896063, 0.182609573035, 0.181863755887, 0.181028442121, 0.180224994041, 0.179381357649, 0.178474862566, 0.177586557723, 0.176697226046, 0.175701928521, 0.174635658965, 0.17350939096, 0.172175970531, 0.168995748617, 0.165735941283, 0.162440184803, 0.160415869749, 0.158957310826, 0.157473121414, 0.155979883857, 0.154430096451, 0.152785336609, 0.151022345404, 0.149900274821, 0.148793261038, 0.147873480083, 0.147190194871, 0.146765078222, 0.146512095475, 0.146301063977, 0.146197658023, 0.146153203837, 0.14612417137, 0.146241604267, 0.146334423564, 0.146512443726, 0.146740819307, 0.146845406329, 0.147084946832, 0.147322538854, ]) hyper_204_std = np.array([ 0.00429063019463, 0.00421556938322, 0.00421326782134, 0.00424388475654, 0.00428636463332, 0.00433271456692, 0.00438247193098, 0.00443299962565, 0.00448713852853, 0.00454708491045, 0.00461157550422, 0.00467946718575, 0.00475101854674, 0.00482716376698, 0.00490837051352, 0.00498946941246, 0.00506774484486, 0.00514309477465, 0.0051925135872, 0.00526480835903, 0.00535829764719, 0.00548445997478, 0.00565378259084, 0.00586489236965, 0.00601440273555, 0.005970148491, 0.00594068114053, 0.00592745749255, 0.00592144445538, 0.00593224420545, 0.00596114517579, 0.00599609616221, 0.00603945896088, 0.00609961389108, 0.00616417685938, 0.00619258440357, 0.00623438905956, 0.00628420787443, 0.00628055551672, 0.00625363016914, 0.00621884501468, 0.0061352346863, 0.00599351875816, 0.00586262239631, 0.00575361016223, 0.00565861485312, 0.00560151576211, 0.00557717177346, 0.00557049875569, 0.00558494534593, 0.00563381330001, 0.00570727512085, 0.0057769231621, 0.00585751955194, 0.00593806224132, 0.00595258611539, 0.00595319388812, 0.00594124665851, 0.00591775837127, 0.0059141902236, 0.00596699872254, 0.00611386079954, 0.00644855789032, 0.00692696261297, 0.00756087973826, 0.00835009031748, 0.00926676960146, 0.0102550364563, 0.0112383325076, 0.0121326939975, 0.0130359966992, 0.0139347779985, 0.0146831125972, 0.0154032868349, 0.0160932585878, 0.0166910923292, 0.0172302476572, 0.0177618612655, 0.0182727020506, 0.0186654340517, 0.0190508131667, 0.0194438153118, 0.0197677705176, 0.0200934680362, 0.0204406290093, 0.0207347602012, 0.0209732759897, 0.0212375094924, 0.0215144401639, 0.0217402755271, 0.0219927754858, 0.0222597296813, 0.0225002660856, 0.0227294251226, 0.022988508771, 0.0232487835734, 0.0234806482899, 0.0237314659303, 0.0239941872785, 0.0241727708999, 0.0243496449606, 0.0245138515016, 0.0246354792911, 0.0247198722799, 0.024818114158, 0.02492103543, 0.0248973159425, 0.0248901019265, 0.0248923472587, 0.0248404151896, 0.0247535516486, 0.0246577808842, 0.0245505008988, 0.0243843574091, 0.0242294347995, 0.0241238050661, 0.0239695315257, 0.0238023597521, 0.023644131753, 0.023513948889, 0.0234134082593, 0.0233344132812, 0.0232932802508, 0.0232432945068, 0.0232309333753, 0.0232681973267, 0.0232199005178, 0.0231528377813, 0.0231294812682, 0.0230352229922, 0.0228088477442, 0.0228078705025, 0.0229006847232, 0.0229372223784, 0.023020913354, 0.0229853138712, 0.0229304161225, 0.0228162353109, 0.0227383430924, 0.0225990666575, 0.0223655419096, 0.022199526403, 0.0221186160642, 0.0219295642769, 0.0217117120569, 0.021496332392, 0.0212742212559, 0.0210190235232, 0.0207939541841, 0.0206486668801, 0.0205093053681, 0.0203708455962, 0.020234317687, 0.0200555277685, 0.0198863727191, 0.0197592893651, 0.0196486847768, 0.0194943242135, 0.0193319664984, 0.0191953265613, 0.0189512298341, 0.0187432517636, 0.0185500554975, 0.0183334317457, 0.0181273213302, 0.0179558892652, 0.0177777718174, 0.0175522171077, 0.0173725481567, 0.0172229668295, 0.0170170874327, 0.0168379915155, 0.016681685115, 0.0165205523857, 0.0163709393398, 0.0162251579454, 0.0160519388455, 0.0155285059359, 0.0150706924364, 0.0147043076919, 0.0144054270517, 0.0141271505124, 0.0138587851936, 0.0135721412487, 0.0132223164765, 0.0128809380303, 0.0125428954494, 0.0122095514875, 0.0118863380612, 0.011623718658, 0.0114240502952, 0.0112818544874, 0.0111888996149, 0.0111072365138, 0.0110557638921, 0.0110133352409, 0.0109718697148, 0.0109764311879, 0.0109805840783, 0.010994444414, 0.0110266496781, 0.0110136046495, 0.0110310144291, 0.0110412626426, ]) # The means and stds also depend on the input mode mode2mean = { 'spixel_51' : hyper_51_mean, 'simage_51' : hyper_51_mean, 'spixel_170' : hyper_170_mean, 'spixel_170_test' : hyper_170_mean, 'simage_170' : hyper_170_mean, 'spixel_204' : hyper_204_mean, 'simage_204' : hyper_204_mean, 'boiko' : None, } mode2std = { 'spixel_51' : hyper_51_std, 'simage_51' : hyper_51_std, 'spixel_170' : hyper_170_std, 'spixel_170_test' : hyper_170_std, 'simage_170' : hyper_170_std, 'spixel_204' : hyper_204_std, 'simage_204' : hyper_204_std, 'boiko' : None, } class LoadImage(torch.nn.Module): """ @class LoadImage is a transform that loads a TIF image and label. This is not supported by MONAI. """ #def __init__(self, keys, mode='rgb', training=True): def __init__(self, keys, mode='rgbimage'): """ @param[in] mode 'rgbpixel': Load an RGB pixel at a time. 'spixel': Load a pixel-wide (1x1x51) multispectral vector. 'rgbimage': Load the RGB version of the input image. 'simage': Load the multispectral version of the input image. """ assert(mode in OdsiDbDataLoader.modes) self.keys = keys self.mode = mode #self._training = training super().__init__() @staticmethod def get_cmfs(cmf_name: str): """ @brief Get colour matching funtions for different standards. @param[in] cmf_name Choose one from [cie1931_2, cie1964_10, cie2012_2, cie2012_10]. From Wikipia: the CIE's color matching functions are the numerical description of the chromatic response of the observer. They can be thought of as the spectral sensitivity curves of three linear light detectors yielding the CIE tristimulus values X, Y and Z. Collectively, these three functions describe the CIE standard observer. """ cmf_full_name = { 'cie_2_1931': 'CIE 1931 2 Degree Standard Observer', 'cie_10_1964': 'CIE 1964 10 Degree Standard Observer', 'cie_2_2012': 'CIE 2012 2 Degree Standard Observer', 'cie_10_2012': 'CIE 2012 10 Degree Standard Observer', } if cmf_name not in cmf_full_name: err_msg = '[ERROR] Wrong CMF name. The available options are: ' err_msg += ', '.join(cmf_full_name.keys()) raise AttributeError(err_msg) standard_obs = colour.colorimetry.MSDS_CMFS_STANDARD_OBSERVER return standard_obs[cmf_full_name[cmf_name]] @staticmethod def gamma_correction(srgb_chan): """ @brief Perform gamma correction on an sRGB image (i.e. the range of the data in the channel is [0., 1.]). @param[in] u One of the three sRGB channels. """ u = srgb_chan[srgb_chan <= 0.0031308] u_r = srgb_chan[srgb_chan > 0.0031308] corrected = np.empty_like(srgb_chan) corrected[srgb_chan <= 0.0031308] = 12.92 * u corrected[srgb_chan > 0.0031308] = 1.055 * np.power(u_r, 0.416) \ - 0.055 return corrected @staticmethod def get_single_wl_im(raw_im_hyper, raw_wl, wl): """ @brief Given an image and the corresponding wavelengths this function returns the image corresponding to a specific channel. @param[in] raw_im_hyper Hyperspectral time, shape (h, w, nbands). @param[in] raw_wl Wavelengths, shape (bands,). @param[in] wl Specific wavelegth of the image you want to extract. @returns an array of dimenstion (h, w). """ idx = (np.abs(raw_wl - wl)).argmin() return raw_im_hyper[:, :, idx].copy() @staticmethod def filter_bands(raw_im_hyper, raw_wl, min_wl, max_wl): """ @brief Given a hyperspectral image this function removes those bands outside the [min_wl, max_wl] range. @param[in] raw_im_hyper Hyperspectral time, shape (h, w, nbands). @param[in] raw_wl Wavelengths, shape (bands,). @param[in] min_wl Minimum wavelength. @param[in] max_wl Maximum wavelength. @returns a tuple (new_im_hyper, new_wl) where new_im_hyper is the new hyperspectral image and new_wl is the new vector of wavelengths of the channels contained in the new_im_hyper. """ # Create new array of wavelengths new_wl = list(filter(lambda x: (x >= min_wl and x <= max_wl), raw_wl)) # Create new hyperspectral image new_im_hyper = np.empty((raw_im_hyper.shape[0], raw_im_hyper.shape[1], len(new_wl)), dtype=raw_im_hyper.dtype) for chan_idx, wl in enumerate(new_wl): new_im_hyper[:, :, chan_idx] = \ OdsiDbDataLoader.LoadImage.get_single_wl_im(raw_im_hyper, raw_wl, wl) return new_im_hyper, np.array(new_wl) @staticmethod def get_additive_correction(cmf_name, wl): """ @brief TODO: although not necessary for ODSI-DB because the data goes until 900nm at least, we should compensate for the data missing the right side of the spectrum too """ # Get sampled colour matching functions cmfs = OdsiDbDataLoader.LoadImage.get_cmfs(cmf_name) xbar_y, ybar_y, zbar_y = colour.utilities.tsplit(cmfs.values) # Get x-coordinates of the missing wavelengths x_corr = [x for x in cmfs.wavelengths if x < min(wl)] epsilon = min(wl) - x_corr[-1] offset = max(x_corr) - min(x_corr) + epsilon x_corr = [x + offset for x in x_corr] # Get additive correction for xbar xbar_y_corr = [y for x, y in zip(cmfs.wavelengths, xbar_y) if x < min(wl)] xbar_y_corr.reverse() # Get additive correction for ybar ybar_y_corr = [y for x, y in zip(cmfs.wavelengths, ybar_y) if x < min(wl)] ybar_y_corr.reverse() # Get additive correction for zbar zbar_y_corr = [y for x, y in zip(cmfs.wavelengths, zbar_y) if x < min(wl)] zbar_y_corr.reverse() # Get interpolated functions for the additive correction f_xbar_corr = scipy.interpolate.PchipInterpolator(x_corr, xbar_y_corr, extrapolate=False) f_ybar_corr = scipy.interpolate.PchipInterpolator(x_corr, ybar_y_corr, extrapolate=False) f_zbar_corr = scipy.interpolate.PchipInterpolator(x_corr, zbar_y_corr, extrapolate=False) return f_xbar_corr, f_ybar_corr, f_zbar_corr @staticmethod def get_corrected_cmf(cmf_name, wl): """ @brief Get colour matching function modified to colour-compensate for missing wavelengths in the input images. @param[in] cmf_name TODO @param[in] wl TODO """ # Get colour matching function (wavelength -> xyz_bar) cmfs = OdsiDbDataLoader.LoadImage.get_cmfs(cmf_name) xbar_y, ybar_y, zbar_y = colour.utilities.tsplit(cmfs.values) f_xbar = scipy.interpolate.PchipInterpolator(cmfs.wavelengths, xbar_y, extrapolate=False) f_ybar = scipy.interpolate.PchipInterpolator(cmfs.wavelengths, ybar_y, extrapolate=False) f_zbar = scipy.interpolate.PchipInterpolator(cmfs.wavelengths, zbar_y, extrapolate=False) # Get additive correction for missing wavelengths (e.g. Nuance EX) f_xbar_corr, f_ybar_corr, f_zbar_corr = \ OdsiDbDataLoader.LoadImage.get_additive_correction(cmf_name, wl) # Get corrected colour matching function xbar_y = np.nan_to_num(f_xbar(wl)) + np.nan_to_num(f_xbar_corr(wl)) ybar_y = np.nan_to_num(f_ybar(wl)) + np.nan_to_num(f_ybar_corr(wl)) zbar_y = np.nan_to_num(f_zbar(wl)) + np.nan_to_num(f_zbar_corr(wl)) f_xbar = scipy.interpolate.PchipInterpolator(wl, xbar_y, extrapolate=False) f_ybar = scipy.interpolate.PchipInterpolator(wl, ybar_y, extrapolate=False) f_zbar = scipy.interpolate.PchipInterpolator(wl, zbar_y, extrapolate=False) return f_xbar, f_ybar, f_zbar @staticmethod def hyper2rgb(raw_im_hyper: np.ndarray, raw_wl: np.ndarray, cmf_name: str = 'cie_2_1931', illuminant_name: str = 'D65', min_wl: float = 300., max_wl: float = 830.): """ @brief Convert hyperspectral image into RGB for debugging purposes. @details The code in this function follows the paper: "Creating RGB images from hyperspectral images using a color matching function" Magnusson et al. 2020. More info here: https://help.commonvisionblox.com/Spectral/ html_english_color-conversion.html @param[in] raw_im_hyper Hyperspectral band image read from ODSI-DB. The intensities are in the range [0.0, 1.0]. Shape (h, w, nbands). @param[in] raw_wl Wavelengths, shape (bands,). In the ODSI-DB, the smallest wavelength is 397.32nm, and the largest is 1003.58nm. @param[in] cmf_name One of [cie_2_1931, cie_10_1964, cie_2_2012, cie_10_2021]. @param[in] illuminant_name See colour.SDS_ILLUMINANTS for more information. @return RGB image in range [0, 255], shape (H, W, 3), dype=np.uint8. """ # Remove bands outside the 300 nm to 830 nm range, they do not # provide any information and this is the range where the D65 illuminant # is defined im_hyper, wl = OdsiDbDataLoader.LoadImage.filter_bands(raw_im_hyper, raw_wl, min_wl, max_wl) h, w, nbands = im_hyper.shape # Get colour matching functions f_xbar, f_ybar, f_zbar = \ OdsiDbDataLoader.LoadImage.get_corrected_cmf(cmf_name, wl) # Get illuminant function (wavelength -> relative power) il = colour.SDS_ILLUMINANTS[illuminant_name] f_illum = scipy.interpolate.PchipInterpolator(il.wavelengths, il.values, extrapolate=True) # Convert hyperspectral image to XYZ (normalisation scales the # values so that y = 1 is the white colour) one_over_n = 1. / np.trapz(f_ybar(wl) * f_illum(wl), wl) x = one_over_n * np.trapz(f_xbar(wl) * im_hyper * f_illum(wl), wl, axis=2) y = one_over_n * np.trapz(f_ybar(wl) * im_hyper * f_illum(wl), wl, axis=2) z = one_over_n * np.trapz(f_zbar(wl) * im_hyper * f_illum(wl), wl, axis=2) # Convert XYZ to RGB (in the range [0, 1]) red = 3.2406255 * x - 1.5372080 * y - 0.4986286 * z; green = -0.9689307 * x + 1.8757561 * y + 0.0415175 * z; blue = 0.0557101 * x - 0.2040211 * y + 1.0569959 * z; # Gamma correction (without gamma correction, images look overly # dark) red = OdsiDbDataLoader.LoadImage.gamma_correction(red) green = OdsiDbDataLoader.LoadImage.gamma_correction(green) blue = OdsiDbDataLoader.LoadImage.gamma_correction(blue) # Reshape images back to 2D im_rgb = np.dstack((red, green, blue)) # Convert sRGB [0., 1.] to RGB in uint8 [0, 255] im_rgb = np.clip((255. * im_rgb).round(), 0, 255).astype(np.uint8) return im_rgb @staticmethod def interp_spectra(im_hyper, wl, new_wl, eps=1., mode='linear'): """ @brief Given a hyperspectral image sampled at certain wavelengths, produce a new hyperspectral image sampled at different wavelengths. @param[in] im_hyper Hyperspectral image of shape (h, w, c). @param[in] wl Wavelengths of each of the channels of 'im_hyper', shape (c,). @param[in] new_wl Wavelengths of each of the channels of the wanted interpolated image to be returned, shape (c,). @param[in] mode Interpolation mode, 'nearest' or 'linear' available. @returns a new hyperspectral image of shape (h, w, c). """ # Make sure that the new wavelengths are not outside the provided ones assert(min(wl) - eps < min(new_wl)) assert(max(wl) + eps > max(new_wl)) # Compute the distance between wavelengths dist = scipy.spatial.distance.cdist(wl.reshape((wl.shape[0], 1)), new_wl.reshape((new_wl.shape[0], 1)), metric='euclidean') # Compute the nearest wavelength for each band nearest = np.argmin(dist, axis=0) # Synthesize new hyperspectral image h, w, _ = im_hyper.shape im_hyper_new = np.empty((h, w, new_wl.shape[0]), dtype=im_hyper.dtype) if mode == 'nearest': # Loop over the new bands for i in range(new_wl.shape[0]): im_hyper_new[:, :, i] = im_hyper[:, :, nearest[i]].copy() elif mode == 'linear': # Loop over the new bands for i in range(new_wl.shape[0]): # Find the bands before and after if wl[nearest[i]] > new_wl[i]: j = nearest[i] - 1 k = nearest[i] else: j = nearest[i] k = nearest[i] + 1 # Clip values at the extremes if j < 0: j = 0 if k == wl.shape[0]: k = wl.shape[0] - 1 # Compute the weights of the bands before and after if j == k: w1 = 0.5 w2 = 0.5 else: d1 = new_wl[i] - wl[j] d2 = wl[k] - new_wl[i] norm = d1 + d2 w1 = d2 / norm w2 = d1 / norm # Add interpolated channel to the image im_hyper_new[:, :, i] = w1 * im_hyper[:, :, j] + w2 * im_hyper[:, :, k] return im_hyper_new @staticmethod def read_image(path, mode, new_wl=np.linspace(400, 1000, 204)): """ @returns an image with shape (c, h, w) containing values [0, 1]. """ if mode not in OdsiDbDataLoader.modes: raise ValueError('[ERROR] The input mode ' + mode \ + ' is unknown.') # Read raw TIFF file im_hyper, wl, _, metadata = torchseg.data_loader.read_stiff(path, silent=True, rgb_only=False) # Select the type of input wanted by the user if mode in ['rgbpixel', 'rgbpixel_test', 'rgbimage']: # Generate RGB reconstruction im_rgb = OdsiDbDataLoader.LoadImage.hyper2rgb(im_hyper, wl) # Convert RGB to float and normalise it to the range [0, 1] im = im_rgb.astype(np.float32) / 255. # Center and scale the image im -= OdsiDbDataLoader.rgb_mean im /= OdsiDbDataLoader.rgb_std # Convert the image to (C, H, W) im = im.transpose((2, 0, 1)).astype(np.float32) elif mode in ['spixel_51', 'simage_51', 'spixel_170', 'spixel_170_test', 'simage_170', 'boiko']: # In the ODSI-DB dataset there are images with 51 bands # (450-950nm, both inclusive) and images with 204 bands # (400-1000nm) # # The boiko mode comes from "Deep Learning for Dental # Image Analysis" by Boiko et al. 2019 # Get the array of interpolated wavelengths according to the # input mode selected by the user new_wl = OdsiDbDataLoader.mode2wl[mode] # Interpolate hyperspectral image to the requested range interp_func = dl.OdsiDbDataLoader.LoadImage.interp_spectra im = interp_func(im_hyper, wl, new_wl) # Center and scale the image im -= OdsiDbDataLoader.mode2mean[mode] im /= OdsiDbDataLoader.mode2std[mode] # Convert the image to (C, H, W) im = im.transpose((2, 0, 1)).astype(np.float32) elif mode in ['spixel_204', 'simage_204']: # Get the array of interpolated wavelengths according to the # input mode selected by the user new_wl = OdsiDbDataLoader.mode2wl[mode] # Get the function that will interpolate the hyperspectral # images into a fixed set of wavelengths interp_func = dl.OdsiDbDataLoader.LoadImage.interp_spectra # If the image comes from the Nuance EX, let's amend it if im_hyper.shape[2] == 51: # Get an array of the wavelengths in [450, 950] inside_indices = np.array([idx for idx, x in enumerate(new_wl.tolist()) \ if x > 450. and x < 950.]) inside_wl = new_wl[inside_indices] # Interpolate hyperspectral image to the requested range inside_im = interp_func(im_hyper, wl, inside_wl) # Extrapolate to 204 bands nbands = 204 im = np.empty((im_hyper.shape[0], im_hyper.shape[1], nbands), dtype=inside_im.dtype) j = 0 for i in range(nbands): if i < np.min(inside_indices): im[:, :, i] = inside_im[:, :, 0] elif i > np.max(inside_indices): im[:, :, i] = inside_im[:, :, -1] else: im[:, :, i] = inside_im[:, :, j] j += 1 # Center and scale the image im -= OdsiDbDataLoader.mode2mean[mode] im /= OdsiDbDataLoader.mode2std[mode] # Convert the image to (C, H, W) im = im.transpose((2, 0, 1)).astype(np.float32) else: # Interpolate hyperspectral image to the requested range im = interp_func(im_hyper, wl, new_wl) # Center and scale the image im -= OdsiDbDataLoader.mode2mean[mode] im /= OdsiDbDataLoader.mode2std[mode] # Convert the image to (C, H, W) im = im.transpose((2, 0, 1)).astype(np.float32) else: raise ValueError('[ERROR] ODSI-DB mode unknown.') return im @staticmethod def read_label(path): raw_tiff = torchseg.data_loader.read_mtiff(path) # Get the shape of the labels (which should be identical to the # height and width of the image) shape = raw_tiff[list(raw_tiff.keys())[0]].shape # Create tensor of labels n_classes = len(OdsiDbDataLoader.OdsiDbDataset.classnames) label = np.zeros((n_classes, *shape), dtype=np.float32) # Build dictionaries to convert quickly from index to class name # and vice versa idx2class = OdsiDbDataLoader.OdsiDbDataset.classnames class2idx = {y: x for x, y in idx2class.items()} # Populate the binary array for each class for k, gt in raw_tiff.items(): if k not in class2idx: raise ValueError('[ERROR] Unknown <' + k + '> label.') # Find the index of the class with name 'k' idx = class2idx[k] # Set the label for the class with name 'k' label[idx] = gt.astype(np.float32) # If all the classes are zero for a pixel, we activate the # background #label[class2idx['Background'], label.sum(axis=0) == 0] = 1 return label def forward(self, data): """ @param[in] data File name to read. """ # Open TIF image im = OdsiDbDataLoader.LoadImage.read_image(data['image'], self.mode) # Open ground truth segmentation label = OdsiDbDataLoader.LoadImage.read_label(data['label']) # Make default output dic = { 'image': im, 'label': label, 'mode' : self.mode, 'path' : data['image'], } ''' # If the user wants pixel-wise multispectral predictions, i.e. what # we call 'spixel' mode, we change the output to contain just a # random pixel from the image #if self.mode == 'spixel' and self.training: if self.mode == 'spixel': # Find all the pixels properly labelled ann_idx = np.array((np.sum(label, axis=0) == 1).nonzero()) # Take only labelled pixels, shape: (npixels, bands) pixels = im[:, ann_idx[0], ann_idx[1]].transpose((1, 0)) pixel_labels = label[:, ann_idx[0], ann_idx[1]].transpose((1, 0)) # Reshape (npixels, bands, 1, 1) pixels = np.expand_dims(pixels, axis=(2, 3)) pixel_labels = np.expand_dims(pixel_labels, axis=(2, 3)) dic = {'image': pixels, 'label': pixel_labels, 'mode': self.mode} # Choose a random pixel #pix = np.random.randint(0, ann_idx.shape[1]) #rand_row = ann_idx[0, pix] #rand_col = ann_idx[1, pix] # Get annotated hyperspectral pixel #pixel = im[:, rand_row, rand_col].reshape( # (im.shape[0], 1, 1)) #pixel_label = label[:, rand_row, rand_col].reshape( # (label.shape[0], 1, 1)) #dic = {'image': pixel, 'label': pixel_label, 'mode': self.mode} # TODO: Make many images of 1x1 pixel instead of picking a # pixel randomly #dic = {'image': [], 'label': []} #for i in range(im.shape[1]): # for j in range(im.shape[2]): # pixel = im[:, i, j].reshape((im.shape[0], 1, 1)) # pixel_label = label[:, i, j].reshape((label.shape[0], 1, 1)) # dic['image'].append(pixel) # dic['label'].append(pixel_label) ''' return dic #@property #def training(self): # return self._training #@training.setter #def training(self, training): # self._training = training class OdsiDbDataset(monai.data.dataset.PersistentDataset): # Class indices and names for the ODSI-DB dataset classnames = { 0 : u'Attached gingiva', 1 : u'Attrition/Erosion', 2 : u'Blood vessel', 3 : u'Calculus', 4 : u'Dentine caries', 5 : u'Enamel', 6 : u'Fibroma', 7 : u'Fluorosis', 8 : u'Gingivitis', 9 : u'Hair', 10: u'Hard palate', 11: u'Inflammation', 12: u'Initial caries', 13: u'Leukoplakia', 14: u'Lip', 15: u'Makeup', 16: u'Malignant lesion', 17: u'Marginal gingiva', 18: u'Metal', 19: u'Microfracture', 20: u'Mole', 21: u'Oral mucosa', 22: u'Out of focus area', 23: u'Pigmentation', 24: u'Plaque', 25: u'Plastic', 26: u'Prosthetics', 27: u'Root', 28: u'Shadow/Noise', 29: u'Skin', 30: u'Soft palate', 31: u'Specular reflection', 32: u'Stain', 33: u'Tongue', 34: u'Ulcer', } classnames_reverse = {v:k for k, v in classnames.items()} #def __init__(self, data_dir, mode='rgb', training='True'): def __init__(self, data_dir, mode='rgbimage'): """ @param[in] mode Input type, see LoadImage constructor for more details. """ # Store parameters assert(mode in OdsiDbDataLoader.modes) self.data_dir = data_dir self.mode = mode # Get list of TIFF files tiffs = [f for f in torchseg.utils.listdir(self.data_dir) \ if '.tif' in f] # Get list of segmentation files segs = [f for f in tiffs if f.endswith('_masks.tif')] # Get list of image files (we ignore images without segmentation) imgs = [f.replace('_masks.tif', '.tif') for f in segs] for im in imgs: assert(os.path.isfile(os.path.join(data_dir, im))) # Build list of dictionaries, where each dictionary # corresponds to an input image data = [{ 'image': os.path.join(data_dir, im), 'label': os.path.join(data_dir, seg)} \ for im, seg in zip(imgs, segs)] # Create data augmentation transform self.image_loader = OdsiDbDataLoader.LoadImage( keys=['image', 'label'], mode=self.mode) #keys=['image', 'label'], mode=self.mode, training=training) trsfm = torchvision.transforms.Compose([ self.image_loader, monai.transforms.ToTensord(keys=['image', 'label']), ]) # Call PersistentDataset constructor super().__init__(data, trsfm) #@property #def training(self): # return self.image_loader.training #@training.setter #def training(self, training): # self.image_loader.training = training """ @class OdsiDbLoader loads data from the ODSI-DB dataset. @details Link to the dataset: https://cs.uef.fi/pub/color/spectra/ODSI-DB """ def __init__(self, data_dir, batch_size, mode='rgbimage', shuffle=True, validation_split=0.0, num_workers=1, ignore_labels=[]): # validation_split=0.0, num_workers=1, training=True): # Store parameters self.data_dir = data_dir self.mode = mode self.ignore_labels = ignore_labels # Create dataset handler self.dataset = OdsiDbDataLoader.OdsiDbDataset(self.data_dir, mode=self.mode) # mode=self.mode, training=training) # Call the constructor of torchseg.base.BaseDataLoader super().__init__(self.dataset, batch_size, shuffle, validation_split, num_workers, collate_fn=self.collate_fn) # num_workers, collate_fn=OdsiDbDataLoader.collate_fn) def _split_sampler(self, split): """ @brief Split the list of inputs between training and validation. @param[in] split Ratio between 0.0 and 1.0 that will correspond to validation. @returns TODO """ if split == 0.0: return None, None if isinstance(split, int): assert(split > 0 and split < self.n_samples) split = float(split) / self.n_samples # Fix seeds for reproducibility np.random.seed(0) random.seed(0) # Make a list of the indices of all the images remaining = np.arange(self.n_samples).tolist() # Get list of labels to check the classes that are present in each image labels = [os.path.join(self.data_dir, im_dic['label']) \ for im_dic in self.dataset.data] # Make a dictionary from each class to the indices of the images that # contain pixels of such class n_classes = len(OdsiDbDataLoader.OdsiDbDataset.classnames) class2img = {c: [] for c in range(n_classes)} for im_idx in remaining: label_path = labels[im_idx] label = OdsiDbDataLoader.LoadImage.read_label(label_path).sum(axis=(1, 2)).astype(np.int64) for c in label.nonzero()[0].tolist(): class2img[c].append(im_idx) # Put an image of each class in the training split, unless we are in # testing mode, in which case self.validation split should be equal to 1.0 train_idx = [] if self.validation_split < 1.0: for c in class2img: random.shuffle(class2img[c]) im_idx = class2img[c][0] if im_idx in remaining: train_idx.append(im_idx) remaining.remove(im_idx) # Split the rest of the images according to the given probability valid_idx = [] while remaining: im_idx = remaining.pop() if np.random.binomial(1, 1 - split): train_idx.append(im_idx) else: valid_idx.append(im_idx) train_idx = np.asarray(train_idx) valid_idx = np.asarray(valid_idx) # Create data samplers for each split train_sampler = torch.utils.data.sampler.SubsetRandomSampler(train_idx) valid_sampler = torch.utils.data.sampler.SubsetRandomSampler(valid_idx) # Turn off shuffle option which is mutually exclusive with sampler self.shuffle = False self.n_samples = len(train_idx) return train_sampler, valid_sampler #@staticmethod def collate_fn(self, batch, max_h=256, max_w=256): """ @brief This function hijacks the batch and crops all the elements of the batch randomly so that they are the same size. The minimum width and height of the images in the batch is used as crop size. @param[in] batch List of dictionaries. Each dictionary represents an input image. @param[in] max_h Maximum height for each image in the batch. If an image has a larger height it will be cropped. This only applies to the 'spixel' mode. @param[in] max_w Maximum width for each image in the batch. If an image is larger it will be cropped. This only applies to the 'spixel' mode. """ # If we are in pixel mode we crop a region of 10 * bs x 10 * bs pixels, # where bs is the batch size #if 'mode' in batch[0] and batch[0]['mode'] == 'spixel': # Reshape the batch, one pixel per batch element #bands = batch[0]['image'].shape[0] #classes = batch[0]['label'].shape[0] #pixels_per_im = min_w * min_h #new_batch = [] #for item in batch: # im = item['image'].reshape((bands, -1)).permute(1, # 0).reshape((pixels_per_im, bands, 1, 1)).unbind() # label = item['label'].reshape((classes, -1)).permute(1, # 0).reshape((pixels_per_im, classes, 1, 1)).unbind() # new_batch += [{'image': pix, 'label': pix_label} \ # for pix, pix_label in zip(im, label)] #batch = new_batch ''' if 'mode' in batch[0] and batch[0]['mode'] == 'spixel': # Find minimum number of pixels min_pixels = np.inf for item in batch: if item['image'].shape[0] < min_pixels: min_pixels = item['image'].shape[0] # Reduce the number of pixels for performance reasons #max_npixels = 128 #min_pixels = min(min_pixels, max_npixels) # Crop number of pixels so that all the elements in the batch have # the same new_batch = [] for item in batch: npixels = item['image'].shape[0] diff = npixels - min_pixels if diff > 0: offset = np.random.randint(0, diff) item['image'] = item['image'][offset:offset + min_pixels, :] item['label'] = item['label'][offset:offset + min_pixels, :] # Reshape the batch, we want a batch of pixels pixels = item['image'].unbind() labels = item['label'].unbind() new_batch += [{'image': pix, 'label': pixl} \ for pix, pixl in zip(pixels, labels)] # Use batch of pixels batch = new_batch else: ''' # Find minimum dimensions min_h = np.inf min_w = np.inf for item in batch: if item['image'].shape[1] < min_h: min_h = item['image'].shape[1] if item['image'].shape[2] < min_w: min_w = item['image'].shape[2] # Crop image for the 'rgbpixel' and 'spixel' cases due to memory and # speed restrictions #if self.training: mode = batch[0].get('mode') if mode in ['rgbpixel', 'spixel_170']: min_h = min(min_h, max_h) min_w = min(min_w, max_w) # Crop image and labels for item in batch: good_crop = False im = None label = None while not good_crop: # Try a crop im, label = OdsiDbDataLoader.random_crop(item['image'], item['label'], min_h, min_w) #item['image'], item['label'] = OdsiDbDataLoader.random_crop(item['image'], item['label'], min_h, min_w) # Does it have annotated pixels? ann_idx = torch.sum(label, dim=0) == 1 num_valid_pixels = ann_idx.sum() if num_valid_pixels > 0: good_crop = True item['image'] = im item['label'] = label return torch.utils.data.dataloader.default_collate(batch) @staticmethod def random_crop(image, label, crop_h, crop_w): """ @param[in] image Torch tensor, shape (C, H, W). @param[in] label Torch tensor, shape (C, H, W). @param[in] crop_h Crop height. @param[in] crop_w Crop width. """ # Get image dimensions h = image.shape[1] w = image.shape[2] assert(h >= crop_h) assert(w >= crop_w) # Compute random crop location if crop_h < h: off_row = np.random.randint(0, h - crop_h, size=1)[0] else: off_row = 0 if crop_w < w: off_col = np.random.randint(0, w - crop_w, size=1)[0] else: off_col = 0 # Crop image and label new_image = image[:, off_row:off_row + crop_h, off_col:off_col + crop_w] new_label = label[:, off_row:off_row + crop_h, off_col:off_col + crop_w] return new_image, new_label
68,696
49.773836
120
py
segodsidb
segodsidb-main/src/config/parser.py
""" @brief Module that contains the ConfigParser class. It is a lightweight module to read the configuration file and the command line parameters and combine them into a single place. @details The code in this module was inspired by: https://github.com/victoresque/pytorch-template @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 1 Jun 2021. """ import os import logging import pathlib import functools import operator import datetime import json # My imports import torchseg.logger import torchseg.utils class ConfigParser: def __init__(self, config, resume=None, modification=None, run_id=None): """ @class ConfigParser parses the configuration JSON file. Handles hyperparameters for training, initializations of modules, checkpoint saving and logging module. @param[in] config Dict containing configurations, hyperparameters for training. Contents of `config.json` file for example. @param[in] resume String, path to the checkpoint being loaded. @param[in] modification Dict keychain:value, specifying position values to be replaced from config dict. @param[in] run_id Unique Identifier for training processes. Used to save checkpoints and training log. Timestamp is being used as default. """ # Load config file and apply modification self._config = _update_config(config, modification) self.resume = resume # Set save_dir where trained model and log will be saved. save_dir = pathlib.Path(self.config['machine']['args']['save_dir']) exper_name = self.config['name'] # If no ID is specified, timestamp is used as default run-id if run_id is None: run_id = datetime.datetime.now().strftime(r'%d%m_%H%M%S') self._save_dir = save_dir / 'models' / exper_name / run_id self._log_dir = save_dir / 'log' / exper_name / run_id # Make directory for saving checkpoints and log. exist_ok = run_id == '' self.save_dir.mkdir(parents=True, exist_ok=exist_ok) self.log_dir.mkdir(parents=True, exist_ok=exist_ok) # Save updated config file to the checkpoint dir torchseg.utils.write_json(self.config, self.save_dir / 'config.json') # Configure logging module torchseg.logger.LoggerSetup(self.log_dir, self.config['logconf']) self.log_levels = { 0: logging.WARNING, 1: logging.INFO, 2: logging.DEBUG } @classmethod def from_args(cls, args, options=''): """ @brief Initialize this class from some CLI arguments. Used in training and testing. """ for opt in options: args.add_argument(*opt.flags, default=None, type=opt.type) if not isinstance(args, tuple): args = args.parse_args() # Select CUDA device (-d option) if args.device is not None: os.environ["CUDA_VISIBLE_DEVICES"] = args.device # Setup environment for resuming train/test (-r option) if args.resume is not None: resume = pathlib.Path(args.resume) cfg_fname = resume.parent / 'config.json' else: msg_no_cfg = "Configuration file need to be specified. \ Add '-c config.json', for example." assert args.conf is not None, msg_no_cfg resume = None cfg_fname = pathlib.Path(args.conf) # Setup config (-c option) config = torchseg.utils.read_json(cfg_fname) if args.conf and resume: # Update new config for fine-tuning config.update(torchseg.utils.read_json(args.conf)) # Setup log config (-l option) if args.logconf: config['logconf'] = args.logconf else: config['logconf'] = '' # Parse custom CLI options into dictionary modification = { opt.target : getattr(args, _get_opt_name(opt.flags)) \ for opt in options } return cls(config, resume, modification) def init_obj(self, name, module, *args, **kwargs): """ @brief Finds a function handle with the name given as 'type' in config, and returns the instance initialized with corresponding arguments given. @details `object = config.init_obj('name', module, a, b=1)` is equivalent to `object = module.name(a, b=1)` """ module_name = self[name]['type'] module_args = dict(self[name]['args']) assert(all([k not in module_args for k in kwargs])) # 'Overwriting kwargs given in config file is not allowed' module_args.update(kwargs) return getattr(module, module_name)(*args, **module_args) def init_ftn(self, name, module, *args, **kwargs): """ @brief Finds a function handle with the name given as 'type' in config, and returns the function with given arguments fixed with functools.partial. @details `function = config.init_ftn('name', module, a, b=1)` is equivalent to `function = lambda *args, **kwargs: module.name(a, *args, b=1, **kwargs)`. """ module_name = self[name]['type'] module_args = dict(self[name]['args']) assert(all([k not in module_args for k in kwargs])) # 'Overwriting kwargs given in config file is not allowed' module_args.update(kwargs) return functools.partial(getattr(module, module_name), *args, **module_args) def __getitem__(self, name): """@brief Access items like ordinary dict.""" return self.config[name] def __str__(self): return json.dumps(self.config, sort_keys=True, indent=4) def get_logger(self, name, verbosity=2): msg_verbosity = 'verbosity option {} is invalid. \ Valid options are {}.'.format(verbosity, self.log_levels.keys()) assert verbosity in self.log_levels, msg_verbosity logger = logging.getLogger(name) logger.setLevel(self.log_levels[verbosity]) return logger # Setting read-only attributes @property def config(self): return self._config @property def save_dir(self): return self._save_dir @property def log_dir(self): return self._log_dir # Helper functions to update config dict with custom CLI options def _update_config(config, modification): if modification is None: return config for k, v in modification.items(): if v is not None: _set_by_path(config, k, v) return config def _get_opt_name(flags): for flg in flags: if flg.startswith('--'): return flg.replace('--', '').replace('-', '_') return flags[0].replace('--', '').replace('-', '_') def _set_by_path(tree, keys, value): """@brief Set a value in a nested object in tree by sequence of keys.""" keys = keys.split(';') _get_by_path(tree, keys[:-1])[keys[-1]] = value def _get_by_path(tree, keys): """@brief Access a nested object in tree by sequence of keys.""" return functools.reduce(operator.getitem, keys, tree)
7,569
35.570048
82
py
segodsidb
segodsidb-main/src/logger/logger.py
""" @brief Module to setup and maintain the logging abilities of the training and validation scripts. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 1 Jun 2021. """ import logging import logging.config import pathlib import torchseg.utils class LoggerSetup(): def __init__(self, save_dir, log_config, default_level=logging.INFO): """ @brief Setup logging configuration. """ log_config = pathlib.Path(log_config) if log_config.is_file(): config = torchseg.utils.read_json(log_config) # Modify logging paths based on run config for _, handler in config['handlers'].items(): if 'filename' in handler: handler['filename'] = str(save_dir / handler['filename']) logging.config.dictConfig(config) else: print("Warning: logging configuration file is not found \ in {}.".format(log_config)) logging.basicConfig(level=default_level)
1,055
33.064516
79
py
segodsidb
segodsidb-main/src/visualization/visualization.py
""" @brief Module with classes for displaying, drawing and plotting information. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 2 Jun 2021. """ import numpy as np import importlib import datetime class TensorboardWriter(): def __init__(self, log_dir, logger, enabled): self.writer = None self.selected_module = "" print('TensorboardWriter.log_dir:', log_dir) print('TensorboardWriter.logger:', logger) print('TensorboardWriter.enabled:', enabled) if enabled: log_dir = str(log_dir) # Retrieve vizualization writer succeeded = False for module in ["torch.utils.tensorboard", "tensorboardX"]: try: self.writer = importlib.import_module(module).SummaryWriter(log_dir) succeeded = True print('TensorboardWriter:', module) break except ImportError: succeeded = False self.selected_module = module if not succeeded: message = "Warning: visualization (Tensorboard) is " \ "configured to use, but currently not installed on " \ "this machine. Please install TensorboardX with " \ "pip install tensorboardx, upgrade PyTorch to " \ "version >= 1.1 to use torch.utils.tensorboard or " \ "turn off the option in the 'config.json' file." logger.warning(message) self.step = 0 self.mode = '' self.tb_writer_ftns = { 'add_scalar', 'add_scalars', 'add_image', 'add_images', 'add_audio', 'add_text', 'add_histogram', 'add_pr_curve', 'add_embedding' } self.tag_mode_exceptions = {'add_histogram', 'add_embedding'} self.timer = datetime.datetime.now() def set_step(self, step, mode='train'): self.mode = mode self.step = step if step == 0: self.timer = datetime.datetime.now() else: duration = datetime.datetime.now() - self.timer self.add_scalar('steps_per_sec', 1 / duration.total_seconds()) self.timer = datetime.datetime.now() def __getattr__(self, name): """ If visualization is configured to use: return add_data() methods of tensorboard with additional information (step, tag) added. Otherwise: return a blank function handle that does nothing """ if name in self.tb_writer_ftns: add_data = getattr(self.writer, name, None) def wrapper(tag, data, *args, **kwargs): if add_data is not None: # add mode(train/valid) tag if name not in self.tag_mode_exceptions: tag = '{}/{}'.format(tag, self.mode) add_data(tag, data, self.step, *args, **kwargs) return wrapper else: # default action for returning methods defined in this class, # set_step() for instance. try: attr = object.__getattr__(name) except AttributeError: raise AttributeError("type object '{}' has no attribute '{}'".format(self.selected_module, name)) return attr
3,424
35.827957
113
py
segodsidb
segodsidb-main/src/base/base_model.py
""" @brief Base class for deep learning models. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 1 Jun 2021. """ import torch import numpy as np import abc class BaseModel(torch.nn.Module): """ @brief Base class for all models. """ def from_file(): # TODO #model = TheModelClass(*args, **kwargs) #model.load_state_dict(torch.load(PATH)) #model.eval() pass @abc.abstractmethod def forward(self, *inputs): """ @brief Forward pass implementation. @returns The model output. """ raise NotImplementedError def __str__(self): """ @brief Print model and number of trainable parameters. """ model_parameters = filter(lambda p: p.requires_grad, self.parameters()) params = sum([np.prod(p.size()) for p in model_parameters]) return super().__str__() + '\nTrainable parameters: {}'.format(params)
977
24.736842
79
py
segodsidb
segodsidb-main/src/base/base_data_loader.py
""" @brief Module with the definition of the base data loader. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 1 Jun 2021. """ import numpy as np import torch class BaseDataLoader(torch.utils.data.DataLoader): def __init__(self, dataset, batch_size, shuffle, validation_split, num_workers, collate_fn=torch.utils.data.dataloader.default_collate): """ @brief Base class for all data loaders. @param[in] dataset TODO. @param[in] batch_size Training and validation batch size. @param[in] shuffle Boolean flag for shuffling data that is currently not used. @param[in] validation_split TODO. @param[in] num_workers Number of PyTorch workers. That is, how many sub-processes to use for data loading. @param[in] collate_fn Function that merges a list of samples to form a mini-batch of tensor[s]. """ # Sanity check: the dataset must have a training flag, cause we use it # here in the BaseDataLoader #assert(hasattr(dataset, 'training')) # Store parameters into attributes self.shuffle = shuffle self.validation_split = validation_split self.batch_idx = 0 self.n_samples = len(dataset) self.sampler, self.valid_sampler = \ self._split_sampler(self.validation_split) self.init_kwargs = { 'dataset': dataset, 'batch_size': batch_size, 'shuffle': self.shuffle, 'collate_fn': collate_fn, 'num_workers': num_workers } super().__init__(sampler=self.sampler, **self.init_kwargs) def _split_sampler(self, split): if split == 0.0: return None, None idx_full = np.arange(self.n_samples) np.random.seed(0) np.random.shuffle(idx_full) if isinstance(split, int): assert(split > 0 and split < self.n_samples) #"validation set size is configured to be larger than entire dataset." len_valid = split else: len_valid = int(self.n_samples * split) valid_idx = idx_full[0:len_valid] train_idx = np.delete(idx_full, np.arange(0, len_valid)) train_sampler = torch.utils.data.sampler.SubsetRandomSampler(train_idx) valid_sampler = torch.utils.data.sampler.SubsetRandomSampler(valid_idx) # Turn off shuffle option which is mutually exclusive with sampler self.shuffle = False self.n_samples = len(train_idx) return train_sampler, valid_sampler def split_validation(self): if self.valid_sampler is None: return None else: return torch.utils.data.DataLoader(sampler=self.valid_sampler, **self.init_kwargs) #@property #def training(self): # return self.dataset.training #@training.setter #def training(self, training): # self.dataset.training = training
3,185
32.893617
82
py
segodsidb
segodsidb-main/src/base/base_machine.py
""" @brief Module with base learning machine. This class is meant to control the training and validation processes for a particular problem or task. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 2 Jun 2021. """ import torch import abc import numpy as np # My imports import torchseg.visualization class BaseMachine: """ @class BaseMachine is an interface class, parent of for all the convolutional learning machines. """ def __init__(self, model, criterion, metric_ftns, optimizer, config): self.config = config self.logger = config.get_logger('machine', config['machine']['args']['verbosity']) self.model = model self.criterion = criterion self.metric_ftns = metric_ftns self.optimizer = optimizer cfg_machine = config['machine']['args'] self.epochs = cfg_machine['epochs'] self.save_period = cfg_machine['save_period'] self.monitor = cfg_machine.get('monitor', 'off') # Configuration to monitor model performance and save best if self.monitor == 'off': self.mnt_mode = 'off' self.mnt_best = 0 else: self.mnt_mode, self.mnt_metric = self.monitor.split() assert self.mnt_mode in ['min', 'max'] self.mnt_best = np.inf if self.mnt_mode == 'min' else -np.inf self.early_stop = cfg_machine.get('early_stop', np.inf) if self.early_stop <= 0: self.early_stop = np.inf self.start_epoch = 1 self.checkpoint_dir = config.save_dir # Setup visualization writer instance self.writer = torchseg.visualization.TensorboardWriter(config.log_dir, self.logger, cfg_machine['tensorboard']) if config.resume is not None: self._resume_checkpoint(config.resume) @abc.abstractmethod def _train_epoch(self, epoch): """ @brief Training logic for an epoch. @param[in] epoch Current epoch number. """ raise NotImplementedError def train(self): """ @brief Full training logic. """ not_improved_count = 0 for epoch in range(self.start_epoch, self.epochs + 1): result = self._train_epoch(epoch) # Save logged informations into log dict log = {'epoch': epoch} log.update(result) # Print logged informations to the screen for key, value in log.items(): self.logger.info(' {:15s}: {}'.format(str(key), value)) # Evaluate model performance according to configured metric, # save best checkpoint as model_best best = False if self.mnt_mode != 'off': try: # Check whether model performance improved or not, # according to specified metric(mnt_metric) improved = (self.mnt_mode == 'min' \ and log[self.mnt_metric] <= self.mnt_best) \ or (self.mnt_mode == 'max' \ and log[self.mnt_metric] >= self.mnt_best) except KeyError: self.logger.warning("""Warning: Metric '{}' is not found. Model performance monitoring is disabled.""".format(self.mnt_metric)) self.mnt_mode = 'off' improved = False if improved: self.mnt_best = log[self.mnt_metric] not_improved_count = 0 best = True else: not_improved_count += 1 if not_improved_count > self.early_stop: self.logger.info('Validation performance did not improve ' \ + 'for ' + str(self.early_stop) \ + ' epochs. Training stops.') break if epoch % self.save_period == 0: self._save_checkpoint(epoch, save_best=best) def _save_checkpoint(self, epoch, save_best=False): """ @brief Method to save checkpoints. @param[in] epoch Current epoch number. @param[in] log Logging information of the epoch. @param[in] save_best If True, rename the saved checkpoint to 'model_best.pth'. """ arch = type(self.model).__name__ state = { 'arch': arch, 'epoch': epoch, 'state_dict': self.model.state_dict(), 'optimizer': self.optimizer.state_dict(), 'monitor_best': self.mnt_best, 'config': self.config } filename = str(self.checkpoint_dir / 'checkpoint-epoch{}.pth'.format( epoch)) torch.save(state, filename) self.logger.info("Saving checkpoint: {} ...".format(filename)) if save_best: best_path = str(self.checkpoint_dir / 'model_best.pth') torch.save(state, best_path) self.logger.info("Saving current best: model_best.pth ...") def _resume_checkpoint(self, resume_path): """ @brief Resume from saved checkpoints. @param[in] resume_path Checkpoint path to be resumed. """ resume_path = str(resume_path) self.logger.info("Loading checkpoint: {} ...".format(resume_path)) checkpoint = torch.load(resume_path) self.start_epoch = checkpoint['epoch'] + 1 self.mnt_best = checkpoint['monitor_best'] # Load architecture params from checkpoint #if checkpoint['config']['arch'] != self.config['arch']: # self.logger.warning('Warning: Architecture configuration given in \ # config file is different from that of \ # checkpoint. This may yield an exception while \ # state_dict is being loaded.') self.model.load_state_dict(checkpoint['state_dict']) # Load optimizer state from checkpoint only when optimizer type is not # changed if checkpoint['config']['optimizer']['type'] != \ self.config['optimizer']['type']: self.logger.warning("""Warning: Optimizer type given in config file is different from that of checkpoint. Optimizer parameters not being resumed.""") else: self.optimizer.load_state_dict(checkpoint['optimizer']) self.logger.info('Checkpoint loaded. Resuming training from epoch ' \ + str(self.start_epoch))
6,879
37.651685
90
py
segodsidb
segodsidb-main/src/utils/utils.py
""" @brief Collection of functions for general purpose use. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 1 Jun 2021. """ import os import re import pathlib import json import collections import torch import pandas as pd ## ----- Functions ----- ## def read_json(fname): """@brief Reads a JSON file and returns a dictionary.""" fname = pathlib.Path(fname) with fname.open('rt') as handle: return json.load(handle, object_hook=collections.OrderedDict) def write_json(content, fname): """@brief Dumps a dictionary into a JSON file.""" fname = pathlib.Path(fname) with fname.open('wt') as handle: json.dump(content, handle, indent=4, sort_keys=False) def setup_gpu_devices(n_gpu_use): """ @brief If available, setup GPU devices. Otherwise, CPU will be used. @param[in] n_gpu_use Number of GPUs you want to use. @returns a tuple (torch.device, list) where the list has the IDs of the GPUs that will be used. """ n_gpu = torch.cuda.device_count() if n_gpu_use > 0 and n_gpu == 0: print("""Warning: There\'s no GPU available on this machine, training will be performed on CPU.""") n_gpu_use = 0 if n_gpu_use > n_gpu: print(f"""Warning: The number of GPU\'s configured to use is {n_gpu_use}, but only {n_gpu} are available on this machine.""") n_gpu_use = n_gpu device = torch.device('cuda:0' if n_gpu_use > 0 else 'cpu') list_ids = list(range(n_gpu_use)) return device, list_ids def natsort(l): """ @brief Natural sort of a list, i.e. using alphabetic order. @param[in] l List to sort. @returns A new list sorted taking into account numbers and not just their ASCII codes. """ convert = lambda text: int(text) if text.isdigit() else text.lower() alphanum_key = lambda key: [ convert(c) for c in re.split('([0-9]+)', str(key)) ] return sorted(l, key = alphanum_key) def listdir(path, hidden=False): """ @brief Lists a directory removing the extension of the files and the hidden files. @param[in] path String containing the path (relative or absolute) to the folder that you want to list. @param[in] hidden Set to True if you want to list the hidden files too. By default, it is set to False. @returns A list of visible files and folders inside the given path. """ files = [] if hidden: files = natsort(os.listdir(path)) else: files = natsort([f for f in os.listdir(path) if not f.startswith('.')]) return files ## ----- Classes ----- ## class MetricTracker: def __init__(self, *keys, writer=None): self.writer = writer self._data = pd.DataFrame(index=keys, columns=['total', 'counts', 'average']) self.reset() def reset(self): for col in self._data.columns: self._data[col].values[:] = 0 def update(self, key, value, n=1): if self.writer is not None: self.writer.add_scalar(key, value) self._data.total[key] += value * n self._data.counts[key] += n self._data.average[key] = self._data.total[key] / self._data.counts[key] def avg(self, key): return self._data.average[key] def result(self): return dict(self._data.average)
3,464
29.9375
82
py
segodsidb
segodsidb-main/src/model/loss.py
""" @brief Module that contains loss functions. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 1 Jun 2021. """ import numpy as np import sys import torch import monai.losses def nll_loss(pred, gt): """ @brief Negative log-likelihood loss (also called multi-class cross entropy). @param[in] pred (N,C) where C = number of classes. @param[in] gt (N) where each value is 0 <= targets[i] <= C - 1. @returns a scalar with the mean loss. """ return torch.nn.functional.nll_loss(pred, gt) def focal_loss(pred, gt): """ @param[in] pred Tensor of probability predictions, shape (B, C, H, W). @param[in] gt Tensor of ground truth, shape (B, C, H, W). """ # FIXME: not working raise NotImplemented() fl = monai.losses.FocalLoss() return fl(pred, gt) def odsi_db_pw_ce_logprob_loss(raw_pred, raw_gt): """ @brief Pixel-wise cross entropy loss. @details Reduces the loss for the batch using the mean. @param[in] raw_pred Tensor of predicted log-probabilities, shape (B, C, H, W). @param[in] raw_gt Tensor of ground truth probabilities, shape (B, C, H, W). @returns a scalar with the mean loss over the images of the batch. """ # Class weights according to the number of pixels per class in ODSI-DB, # a higher number indicates less represented classes class_weights = torch.cuda.FloatTensor([ 0.9682632484506389, 0.9983340617620862, 0.9999394663490677, 0.9995276328275359, 0.9998907852100988, 0.9188844950588891, 0.9999902109476404, 0.9997049747997107, 0.9973278363209698, 0.97715387049886, 0.9607777928694627, 0.9986612604245154, 0.9996366990483456, 0.9999236850100192, 0.9352800022239143, 0.9999932978832073, 0.9999784739894149, 0.986721340312427, 0.9967532371059031, 0.9997563631976796, 0.99995392707397, 0.8692997588756661, 0.860435994817646, 0.9992877927908862, 0.9998345270474653, 0.9957902617780788, 0.9875275587410823, 0.9997695198161124, 0.9879129304566572, 0.6864583697190043, 0.9769124622329116, 0.9681097270561285, 0.9996792888545647, 0.9326207970711948, 0.9999083493782449]) # Flatten predictions and labels bs = raw_pred.shape[0] chan = raw_pred.shape[1] pred_log = torch.reshape(raw_pred, (bs, chan, -1)) gt = torch.reshape(raw_gt, (bs, chan, -1)) #npixels = pred_log.shape[2] # Loop over the batch (each image might have different number of annotated # pixels) ce_loss = torch.empty((bs)) for i in range(bs): # Filter out those pixels without labels ann_idx = torch.sum(gt[i, ...], dim=0) == 1 num_valid_pixels = ann_idx.float().sum() # Convert prediction into nll_loss() preferred input shape (N, C) log_yhat = pred_log[i, :, ann_idx].permute(1, 0) # Convert vector of probabilities into target vector y = torch.argmax(gt[i, :, ann_idx], dim=0) # Compute pixel-wise multi-class (exclusive) cross-entropy loss ce_loss[i] = torch.nn.functional.nll_loss(log_yhat, y, weight=class_weights) # NOTE: The cross entropy for some of the images might be nan, this happens # when an image does not have properly annotated pixels # Reduce the loss for the batch using the mean is_nan = torch.isnan(ce_loss) ce_loss[is_nan] = 0 loss = ce_loss.sum() / (~is_nan).float().sum() return loss def ResNet_18_CAM_DS_ce_loss(tuple_pred, raw_gt): """ @brief Computes the cross entropy loss for all the images in the batch (taking into account the predictions at all resolution levels). @details This loss estimates how good is the network at estimating which classes are present in a given image. We treat this problem as many binary classification problems, each class versus all the others. @param[in] tuple_pred Tuple of six elements, the first one is a tensor containing the predicted class presence log-probabilities, shape (R, B, K, 1, 1). * B is the batch size. * R will be equivalent to the number of resolution levels + 1 (the global prediction). * K is the number of classes. The rest of the elements of the tuple are the pseudo-segmentation predictions at different resolutions. @param[in] raw_gt Tensor of ground truth probabilities, shape (B, C, H, W). """ # Get the classification predictions, we do not use the # pseudo-segmentations to calculate the loss raw_pred = tuple_pred[0] # Get dimensions bs = raw_pred.shape[1] classes = raw_gt.shape[1] rl = raw_pred.shape[0] # Resolution levels + 1 (the global loss) # Loop over the images of the batch ce_loss = torch.cuda.FloatTensor(bs).fill_(0) for i in range(bs): # Build ground truth vector of class-presence probabilities from the # segmentation ground truth that ODSI-DB provides gt = torch.cuda.FloatTensor(classes).fill_(0) for j in range(classes): gt[j] = 1 if 1 in raw_gt[i, j] else 0 # Compute the losses at all resolution levels + global for j in range(rl): # Build vector of class-presence log-probability predictions pred = torch.flatten(raw_pred[j, i]) # Compute loss at this resolution level and add it to the accumulator ce_loss_per_class = - (gt * pred + (1 - gt) * torch.log(1 - torch.exp(pred))) # Mean over classes ce_loss[i] += ce_loss_per_class.mean() # Reduce the loss for the batch using the mean is_nan = torch.isnan(ce_loss) ce_loss[is_nan] = 0 loss = ce_loss.sum() / (~is_nan).float().sum() return loss if __name__ == '__main__': raise RuntimeError('The loss.py module is not is not a script.')
6,327
37.120482
89
py
segodsidb
segodsidb-main/src/model/model.py
""" @brief Collection of deep learning models. @author Luis C. Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 4 Jun 2021. """ import torch import torch.nn.functional as F import torch.utils.model_zoo import numpy as np import monai.networks import torchvision.models import collections import math import unet # My imports import torchseg.base # PyTorch pretrained model URLs model_urls = { "deeplabv3_resnet50_coco": "https://download.pytorch.org/models/deeplabv3_resnet50_coco-cd0a2569.pth", "deeplabv3_resnet101_coco": "https://download.pytorch.org/models/deeplabv3_resnet101_coco-586e9e4e.pth", "deeplabv3_mobilenet_v3_large_coco": "https://download.pytorch.org/models/deeplabv3_mobilenet_v3_large-fc3c493d.pth", } class MnistModel(torchseg.base.BaseModel): def __init__(self, num_classes=10): super().__init__() self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5) self.conv2 = torch.nn.Conv2d(10, 20, kernel_size=5) self.conv2_drop = torch.nn.Dropout2d() self.fc1 = torch.nn.Linear(320, 50) self.fc2 = torch.nn.Linear(50, num_classes) def forward(self, x): x = F.relu(F.max_pool2d(self.conv1(x), 2)) x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2)) x = x.view(-1, 320) x = F.relu(self.fc1(x)) x = F.dropout(x, training=self.training) x = self.fc2(x) return F.log_softmax(x, dim=1) class SimpleUnet(torchseg.base.BaseModel): """ @brief Cut-down implementation of a small U-Net. This network is here for learning and debugging purposes. @details Four resolution levels with 32, 64, 128, and 128 neurons. """ def __init__(self, in_channels=3, out_channels=35): """ @param[in] in_channels Number of input channels, e.g. set it to three for RGB images. @param[in] out_channels Number of classes. """ super().__init__() # Way down self.conv1 = self.down_block(in_channels, 32, 7, 3) self.conv2 = self.down_block(32, 64, 3, 1) self.conv3 = self.down_block(64, 128, 3, 1) # Way up self.upconv3 = self.up_block(128, 64, 3, 1) self.upconv2 = self.up_block(64 * 2, 32, 3, 1) self.upconv1 = self.up_block(32 * 2, out_channels, 3, 1) def down_block(self, in_channels, out_channels, kernel_size, padding): layer = torch.nn.Sequential( # First convolution torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.BatchNorm2d(out_channels), torch.nn.ReLU(), # Second convolution torch.nn.Conv2d(out_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.BatchNorm2d(out_channels), torch.nn.ReLU(), # Max-pooling torch.nn.MaxPool2d(kernel_size=3, stride=2, padding=1) ) return layer def up_block(self, in_channels, out_channels, kernel_size, padding): layer = torch.nn.Sequential( # First convolution torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.BatchNorm2d(out_channels), torch.nn.ReLU(), # Second convolution torch.nn.Conv2d(out_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.BatchNorm2d(out_channels), torch.nn.ReLU(), # Up-convolution torch.nn.ConvTranspose2d(out_channels, out_channels, kernel_size=3, stride=2, padding=1, output_padding=1) ) return layer def crop(self, x, ref): if x.shape != ref.shape: crop_h = ref.shape[2] - x.shape[2] crop_up = int(np.floor(crop_h / 2)) crop_down = crop_h - crop_up crop_w = ref.shape[3] - x.shape[3] crop_left = int(np.floor(crop_w / 2)) crop_right = crop_w - crop_left x = F.pad(x, (crop_left, crop_right, crop_up, crop_down)) return x def forward(self, x): conv1 = self.conv1(x) conv2 = self.conv2(conv1) conv3 = self.conv3(conv2) upconv3 = self.upconv3(conv3) upconv3 = self.crop(upconv3, conv2) upconv2 = self.upconv2(torch.cat([upconv3, conv2], 1)) upconv2 = self.crop(upconv2, conv1) upconv1 = self.upconv1(torch.cat([upconv2, conv1], 1)) upconv1 = self.crop(upconv1, x) return F.log_softmax(upconv1, dim=1) class Unet(torchseg.base.BaseModel): """ @class Wraps the 'pip install unet' U-Net implementation. """ def __init__(self, in_channels: int = 3, out_channels: int = 35) -> None: """ @param[in] in_channels Number of channels of the input image. @param[in] out_channels Number of output classes. """ super().__init__() # Get segmentation model self.backbone = unet.UNet(in_channels=in_channels, out_classes=out_channels, dimensions=2, num_encoding_blocks=5, out_channels_first_layer=64, normalization='batch', padding=1, preactivation=True, residual=False) def forward(self, x: torch.Tensor) -> dict[str, torch.Tensor]: # Get input shape bs, channels, old_height, old_width = x.shape # Compute form factor ff = float(old_width) / old_height # Resize width to 512 width = 512 height = int(round(width / ff)) x = torchvision.transforms.functional.resize(x, [height, width]) # Find how much we have to pad width and height powers = [2**i for i in range(9, 13)] height_padding = pow(2, math.ceil(np.log2(height))) - height width_padding = pow(2, math.ceil(np.log2(width))) - width # Add padding to the image x = F.pad(x, (0, width_padding, 0, height_padding), 'constant', 0) # Perform forward pass with unet.UNet x = self.backbone(x) # Remove padding and convet the image to the original size x = F.pad(x, (0, -width_padding, 0, -height_padding), 'constant', 0) # Resize back to original size x = torchvision.transforms.functional.resize(x, [old_height, old_width]) # Return log-probabilities return F.log_softmax(x, dim=1) class Unet_obsolete(torchseg.base.BaseModel): """ @brief Classical U-Net implementation. """ def __init__(self, in_channels=3, out_channels=35): """ @param[in] in_channels Number of input channels, e.g. set it to three for RGB images. @param[in] out_channels Number of classes. """ super().__init__() # Way down self.conv1 = self.down_block(in_channels, 64, 7, 3) self.conv2 = self.down_block(64, 128, 3, 1) self.conv3 = self.down_block(128, 256, 3, 1) self.conv4 = self.down_block(256, 512, 3, 1) self.conv5 = self.down_block(512, 1024, 3, 1) # Way up self.upconv5 = self.up_block(1024, 512, 3, 1) self.upconv4 = self.up_block(512 * 2, 256, 3, 1) self.upconv3 = self.up_block(256 * 2, 128, 3, 1) self.upconv2 = self.up_block(128 * 2, 64, 3, 1) self.upconv1 = self.up_block(64 * 2, out_channels, 3, 1) def down_block(self, in_channels, out_channels, kernel_size, padding): layer = torch.nn.Sequential( # First convolution torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.ReLU(), torch.nn.BatchNorm2d(out_channels), # Second convolution torch.nn.Conv2d(out_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.ReLU(), torch.nn.BatchNorm2d(out_channels), # Max-pooling torch.nn.MaxPool2d(kernel_size=3, stride=2, padding=1) ) return layer def up_block(self, in_channels, out_channels, kernel_size, padding): layer = torch.nn.Sequential( # First convolution torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.ReLU(), torch.nn.BatchNorm2d(out_channels), # Second convolution torch.nn.Conv2d(out_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.ReLU(), torch.nn.BatchNorm2d(out_channels), # Up-convolution torch.nn.ConvTranspose2d(out_channels, out_channels, kernel_size=3, stride=2, padding=1, output_padding=1) ) return layer def crop(self, x, ref): if x.shape != ref.shape: crop_h = ref.shape[2] - x.shape[2] crop_up = int(np.floor(crop_h / 2)) crop_down = crop_h - crop_up crop_w = ref.shape[3] - x.shape[3] crop_left = int(np.floor(crop_w / 2)) crop_right = crop_w - crop_left x = F.pad(x, (crop_left, crop_right, crop_up, crop_down)) return x def forward(self, x): conv1 = self.conv1(x) conv2 = self.conv2(conv1) conv3 = self.conv3(conv2) conv4 = self.conv4(conv3) conv5 = self.conv5(conv4) upconv5 = self.upconv5(conv5) upconv5 = self.crop(upconv5, conv4) upconv4 = self.upconv4(torch.cat([upconv5, conv4], 1)) upconv4 = self.crop(upconv4, conv3) upconv3 = self.upconv3(torch.cat([upconv4, conv3], 1)) upconv3 = self.crop(upconv3, conv2) upconv2 = self.upconv2(torch.cat([upconv3, conv2], 1)) upconv2 = self.crop(upconv2, conv1) upconv1 = self.upconv1(torch.cat([upconv2, conv1], 1)) upconv1 = self.crop(upconv1, x) return F.log_softmax(upconv1, dim=1) class VectorUnet(torchseg.base.BaseModel): """ @brief Version of the SimpleUnet but without downsampling/upsampling. @details Four levels with 32, 64, 128, and 128 neurons. """ def __init__(self, in_channels=3, out_channels=35): """ @param[in] in_channels Number of input channels, e.g. set it to three for RGB images. @param[in] out_channels Number of classes. """ super().__init__() # Fake way down self.conv1 = self.conv_block(in_channels, 64, 1, 0) self.conv2 = self.conv_block(64, 128, 1, 0) self.conv3 = self.conv_block(128, 256, 1, 0) self.conv4 = self.conv_block(256, 512, 1, 0) self.conv5 = self.conv_block(512, 1024, 1, 0) # Fake way up self.upconv5 = self.conv_block(1024, 512, 1, 0) self.upconv4 = self.conv_block(512 * 2, 256, 1, 0) self.upconv3 = self.conv_block(256 * 2, 128, 1, 0) self.upconv2 = self.conv_block(128 * 2, 64, 1, 0) self.upconv1 = self.conv_block(64 * 2, out_channels, 1, 0, scoring=True) def conv_block(self, in_channels, out_channels, kernel_size, padding, scoring=False): # Create list of operations modules = [ # First convolution torch.nn.BatchNorm2d(in_channels), torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.ReLU(), # Second convolution torch.nn.BatchNorm2d(out_channels), torch.nn.Conv2d(out_channels, out_channels, kernel_size, stride=1, padding=padding), torch.nn.ReLU(), ] # If we are not before softmax, we use BN+ReLU as last operation if scoring: modules.append(torch.nn.BatchNorm2d(out_channels)) modules.append(torch.nn.Conv2d(out_channels, out_channels, kernel_size, stride=1, padding=padding)) # modules.append(torch.nn.ReLU()) layer = torch.nn.Sequential(*modules) return layer def forward(self, x): conv1 = self.conv1(x) conv2 = self.conv2(conv1) conv3 = self.conv3(conv2) conv4 = self.conv4(conv3) conv5 = self.conv5(conv4) upconv5 = self.upconv5(conv5) upconv4 = self.upconv4(torch.cat([upconv5, conv4], 1)) upconv3 = self.upconv3(torch.cat([upconv4, conv3], 1)) upconv2 = self.upconv2(torch.cat([upconv3, conv2], 1)) upconv1 = self.upconv1(torch.cat([upconv2, conv1], 1)) return F.log_softmax(upconv1, dim=1) class DeepLabV3(torchseg.base.BaseModel): class ASPPConv(torch.nn.Sequential): def __init__(self, in_channels, out_channels, dilation): modules = [ torch.nn.Conv2d(in_channels, out_channels, 3, padding=dilation, dilation=dilation, bias=False), # FIXME: commented out because it does not support bs=1 #torch.nn.BatchNorm2d(out_channels), torch.nn.ReLU() ] super(DeepLabV3.ASPPConv, self).__init__(*modules) class ASPPPooling(torch.nn.Sequential): def __init__(self, in_channels, out_channels): modules = [ torch.nn.AdaptiveAvgPool2d(1), torch.nn.Conv2d(in_channels, out_channels, 1, bias=False), # FIXME: commented out because it does not support bs=1 #torch.nn.BatchNorm2d(out_channels), torch.nn.ReLU() ] super(DeepLabV3.ASPPPooling, self).__init__(*modules) def forward(self, x): size = x.shape[-2:] for mod in self: x = mod(x) return F.interpolate(x, size=size, mode='bilinear', align_corners=False) class ASPP(torch.nn.Module): def __init__(self, in_channels, atrous_rates, out_channels=256): super(DeepLabV3.ASPP, self).__init__() modules = [] modules.append(torch.nn.Sequential( torch.nn.Conv2d(in_channels, out_channels, 1, bias=False), # FIXME: commented out because it does not support bs=1 #torch.nn.BatchNorm2d(out_channels), torch.nn.ReLU())) rates = tuple(atrous_rates) for rate in rates: modules.append(DeepLabV3.ASPPConv(in_channels, out_channels, rate)) modules.append(DeepLabV3.ASPPPooling(in_channels, out_channels)) self.convs = torch.nn.ModuleList(modules) self.project = torch.nn.Sequential( torch.nn.Conv2d(len(self.convs) * out_channels, out_channels, 1, bias=False), # FIXME: commented out because it does not support bs=1 #torch.nn.BatchNorm2d(out_channels), torch.nn.ReLU(), torch.nn.Dropout(0.5)) def forward(self, x): res = [] for conv in self.convs: res.append(conv(x)) res = torch.cat(res, dim=1) return self.project(res) class DeepLabHead(torch.nn.Sequential): def __init__(self, in_channels, num_classes): super(DeepLabV3.DeepLabHead, self).__init__( DeepLabV3.ASPP(in_channels, [12, 24, 36]), torch.nn.Conv2d(256, 256, 3, padding=1, bias=False), #torch.nn.BatchNorm2d(256), torch.nn.ReLU(), torch.nn.Conv2d(256, num_classes, 1) ) def __init__(self, in_channels: int = 3, out_channels: int = 35, pretrained: bool = False) -> None: """ @param[in] in_channels Number of channels of the input image. @param[in] out_channels Number of output classes. """ super().__init__() # Get ResNet-101 encoder self.backbone = torchvision.models.resnet.resnet101( pretrained=True, replace_stride_with_dilation=[False, True, True]) # Freeze ResNet-101 encoder layers for param in self.backbone.parameters(): param.requires_grad = False # Get segmentation model return_layers = {'layer4': 'out'} self.backbone = torchvision.models._utils.IntermediateLayerGetter(self.backbone, return_layers=return_layers) self.decoder = DeepLabV3.DeepLabHead(2048, out_channels) # TODO: Load pretrained weights from a DeepLabV3 trained on MS-COCO (train2017) def forward(self, x: torch.Tensor) -> dict[str, torch.Tensor]: # Encoder: get latent space vector input_shape = x.shape[-2:] latent_tensor_dict = self.backbone(x) # Decoder: get per-class output scores raw_decoder_output = self.decoder(latent_tensor_dict['out']) # Final touch: interpolate segmentation maps to the size of the # original image interp_decoder_output = F.interpolate(raw_decoder_output, size=input_shape, mode='bilinear', align_corners=False) return F.log_softmax(interp_decoder_output, dim=1) class ResNet_18_CAM_DS(torchseg.base.BaseModel): """ @brief Classification network that produces segmentation predictions as an intermediary step. @details Implementation of the architecture presented in "Intrapapillary capillary loop classification in magnification endoscopy: open dataset and baseline methodology" by Luis C. Garcia Peraza Herrera et al., 2020. """ def __init__(self, in_channels: int = 3, out_channels: int = 35, pretrained: bool = False) -> None: """ @param[in] in_channels Number of input channels, e.g. set it to three for RGB images. @param[in] out_channels Number of classes. """ super().__init__() # Store parameters self.in_channels = in_channels self.out_channels = out_channels self.pretrained = pretrained # Get ResNet-18 encoder and modify the number of input channels self.resnet = torchvision.models.resnet.resnet18(pretrained=self.pretrained) self.resnet.conv1 = torch.nn.Conv2d(self.in_channels, 64, kernel_size=7, stride=2, padding=3, bias=False) # Create "segmentation" layers self.seg1 = torch.nn.Conv2d(64, self.out_channels, kernel_size=1) self.seg2 = torch.nn.Conv2d(64, self.out_channels, kernel_size=1) self.seg3 = torch.nn.Conv2d(128, self.out_channels, kernel_size=1) self.seg4 = torch.nn.Conv2d(256, self.out_channels, kernel_size=1) self.seg5 = torch.nn.Conv2d(512, self.out_channels, kernel_size=1) # Create GAP layers self.gap1 = torch.nn.AdaptiveAvgPool2d(output_size=(1, 1)) self.gap2 = torch.nn.AdaptiveAvgPool2d(output_size=(1, 1)) self.gap3 = torch.nn.AdaptiveAvgPool2d(output_size=(1, 1)) self.gap4 = torch.nn.AdaptiveAvgPool2d(output_size=(1, 1)) self.gap5 = torch.nn.AdaptiveAvgPool2d(output_size=(1, 1)) # Add the extra stuff to make ResNet-18-CAM-DS ilg = torchvision.models._utils.IntermediateLayerGetter self.conv1_tensor = ilg(self.resnet, return_layers={'bn1': 'out'}) self.conv2_tensor = ilg(self.resnet, return_layers={'layer1': 'out'}) self.conv3_tensor = ilg(self.resnet, return_layers={'layer2': 'out'}) self.conv4_tensor = ilg(self.resnet, return_layers={'layer3': 'out'}) self.conv5_tensor = ilg(self.resnet, return_layers={'layer4': 'out'}) # Create layer to convert scores into fake probabilities self.log_sigmoid = torch.nn.LogSigmoid() def forward(self, x: torch.Tensor): """ @returns a dictionary of log-probability predictions from the side branches and the global branch. The key 'logprob6_output' represents the global prediction. The tensors should be one-dimensional, with each element indicating the probability of that class being in the image. """ # Get tensors at every resolution level conv1_output = self.conv1_tensor(x)['out'] conv2_output = self.conv2_tensor(x)['out'] conv3_output = self.conv3_tensor(x)['out'] conv4_output = self.conv4_tensor(x)['out'] conv5_output = self.conv5_tensor(x)['out'] # Produce a feature map per class at every resolution level seg1_output = self.seg1(conv1_output) seg2_output = self.seg2(conv2_output) seg3_output = self.seg3(conv3_output) seg4_output = self.seg4(conv4_output) seg5_output = self.seg5(conv5_output) # Compute class scores (with GAP) at every resolution level # (and globally too) score1_output = self.gap1(seg1_output) score2_output = self.gap2(seg2_output) score3_output = self.gap3(seg3_output) score4_output = self.gap4(seg4_output) score5_output = self.gap5(seg5_output) score6_output = score1_output + score2_output + score3_output \ + score4_output + score5_output # Compute sigmoid function to estimate the side "probabilities" logprob1_output = self.log_sigmoid(score1_output) logprob2_output = self.log_sigmoid(score2_output) logprob3_output = self.log_sigmoid(score3_output) logprob4_output = self.log_sigmoid(score4_output) logprob5_output = self.log_sigmoid(score5_output) # Compute the global "probabilities" logprob6_output = self.log_sigmoid(score6_output) # Stack the class-presence log-probability predictions at every # resolution level class_presence = torch.stack([logprob1_output, logprob2_output, logprob3_output, logprob4_output, logprob5_output, logprob6_output]) return class_presence, seg1_output, seg2_output, seg3_output, seg4_output, seg5_output if __name__ == '__main__': raise RuntimeError('The model.py module is not is not a script.')
23,154
38.446337
111
py
segodsidb
segodsidb-main/src/model/metric.py
""" @brief Module of evaluation measures or metrics. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 2 Jun 2021. """ import torch import numpy as np import monai.metrics import torchvision # My imports import torchseg.data_loader as dl def accuracy(pred, gt): """ @brief Given a set of data points from repeated measurements of the same quantity, the set can be said to be accurate if their average is close to the true value of the quantity being measured. @param[in] pred TODO. @param[in] gt TODO. @returns TODO """ with torch.no_grad(): pred = torch.argmax(pred, dim=1) assert pred.shape[0] == len(gt) correct = 0 correct += torch.sum(pred == gt).item() return correct / len(gt) def top_k_acc(pred, gt, k=3): """ @brief Top-k accuracy means that the correct class gets to be in the top-k predicted probabilities for it to count as correct. @param[in] pred TODO. @param[in] gt TODO. @returns TODO """ with torch.no_grad(): pred = torch.topk(pred, k, dim=1)[1] assert pred.shape[0] == len(gt) correct = 0 for i in range(k): correct += torch.sum(pred[:, i] == gt).item() return correct / len(gt) def iou(pred, gt, k=1): """ @brief Computes the soft intersection over union for binary segmentation. @param[in] pred TODO. @param[in] gt TODO. @param[in] k Index of the positive class (starting from zero). @returns TODO. """ with torch.no_grad(): # TODO raise NotImplemented() return 0 def mean_iou(pred, gt, eps=1e-6, thresh=0.5): """ @brief Computes the soft mean (over classes) intersection over union for all the images of the batch. Then all the mIoU per image are averaged over all the images in the batch. @param[in] pred Tensor of predicted log-probabilities, shape (B, C, H, W). @param[in] gt Tensor of ground truth, shape (B, C, H, W). @returns a floating point with the loss value. """ with torch.no_grad(): # Flatten predictions and labels bs = pred.shape[0] chan = pred.shape[1] flat_pred = torch.reshape(torch.exp(pred), (bs, chan, -1)) flat_gt = torch.reshape(gt, (bs, chan, -1)) # Binarise prediction bin_pred = (flat_pred > thresh).float() argmax_pred = torch.argmax(flat_pred, dim=1) one_hot_argmax_pred = torch.zeros_like(bin_pred) for i in range(bs): one_hot_argmax_pred[i] = torch.nn.functional.one_hot(argmax_pred[i], num_classes=chan).transpose(0, 1) bin_pred *= one_hot_argmax_pred # Intersection inter = torch.sum(bin_pred * flat_gt, dim=2) # Union union = torch.sum(bin_pred, dim=2) + torch.sum(flat_gt, dim=2) - inter # Mean IoU over classes miou = torch.mean((inter + eps) / (union + eps), dim=1) # Mean over the images in the batch batch_miou = torch.mean(miou).item() return batch_miou def odsi_db_mean_iou(pred, gt, eps=1e-6, thresh=0.5): """ @brief Computes the soft mean (over classes) intersection over union for all the images of the batch. Then all the mIoU per image are averaged over all the images in the batch. @param[in] pred Tensor of predicted log-probabilities, shape (B, C, H, W). @param[in] gt Tensor of ground truth, shape (B, C, H, W). @returns a floating point with the loss value. """ with torch.no_grad(): # Flatten predictions and labels bs = pred.shape[0] chan = pred.shape[1] flat_pred = torch.reshape(torch.exp(pred), (bs, chan, -1)) flat_gt = torch.reshape(gt, (bs, chan, -1)) # Binarise prediction flat_bin_pred = (flat_pred > thresh).float() argmax_pred = torch.argmax(flat_pred, dim=1) one_hot_argmax_pred = torch.zeros_like(flat_bin_pred) for i in range(bs): one_hot_argmax_pred[i] = torch.nn.functional.one_hot(argmax_pred[i], num_classes=chan).transpose(0, 1) flat_bin_pred *= one_hot_argmax_pred # NOTE: Every image in the batch can have a different number of annotated # pixels miou = torch.empty((bs)) for i in range(bs): # Filter predictions without annotation, i.e. when all the ground # truth classes are zero ann_idx = torch.sum(flat_gt[i, ...], dim=0) == 1 y_true = flat_gt[i, :, ann_idx] y_pred = flat_bin_pred[i, :, ann_idx] # Intersection inter = torch.sum(y_pred * y_true, dim=1) # Union union = torch.sum(y_pred, dim=1) + torch.sum(y_true, dim=1) - inter # Mean IoU over classes miou[i] = torch.mean((inter + eps) / (union + eps), dim=0) # Mean over the images in the batch batch_miou = torch.mean(miou).item() return batch_miou def odsi_db_monai_metric(pred, gt, metric_name, ignore_labels=[], thresh=0.5): """ @brief Computes the balanced accuracy averaged over all classes. @details True negatives: only the labelled pixels of the batch are counted. @param[in] pred Tensor of predicted log-probabilities, shape (B, C, H, W). @param[in] gt Tensor of ground truth, shape (B, C, H, W). @param[in] ignore_labels List of classes (class names) that should be ignored when averaging the metric over classes. @param[in] thresh Theshold used to binarise predictions. @returns a scalar with the metric for the whole batch. """ retval = None with torch.no_grad(): # Flatten predictions and labels, and convert log-probabilities into # probabilities bs = pred.shape[0] chan = pred.shape[1] height = pred.shape[2] width = pred.shape[3] flat_pred = torch.reshape(torch.exp(pred), (bs, chan, -1)) flat_gt = torch.reshape(gt, (bs, chan, -1)) # Binarise prediction flat_bin_pred = (flat_pred > thresh).float() # Shapes of bin_pred and flat_gt now are (B, C, H*W) assert(flat_bin_pred.shape == (bs, chan, height * width)) assert(flat_gt.shape == (bs, chan, height * width)) # NOTE: Every image in the batch can have a different number of # annotated pixels conf_mat = torch.zeros((1, chan, 4)).cuda() for i in range(bs): # Get tensor of booleans of valid annotated pixels (i.e. they # have been assigned to only one class) valid_idx = torch.sum(flat_gt[i, ...], dim=0) == 1 # TODO: Add a case for when valid_idx is empty, i.e. what # happens if the image does not have annotated pixels? # Get prediction and ground truth tensors, the shapes of the # tensors are (1, nclasses, npixels) y_true = flat_gt[i, :, valid_idx][None, :, :] y_pred = flat_bin_pred[i, :, valid_idx][None, :, :] # Accumulate the confusion matrix over the images in the batch conf_mat += monai.metrics.get_confusion_matrix(y_pred, y_true) if metric_name == 'confusion matrix': retval = conf_mat else: # Compute the metric with MONAI met = monai.metrics.compute_confusion_matrix_metric(metric_name, conf_mat)[0] # We might want to ignore some classes for whatever reason idx2class = dl.OdsiDbDataLoader.OdsiDbDataset.classnames class2idx = {y: x for x, y in idx2class.items()} if ignore_labels: ignore_labels = [class2idx[x] for x in ignore_labels] for idx in ignore_labels: met[idx] = float('nan') # Mean metric over classes (ignoring classes whose metric is nan) is_nan = torch.isnan(met) met[is_nan] = 0 mean_over_classes = (met.sum() / (~is_nan).float().sum()).item() retval = mean_over_classes return retval def odsi_db_accuracy(pred, gt, ignore_labels=[]): return odsi_db_monai_metric(pred, gt, 'accuracy', ignore_labels=ignore_labels) def odsi_db_balanced_accuracy(pred, gt, ignore_labels=[]): sens = odsi_db_monai_metric(pred, gt, 'sensitivity', ignore_labels=ignore_labels) spec = odsi_db_monai_metric(pred, gt, 'specificity', ignore_labels=ignore_labels) return .5 * sens + .5 * spec #return odsi_db_monai_metric(pred, gt, 'balanced accuracy', # ignore_labels=ignore_labels) def odsi_db_sensitivity(pred, gt, ignore_labels=[]): return odsi_db_monai_metric(pred, gt, 'sensitivity', ignore_labels=ignore_labels) def odsi_db_specificity(pred, gt, ignore_labels=[]): return odsi_db_monai_metric(pred, gt, 'specificity', ignore_labels=ignore_labels) def odsi_db_precision(pred, gt, ignore_labels=[]): return odsi_db_monai_metric(pred, gt, 'precision', ignore_labels=ignore_labels) def odsi_db_f1_score(pred, gt, ignore_labels=[]): return odsi_db_monai_metric(pred, gt, 'f1 score', ignore_labels=ignore_labels) def odsi_db_conf_mat(pred, gt, ignore_labels=[]): return odsi_db_monai_metric(pred, gt, 'confusion matrix', ignore_labels=ignore_labels) def odsi_db_ResNet_18_CAM_DS_accuracy(tuple_pred, raw_gt, ignore_labels=[], thresh=0.5, pred_index=5): """ @brief Class-presence accuracy metric for images. @details TP, TN, FP, FN explanation: * True positive: the network estimates that the image contains pixels of class K, and the image contains them. * True negative: the network estimates that the image does not contain pixels of class K, and the image does not contain them. * False positive: the network estimates that the image contains pixels of class K, and the image does not contain them. * False negative: the network estimates that the image does not contain pixels of class K, but the image contains them. Given an image, we have a vector of predictions, and a ground truth vector. Each element of the vector represents a class, a 1 means that the class is present in the image (i.e. there is at least one pixel belonging to this class), and a 0 means that the class is not present in the image. Based on these two vectors (prediction and ground truth) we can build a confusion matrix, and then compute an accuracy. This accuracy tells you how good is the system in predicting the classes that are present (and not present) in the image. @param[in] tuple_pred Tuple of six elements, the first one is a tensor containing the predicted class presence log-probabilities, shape (R, B, K, 1, 1). * B is the batch size. * R will be equivalent to the number of resolution levels + 1 (the global prediction). * K is the number of classes. The rest of the elements of the tuple are the pseudo-segmentation predictions at different resolutions. @param[in] raw_gt Tensor of ground truth probabilities, shape (B, C, H, W). @param[in] thresh Threshold to convert soft probability predictions into binary predictions. @param[in] pred_index tuple_pred[0] contains a tensor of dimensions (R, B, K, 1, 1), where each element of R contains a class-presence prediction at a different resolution. The last item of R contains the global prediction. Using this parameter you can choose to use the class-presence prediction at any specific resolution level (or the global one). By default, pred_index=5 because that is where the global prediction is stored by the forward pass of the model. @returns the average accuracy over the images in the batch. """ retval = None with torch.no_grad(): # Get the classification predictions, we will not use the # pseudo-segmentations to calculate the this metric raw_pred = tuple_pred[0] # Get dimensions bs = raw_pred.shape[1] classes = raw_gt.shape[1] rl = raw_pred.shape[0] # Resolution levels + 1 (the global loss) # Loop over the batch acc = torch.cuda.FloatTensor(bs) for i in range(bs): # Build ground truth vector of class-presence probabilities from # the segmentation ground truth that ODSI-DB provides gt = torch.cuda.FloatTensor(classes).fill_(0) for j in range(classes): gt[j] = 1 if 1 in raw_gt[i, j] else 0 # Get the class-presence prediction vector, the five is to get the # global prediction (computed as the sum of the side predictions) soft_pred = torch.exp(torch.flatten(raw_pred[pred_index, i])) pred = (soft_pred > thresh).float() # Compute confusion matrix tp = torch.sum(torch.logical_and(pred == 1, gt == 1)) tn = torch.sum(torch.logical_and(pred == 0, gt == 0)) fp = torch.sum(torch.logical_and(pred == 1, gt == 0)) fn = torch.sum(torch.logical_and(pred == 0, gt == 1)) # Compute "accuracy" of class-presence detection for this # batch image acc[i] = (tp + tn) / (tp + tn + fp + fn) # Compute the mean accuracy over the images of the batch retval = acc.mean().item() return retval def convert_tuple_pred_to_segmap(tuple_pred, raw_gt, pred_index): """ @brief Convert segmentation feature maps into log-probability maps. """ # Get "segmentation" feature maps segmap = tuple_pred[1 + pred_index] bs, classes, small_h, small_w = segmap.shape _, _, big_h, big_w = raw_gt.shape # Resize the "segmentation" feature maps to the original image size segmap = torchvision.transforms.Resize((big_h, big_w), interpolation=torchvision.transforms.InterpolationMode.BICUBIC)(segmap) # Apply sigmoid to convert the "segmentation" feature maps into # "probability" maps segmap = torch.nn.functional.logsigmoid(segmap) return segmap def odsi_db_ResNet_18_CAM_DS_pw_accuracy(tuple_pred, raw_gt, ignore_labels=[], pred_index=0): """ @brief Evaluate the pixel-wise accuracy of the class-presence classification model. @param[in] tuple_pred See the comments in odsi_db_ResNet_18_CAM_DS_accuracy(). @param[in] raw_gt See the comments in odsi_db_ResNet_18_CAM_DS_accuracy(). @param[in] pred_index Indicates the resolution level where to take the segmentation from, starting from zero with the one of largest resolution. @details The model and the method to obtain the segmentation is as explained in: Garcia-Peraza Herrera et al. 2020 "Intrapapillary capillary loop classification in magnification endoscopy: open dataset and baseline methodology". """ with torch.no_grad(): segmap = convert_tuple_pred_to_segmap(tuple_pred, raw_gt, pred_index) return odsi_db_accuracy(segmap, raw_gt, ignore_labels) def odsi_db_ResNet_18_CAM_DS_pw_balanced_accuracy(tuple_pred, raw_gt, ignore_labels=[], pred_index=0): """ @brief Evaluate the pixel-wise balanced accuracy of the class-presence classification model. @details The model and the method to obtain the segmentation is as explained in: Garcia-Peraza Herrera et al. 2020 "Intrapapillary capillary loop classification in magnification endoscopy: open dataset and baseline methodology". """ with torch.no_grad(): segmap = convert_tuple_pred_to_segmap(tuple_pred, raw_gt, pred_index) return odsi_db_balanced_accuracy(segmap, raw_gt, ignore_labels) if __name__ == '__main__': raise RuntimeError('The metric.py module is not is not a script.')
17,446
37.944196
82
py
segodsidb
segodsidb-main/src/machine/machine.py
""" @brief Module that contains the learning machine for each problem/task. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 2 Jun 2021. """ import numpy as np import torch import torchvision.utils # My imports import torchseg.base import torchseg.utils class GenericMachine(torchseg.base.BaseMachine): """ @class Generic learning machine. @param[in] model TODO @param[in] criterion TODO @param[in] metric_ftns TODO @param[in] optimizer TODO @param[in] config TODO @param[in] device TODO @param[in] data_loader TODO @param[in] valid_data_loader TODO @param[in] lr_scheduler TODO @param[in] len_epoch TODO @param[in] acc_steps Number of batches used to accumulate gradients. Increasing this value is (almost) equivalent to increasing the batch size, but without using more GPU memory. Obviously, this will slow down the training. """ def __init__(self, model, criterion, metric_ftns, optimizer, config, device, data_loader, valid_data_loader=None, lr_scheduler=None, len_epoch=None, acc_steps=1): super().__init__(model, criterion, metric_ftns, optimizer, config) self.config = config self.device = device self.data_loader = data_loader if len_epoch is None: # Epoch-based training self.len_epoch = len(self.data_loader) else: # Iteration-based training self.data_loader = torchseg.utils.inf_loop(data_loader) self.len_epoch = len_epoch self.acc_steps = acc_steps self.valid_data_loader = valid_data_loader self.do_validation = self.valid_data_loader is not None self.lr_scheduler = lr_scheduler self.log_step = int(np.sqrt(data_loader.batch_size)) self.train_metrics = torchseg.utils.MetricTracker('loss', *[m.__name__ for m in self.metric_ftns], writer=self.writer) self.valid_metrics = torchseg.utils.MetricTracker('loss', *[m.__name__ for m in self.metric_ftns], writer=self.writer) def _train_epoch(self, epoch): """ @brief Training logic for an epoch. @param[in] epoch Integer, current training epoch. @returns a log that contains average loss and metric in this epoch. """ self.model.train() self.train_metrics.reset() self.data_loader.training = True # Data structures for accumulated gradients acc_batches = 0 acc_loss = 0 self.optimizer.zero_grad() # For all the batches of this epoch for batch_idx, raw_data in enumerate(self.data_loader): # Machinery for accumulated gradients, which can be used by users when they want to # use a larger batch size than what fits in GPU acc_batches += 1 # Load batch (and expected output) into the GPU, len(data) here must be equal # to your batch size data = raw_data['image'] target = raw_data['label'] data, target = data.to(self.device), target.to(self.device) #import cv2 #foo = raw_data['image'][0].numpy().transpose((1, 2, 0)).astype( # np.uint8)[...,::-1] #print('Foo:', foo.shape) #cv2.imshow('image', foo) #fool = raw_data['label'][0, 0, :, :].numpy() * 255 #cv2.imshow('label', fool) #cv2.waitKey(0) # Perform inference output = self.model(data) # Compute loss loss = self.criterion(output, target) / self.acc_steps # Update weights with backpropagation loss.backward() acc_loss += loss.item() # Machinery for accumulated gradients if acc_batches == self.acc_steps: self.logger.debug('Train Epoch: {} {} LR: {} Accumulated loss: {:.6f}'.format( epoch, self._progress(batch_idx), [ group['lr'] for group in self.optimizer.param_groups ], acc_loss)) self.optimizer.step() acc_batches = 0 acc_loss = 0 self.optimizer.zero_grad() self.writer.set_step((epoch - 1) * self.len_epoch + batch_idx) self.train_metrics.update('loss', loss.item() * self.acc_steps) for met in self.metric_ftns: self.train_metrics.update(met.__name__, met(output, target)) if batch_idx % self.log_step == 0: self.logger.debug('Train Epoch: {} {} LR: {} Loss: {:.6f}'.format( epoch, self._progress(batch_idx), [ group['lr'] for group in self.optimizer.param_groups ], loss.item() * self.acc_steps)) # FIXME: Add image only if it has three channels #self.writer.add_image('input', # torchvision.utils.make_grid(data.cpu(), nrow=8, # normalize=True)) if batch_idx == self.len_epoch: break log = self.train_metrics.result() if self.do_validation: val_log = self._valid_epoch(epoch) log.update(**{'val_' + k : v for k, v in val_log.items()}) if self.lr_scheduler is not None: if type(self.lr_scheduler) == \ torch.optim.lr_scheduler.ReduceLROnPlateau: #self.lr_scheduler.step(self.train_metrics.avg(self.mnt_metric)) self.lr_scheduler.step(val_log['loss']) else: self.lr_scheduler.step() return log def _valid_epoch(self, epoch): """ @brief Validate after training an epoch. @param[in] epoch Integer, current training epoch. @returns A log that with the validation information. """ self.model.eval() self.valid_metrics.reset() self.data_loader.training = False with torch.no_grad(): #for batch_idx, (data, target) in enumerate(self.valid_data_loader): for batch_idx, raw_data in enumerate(self.data_loader): data = raw_data['image'] target = raw_data['label'] data, target = data.to(self.device), target.to(self.device) output = self.model(data) loss = self.criterion(output, target) print('Validation epoch:', epoch - 1) print('len(self.valid_data_loader):', len(self.valid_data_loader)) print('batch_idx:', batch_idx) self.writer.set_step((epoch - 1) \ * len(self.valid_data_loader) + batch_idx, 'valid') self.valid_metrics.update('loss', loss.item()) for met in self.metric_ftns: self.valid_metrics.update(met.__name__, met(output, target)) # FIXME: Add image only if it has three channels #self.writer.add_image('input', # torchvision.utils.make_grid(data.cpu(), nrow=8, # normalize=True)) # Add histogram of model parameters to the tensorboard # FIXME: disabled as it is not currently used #for name, p in self.model.named_parameters(): # self.writer.add_histogram(name, p, bins='auto') return self.valid_metrics.result() def _progress(self, batch_idx): """ @brief TODO @param[in] batch_idx TODO """ base = '[{}/{} ({:.0f}%)]' if hasattr(self.data_loader, 'n_samples'): current = batch_idx * self.data_loader.batch_size total = self.data_loader.n_samples else: current = batch_idx total = self.len_epoch return base.format(current, total, 100.0 * current / total)
8,241
38.435407
96
py
segodsidb
segodsidb-main/test/test_hyper2rgb.py
""" @brief Unit tests for the reconstruction of RGB images from hyperspectral data. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 9 Feb 2022. """ import unittest import numpy as np import colour import scipy # My imports import torchseg.data_loader class TestReconstructionMethods(unittest.TestCase): def test_reconstruction_of_nuance_d65(self): # Get function for the D65 illuminant il = colour.SDS_ILLUMINANTS['D65'] f_illum = scipy.interpolate.PchipInterpolator(il.wavelengths, il.values, extrapolate=True) # Generate a hyperspectral image of 51 bands containing the D65 illuminant min_wl = 450 max_wl = 950 nbands = max_wl - min_wl bands = np.linspace(min_wl, max_wl, nbands) h = 1 w = 1 im_hyper = np.empty((h, w, nbands)) im_hyper[:, :, :] = f_illum(bands) # Convert illuminant image to RGB image_loader = torchseg.data_loader.OdsiDbDataLoader.LoadImage im_rgb = image_loader.hyper2rgb(im_hyper, bands) print(im_rgb) print(im_rgb.shape) if __name__ == '__main__': unittest.main()
1,182
27.166667
98
py
segodsidb
segodsidb-main/test/test_metric.py
""" @brief Unit tests for the reconstruction of RGB images from hyperspectral data. @author Luis Carlos Garcia Peraza Herrera (luiscarlos.gph@gmail.com). @date 9 Feb 2022. """ import unittest import numpy as np import colour import scipy # My imports import torchseg.data_loader class TestReconstructionMethods(unittest.TestCase): def test_reconstruction_of_nuance_d65(self): # Get function for the D65 illuminant il = colour.SDS_ILLUMINANTS['D65'] f_illum = scipy.interpolate.PchipInterpolator(il.wavelengths, il.values, extrapolate=True) # Generate a hyperspectral image of 51 bands containing the D65 illuminant min_wl = 450 max_wl = 950 nbands = max_wl - min_wl bands = np.linspace(min_wl, max_wl, nbands) h = 1 w = 1 im_hyper = np.empty((h, w, nbands)) im_hyper[:, :, :] = f_illum(bands) # Convert illuminant image to RGB image_loader = torchseg.data_loader.OdsiDbDataLoader.LoadImage im_rgb = image_loader.hyper2rgb(im_hyper, bands) print(im_rgb) print(im_rgb.shape) if __name__ == '__main__': unittest.main()
1,182
27.166667
98
py
DiffProxy
DiffProxy-main/generate.py
import os import os.path as pth import json import numpy as np import torch from sbs_generators import generator_lookup_table from differentiable_generator import StyleGANCond from utils import read_image, write_image sat_dir = 'C:/Program Files/Allegorithmic/Substance Automation Toolkit' def compare2real(): generator_name = 'arc_pavement' out_dir = pth.join('./pretrained/synthesis', generator_name) graph_filename = f'./data/sbs/{generator_name}16.sbs' model_path = './pretrained/arc_pavement_20000_nogan.pkl' json_input = None image_res = 256 n_samples = 32 if pth.exists(out_dir) is False: os.makedirs(out_dir) sampler = generator_lookup_table[generator_name](graph_filename, 'generator', sat_dir, image_res) if json_input is None: sampler.sample(out_dir, n_samples=n_samples, vis_every=1) json_file = pth.join(out_dir, 'dataset.json') with open(json_file) as f: sampled_params = json.load(f)['labels'] else: sampler.sample_with_json(out_dir, json_input) with open(json_input) as f: sampled_params = json.load(f)['labels'] init = {'method': 'avg'} G = StyleGANCond(generator_name, model_path, init, model_type='norm') for params in sampled_params: image_name = pth.basename(params[0]) real_np = read_image(pth.join(out_dir, image_name)) parameters = params[1] p = torch.as_tensor(parameters, dtype=torch.float64, device=G.device).unsqueeze(0) comp = real_np.copy() G.set_params(p) fake = G().detach().squeeze().cpu().numpy() comp = np.concatenate((comp, fake), axis=1) write_image(pth.join(out_dir, f'fake_{image_name}'), fake) write_image(pth.join(out_dir, f'comp_{image_name}'), comp) if __name__ == "__main__": compare2real()
1,856
33.388889
101
py
DiffProxy
DiffProxy-main/sbs_generators.py
import os import os.path as pth import numbers import shutil import glob import random import torch import numpy as np from abc import ABC, abstractmethod import xml.etree.ElementTree as ET import json import subprocess from collections import OrderedDict from utils import Timer, read_image, write_image class SimpleSBSGraph: def __init__(self, sbs_file_name, params): self.sbs_file_name = sbs_file_name self.params = params self.xml_tree, self.n_nodes, self.node_params_dict = self.parse() @staticmethod def set_param(param_val_xml, val, is_int): if isinstance(val, (list, tuple)): if not all(isinstance(x, numbers.Number) for x in val): raise RuntimeError('Unknown parameter type.') param_val_str = ' '.join(str(x) for x in val) else: raise RuntimeError(f'Unknown parameter type: {type(val)}') param_tag = f'constantValueInt{len(val)}' if is_int else f'constantValueFloat{len(val)}' if len(val) == 1 and is_int: param_tag = 'constantValueInt32' param_val_xml_ = param_val_xml.find(param_tag) if param_val_xml_ is None: ET.SubElement(param_val_xml, param_tag).set('v', param_val_str) else: param_val_xml_.set('v', param_val_str) def parse(self): # Parse XML file doc_xml = ET.parse(self.sbs_file_name) graph_xml = doc_xml.getroot().find('content/graph') # find graph outputs graph_outputs_by_uid = {} for output_ in graph_xml.iter('graphoutput'): output_name = output_.find('identifier').get('v') output_uid = output_.find('uid').get('v') graph_outputs_by_uid[output_uid] = output_name n_nodes = 0 node_params_dict = OrderedDict() output_name_list = [] # check generator nodes for node_xml in graph_xml.iter('compNode'): node_uid = int(node_xml.find('uid').get('v')) node_imp = node_xml.find('compImplementation')[0] if node_imp.tag == 'compInstance': node_params = {} for param_xml in node_imp.iter('parameter'): param_name = param_xml.find('name').get('v') if param_name in self.params: node_params[param_name] = param_xml.find('paramValue') # add unregistered params to nodes unregistered_param_names = set(self.params) - set(node_params) print(f"In Node {n_nodes}, found registered params:{set(node_params)}") print(f"In Node {n_nodes}, found unregistered params: {unregistered_param_names}") params = node_imp.find('parameters') for param_name in unregistered_param_names: param_xml = ET.SubElement(params, 'parameter') ET.SubElement(param_xml, 'name').set('v', param_name) param_val_xml = ET.SubElement(param_xml, 'paramValue') self.set_param(param_val_xml, val=self.params[param_name].default_val, is_int=self.params[param_name].is_discrete) node_params[param_name] = param_val_xml n_nodes += 1 node_params_dict[node_uid] = {'params': node_params, 'name': None} elif node_imp.tag == 'compOutputBridge': pass else: raise NotImplementedError(f'This simple sbs parse cannot recognize this types of node: {node_imp.tag}') for node_xml in graph_xml.iter('compNode'): node_imp = node_xml.find('compImplementation')[0] if node_imp.tag == 'compInstance': pass elif node_imp.tag == 'compOutputBridge': output_uid = node_imp.find('output').get('v') output_name_list.append(graph_outputs_by_uid[output_uid]) connections = node_xml.findall('connections/connection') if len(connections) != 1: raise RuntimeError('A output node is not connected.') gen_uid = int(connections[0].find('connRef').get('v')) if gen_uid not in node_params_dict: raise RuntimeError('Cannot find input generator node for this output node.') node_params_dict[gen_uid]['name'] = graph_outputs_by_uid[output_uid] else: raise NotImplementedError(f'This simple sbs parse cannot recognize this types of node: {node_imp.tag}') return doc_xml, n_nodes, node_params_dict class Sampler(ABC): @abstractmethod def sample(self): pass @abstractmethod def size(self): pass class RandomSampler(Sampler): def __init__(self, min_val, max_val, default_val=None, is_discrete=False): self.min_val = min_val self.max_val = max_val self.default_val = default_val if default_val is not None else self.min_val self.is_discrete = is_discrete if is_discrete: self.func = random.randint else: self.func = random.uniform def sample(self): val = [] for min_v, max_v in zip(self.min_val, self.max_val): val.append(self.func(min_v, max_v)) return val def size(self): return len(self.min_val) class GaussianRandomSampler(Sampler): def __init__(self, min_val, max_val, mean_val, std_val, default_val=None, is_discrete=False): self.min_val = min_val self.max_val = max_val self.mean_val = mean_val self.std_val = std_val self.default_val = default_val if default_val is not None else mean_val self.is_discrete = is_discrete def get_sample_np(self): val = np.random.normal(self.mean_val, self.std_val) val = np.clip(val, self.min_val, self.max_val) return val def sample(self): val = self.get_sample_np() if self.is_discrete: val = np.rint(val).astype(np.int) return val.tolist() def size(self): return len(self.mean_val) class ParameterNormalizer: def __init__(self, min_, max_): self.min_ = min_.clone() self.max_ = max_.clone() self.range = self.max_ - self.min_ def normalize(self, x): return torch.nan_to_num((x - self.min_) / self.range) def denormalize(self, x): return x * self.range + self.min_ def __str__(self): return 'Parameter Normalizer' class ParameterStandarizer: def __init__(self, mean, std): self.mean = mean.clone() self.std = std.clone() def normalize(self, x): return torch.nan_to_num((x - self.mean) / self.std) def denormalize(self, x): return x * self.std + self.mean def __str__(self): return 'Parameter Standarizer' class ParameterRegularizer: def __init__(self, min_, max_): self.min_ = min_.clone() self.max_ = max_.clone() def regularize(self, x): return torch.clamp(x, self.min_, self.max_) def regularize_(self, x): x.clamp_(self.min_, self.max_) def check_valid(self, x): all_min = x >= self.min_ if not torch.all(all_min): l = x.shape[1] for k in range(l): i, j = x[0, k], self.min_[0, k] if i < j: print(f'For {k}th params: {i} < {j}') raise RuntimeError('Invalid parameters') all_max = x <= self.max_ if not torch.all(all_max): l = x.shape[1] for k in range(l): i, j = x[0, k], self.max_[0, k] if i > j: print(f'For {k}th params: {i} > {j}') raise RuntimeError('Invalid parameters') def get_normalizer(generator_name, normalization_type, batch_size, device): if normalization_type == 'norm': params = generator_lookup_table[generator_name].get_params() normalizer = SBSGenerators.get_normalizer(params, batch_size, device) elif normalization_type == 'std': params = generator_lookup_table[generator_name].get_params() normalizer = SBSGenerators.get_standarizer(params, batch_size, device) else: normalizer = None return normalizer class SBSGenerators: @staticmethod def get_params(): pass @staticmethod def get_normalizer(params, batch_size, device): min_, max_ = [], [] for param_name, param_sampler in params.items(): min_val = param_sampler.min_val max_val = param_sampler.max_val min_.extend(min_val) max_.extend(max_val) min_tensor = torch.as_tensor(min_, dtype=torch.float64, device=device) max_tensor = torch.as_tensor(max_, dtype=torch.float64, device=device) min_tensor = min_tensor.expand((batch_size, -1)) max_tensor = max_tensor.expand((batch_size, -1)) return ParameterNormalizer(min_tensor, max_tensor) @staticmethod def get_standarizer(params, batch_size, device): mean, std = [], [] for param_name, param_sampler in params.items(): mean_v = param_sampler.mean_val std_v = param_sampler.std_val mean.extend(mean_v) std.extend(std_v) mean_tensor = torch.as_tensor(mean, dtype=torch.float64, device=device) std_tensor = torch.as_tensor(std, dtype=torch.float64, device=device) mean_tensor = mean_tensor.expand((batch_size, -1)) std_tensor = std_tensor.expand((batch_size, -1)) return ParameterStandarizer(mean_tensor, std_tensor) @staticmethod def get_regularizer(params, batch_size, device): min_, max_ = [], [] for param_name, param_sampler in params.items(): min_val = param_sampler.min_val max_val = param_sampler.max_val min_.extend(min_val) max_.extend(max_val) min_tensor = torch.as_tensor(min_, dtype=torch.float64, device=device) max_tensor = torch.as_tensor(max_, dtype=torch.float64, device=device) min_tensor = min_tensor.expand((batch_size, -1)) max_tensor = max_tensor.expand((batch_size, -1)) return ParameterRegularizer(min_tensor, max_tensor) def __init__(self, graph_filename, graph_name, sat_dir, image_res): self.grah_filename = graph_filename self.graph_name = graph_name self.sat_dir = sat_dir self.image_res = image_res # load parameters self.params = self.get_params() # load graph self.graph = SimpleSBSGraph(graph_filename, self.params) @staticmethod def save_params(all_params, all_image_names, output_dir, i_batch): assert len(all_params) == len(all_image_names) data = dict() data['labels'] = [] for params, image_name in zip(all_params, all_image_names): data['labels'].append([image_name, params]) with open(pth.join(output_dir, f'dataset{i_batch}.json'), 'w') as outfile: json.dump(data, outfile, indent=4) @staticmethod def combine_params(input_path, n_batch, move_to_folder=None): assert (n_batch >= 1) if move_to_folder is not None: os.makedirs(move_to_folder, exist_ok=True) data_path = pth.join(input_path, 'dataset0.json') with open(data_path) as f: data = json.load(f) if move_to_folder is not None: shutil.move(data_path, pth.join(move_to_folder, f'dataset0.json')) for i in range(1, n_batch): data_path = pth.join(input_path, f'dataset{i}.json') with open(data_path) as f: data_i = json.load(f) data['labels'].extend(data_i['labels']) if move_to_folder is not None: shutil.move(data_path, pth.join(move_to_folder, f'dataset{i}.json')) output_path = pth.join(input_path, 'dataset.json') with open(output_path, 'w') as outfile: json.dump(data, outfile, indent=4) def sample(self, output_dir, n_samples, vis_every): n_samples = n_samples // self.graph.n_nodes * self.graph.n_nodes n_batch = n_samples // self.graph.n_nodes timer = Timer() timer.begin("Begin Sampling") for i in range(n_batch): params_list = [] # sample parameters for nodes in self.graph.node_params_dict.values(): params = [] for param_name, param_sampler in self.params.items(): val = param_sampler.sample() self.graph.set_param(nodes['params'][param_name], val, param_sampler.is_discrete) params.extend(val) params_list.append(params) # save sbs output_graph_filename = pth.join(output_dir, f'tmp{i}.sbs') self.save_graph(output_graph_filename) image_names_list = self.save_sample(output_graph_filename, output_dir, i) self.save_params(params_list, image_names_list, output_dir, i) if i % vis_every == 0 or i == n_batch - 1: timer.end(f'Generated {(i+1)*self.graph.n_nodes}/{n_samples} samples') timer.begin() # combine parameter json into one file self.combine_params(output_dir, n_batch, move_to_folder=pth.join(output_dir, 'params')) # move generated sbs and sbsar files to an sbs folder sbs_files = glob.glob(pth.join(output_dir, '*.sbs')) + glob.glob(pth.join(output_dir, '*.sbsar')) sbs_out_dir = pth.join(output_dir, 'sbs') if pth.exists(sbs_out_dir): shutil.rmtree(sbs_out_dir) os.makedirs(sbs_out_dir) for sbs_file in sbs_files: shutil.move(sbs_file, sbs_out_dir) def sample_with_json(self, output_dir, json_file): if isinstance(json_file, str): with open(json_file) as f: params = json.load(f)['labels'] else: params = json_file n_samples = len(params) n_batch = n_samples // self.graph.n_nodes timer = Timer() timer.begin("Begin Sampling") for i in range(n_batch): image_names = [] for k, node in enumerate(self.graph.node_params_dict.values()): idx = i * self.graph.n_nodes + k # set parameters s = 0 for param_name, param_sampler in self.params.items(): r = param_sampler.size() val = [int(np.rint(x)) if param_sampler.is_discrete else x for x in params[idx][1][s:s + r]] self.graph.set_param(node['params'][param_name], val, param_sampler.is_discrete) s += r # record image name image_name = pth.join(output_dir, pth.basename(params[idx][0])) image_names.append(image_name) # save sbs output_graph_filename = pth.join(output_dir, f'tmp{i}.sbs') self.save_graph(output_graph_filename) self.save_sample(output_graph_filename, output_dir, i, image_names) # move generated sbs and sbsar files to an sbs folder sbs_files = glob.glob(pth.join(output_dir, '*.sbs')) + glob.glob(pth.join(output_dir, '*.sbsar')) sbs_out_dir = pth.join(output_dir, 'sbs') if pth.exists(sbs_out_dir): shutil.rmtree(sbs_out_dir) os.makedirs(sbs_out_dir) for sbs_file in sbs_files: shutil.move(sbs_file, sbs_out_dir) # save sbs graph back to an sbs file def save_graph(self, output_graph_filename): self.graph.xml_tree.write(output_graph_filename) # cook and output images def save_sample(self, input_graph_filename, output_dir, i_batch, image_names=None): tmp_output_dir = pth.join(output_dir, 'tmp') os.makedirs(tmp_output_dir, exist_ok=True) command_cooker = ( f'"{os.path.join(self.sat_dir, "sbscooker")}" ' f'--inputs "{input_graph_filename}" ' f'--alias "sbs://{os.path.join(self.sat_dir, "resources", "packages")}" ' f'--output-path {{inputPath}}') completed_process = subprocess.run(command_cooker, shell=True, capture_output=True, text=True) if completed_process.returncode != 0: raise RuntimeError(f'Error while running sbs cooker:\n{completed_process.stderr}') # import pdb; pdb.set_trace() cooked_input_graph_filename = pth.splitext(input_graph_filename)[0] + '.sbsar' image_format = 'png' command_render = ( f'"{os.path.join(self.sat_dir, "sbsrender")}" render ' f'--inputs "{cooked_input_graph_filename}" ' f'--input-graph "{self.graph_name}" ' f'--output-format "{image_format}" ' f'--output-path "{tmp_output_dir}" ' f'--output-name "{{outputNodeName}}"') completed_process = subprocess.run(command_render, shell=True, capture_output=True, text=True) if completed_process.returncode != 0: raise RuntimeError(f'Error while running sbs render:\n{completed_process.stderr}') image_list = [pth.join(tmp_output_dir, f'{node["name"]}.png') for node in self.graph.node_params_dict.values()] assert len(image_list) == self.graph.n_nodes image_names_list = [] dir_name = pth.basename(output_dir) for i, image_filename in enumerate(image_list): if image_names is None: image_name = pth.join(output_dir, '{:08d}.png'.format(i_batch*self.graph.n_nodes + i)) else: image_name = image_names[i] shutil.move(image_filename, image_name) # convert to 8bit self.convert(image_name) image_name = f'{dir_name}/{pth.basename(image_name)}' image_names_list.append(image_name) os.rmdir(tmp_output_dir) return image_names_list @staticmethod def convert(image_file): im = read_image(image_file) write_image(image_file, im) class ArcPavement(SBSGenerators): @staticmethod def get_params(): params = OrderedDict([('pattern_amount', RandomSampler((4,), (32,), (12,), True)), ('arcs_amount', RandomSampler((4,), (20,), (14,), True)), ('pattern_scale', RandomSampler((0.9,), (1.0,), (1.0,), False)), ('pattern_width', RandomSampler((0.7,), (0.9,), (0.8,), False)), ('pattern_height', RandomSampler((0.8,), (1.0,), (0.9,), False)), ('pattern_width_random', RandomSampler((0.0,), (0.2,), (0.0,), False)), ('pattern_height_random', RandomSampler((0.0,), (0.2,), (0.0,), False)), ('global_pattern_width_random', RandomSampler((0.0,), (0.2,), (0.0,), False)), ('pattern_height_decrease', RandomSampler((0.0,), (0.5,), (0.25,), False)), ('color_random', RandomSampler((0.0,), (1.0,), (0.0,), False)), ]) return params generator_lookup_table = {'arc_pavement': ArcPavement, }
19,510
37.107422
119
py
DiffProxy
DiffProxy-main/vgg.py
import os import torch import torch.nn as nn import torch.nn.functional as F vgg_path = 'pretrained/vgg_conv.pt' # vgg definition that conveniently let's you grab the outputs from any layer class VGG(nn.Module): def __init__(self, pool='max'): super(VGG, self).__init__() # vgg modules self.conv1_1 = nn.Conv2d(3, 64, kernel_size=3, padding=1) self.conv1_2 = nn.Conv2d(64, 64, kernel_size=3, padding=1) self.conv2_1 = nn.Conv2d(64, 128, kernel_size=3, padding=1) self.conv2_2 = nn.Conv2d(128, 128, kernel_size=3, padding=1) self.conv3_1 = nn.Conv2d(128, 256, kernel_size=3, padding=1) self.conv3_2 = nn.Conv2d(256, 256, kernel_size=3, padding=1) self.conv3_3 = nn.Conv2d(256, 256, kernel_size=3, padding=1) self.conv3_4 = nn.Conv2d(256, 256, kernel_size=3, padding=1) self.conv4_1 = nn.Conv2d(256, 512, kernel_size=3, padding=1) self.conv4_2 = nn.Conv2d(512, 512, kernel_size=3, padding=1) self.conv4_3 = nn.Conv2d(512, 512, kernel_size=3, padding=1) self.conv4_4 = nn.Conv2d(512, 512, kernel_size=3, padding=1) self.conv5_1 = nn.Conv2d(512, 512, kernel_size=3, padding=1) self.conv5_2 = nn.Conv2d(512, 512, kernel_size=3, padding=1) self.conv5_3 = nn.Conv2d(512, 512, kernel_size=3, padding=1) self.conv5_4 = nn.Conv2d(512, 512, kernel_size=3, padding=1) if pool == 'max': self.pool1 = nn.MaxPool2d(kernel_size=2, stride=2) self.pool2 = nn.MaxPool2d(kernel_size=2, stride=2) self.pool3 = nn.MaxPool2d(kernel_size=2, stride=2) self.pool4 = nn.MaxPool2d(kernel_size=2, stride=2) self.pool5 = nn.MaxPool2d(kernel_size=2, stride=2) elif pool == 'avg': self.pool1 = nn.AvgPool2d(kernel_size=2, stride=2) self.pool2 = nn.AvgPool2d(kernel_size=2, stride=2) self.pool3 = nn.AvgPool2d(kernel_size=2, stride=2) self.pool4 = nn.AvgPool2d(kernel_size=2, stride=2) self.pool5 = nn.AvgPool2d(kernel_size=2, stride=2) self.normalizer = VGG19Normalizer() def forward(self, x, out_keys): out = {} out['r11'] = F.relu(self.conv1_1(x)) out['r12'] = F.relu(self.conv1_2(out['r11'])) out['p1'] = self.pool1(out['r12']) out['r21'] = F.relu(self.conv2_1(out['p1'])) out['r22'] = F.relu(self.conv2_2(out['r21'])) out['p2'] = self.pool2(out['r22']) out['r31'] = F.relu(self.conv3_1(out['p2'])) out['r32'] = F.relu(self.conv3_2(out['r31'])) out['r33'] = F.relu(self.conv3_3(out['r32'])) out['r34'] = F.relu(self.conv3_4(out['r33'])) out['p3'] = self.pool3(out['r34']) out['r41'] = F.relu(self.conv4_1(out['p3'])) out['r42'] = F.relu(self.conv4_2(out['r41'])) out['r43'] = F.relu(self.conv4_3(out['r42'])) out['r44'] = F.relu(self.conv4_4(out['r43'])) out['p4'] = self.pool4(out['r44']) out['r51'] = F.relu(self.conv5_1(out['p4'])) out['r52'] = F.relu(self.conv5_2(out['r51'])) out['r53'] = F.relu(self.conv5_3(out['r52'])) out['r54'] = F.relu(self.conv5_4(out['r53'])) out['p5'] = self.pool5(out['r54']) return [out[key] for key in out_keys] def extract_features(self, im, feature_layers, style_layers, detach_features=False, detach_styles=False): if im.shape[1] == 1: im = torch.cat((im, im, im), dim=1) x = self.normalizer.normalize(im) out = {} out['r11'] = F.relu(self.conv1_1(x)) out['r12'] = F.relu(self.conv1_2(out['r11'])) out['p1'] = self.pool1(out['r12']) out['r21'] = F.relu(self.conv2_1(out['p1'])) out['r22'] = F.relu(self.conv2_2(out['r21'])) out['p2'] = self.pool2(out['r22']) out['r31'] = F.relu(self.conv3_1(out['p2'])) out['r32'] = F.relu(self.conv3_2(out['r31'])) out['r33'] = F.relu(self.conv3_3(out['r32'])) out['r34'] = F.relu(self.conv3_4(out['r33'])) out['p3'] = self.pool3(out['r34']) out['r41'] = F.relu(self.conv4_1(out['p3'])) out['r42'] = F.relu(self.conv4_2(out['r41'])) out['r43'] = F.relu(self.conv4_3(out['r42'])) out['r44'] = F.relu(self.conv4_4(out['r43'])) out['p4'] = self.pool4(out['r44']) out['r51'] = F.relu(self.conv5_1(out['p4'])) out['r52'] = F.relu(self.conv5_2(out['r51'])) out['r53'] = F.relu(self.conv5_3(out['r52'])) out['r54'] = F.relu(self.conv5_4(out['r53'])) out['p5'] = self.pool5(out['r54']) if not detach_features: extracted_features = [out[layer] for layer in feature_layers] else: extracted_features = [out[layer].detach() for layer in feature_layers] if not detach_styles: extracted_styles = [gram(out[layer]) for layer in style_layers] else: extracted_styles = [gram(out[layer]).detach() for layer in style_layers] return extracted_features, extracted_styles # get network def get_vgg19(): vgg = VGG() vgg.load_state_dict(torch.load(vgg_path)) for param in vgg.parameters(): param.requires_grad = False if torch.cuda.is_available(): vgg.cuda() return vgg def gram(x): b, c, h, w = x.shape F = x.view(b, c, h*w) G = torch.bmm(F, F.transpose(1,2)) G.div_(h*w) return G class WeightedLoss(nn.Module): def __init__(self, weights, metric='l2'): super(WeightedLoss, self).__init__() self.weights = weights if metric == 'l2': self.criterion = nn.MSELoss().cuda() elif metric == 'l1': self.criterion = nn.L1Loss().cuda() else: raise NotImplementedError('Unknown metric {}'.format(metric)) def forward(self, x, y): loss = torch.tensor(0.0, requires_grad=True).cuda() for w, x_, y_ in zip(self.weights, x, y): loss = loss + w * self.criterion(x_, y_) return loss class VGG19Normalizer(nn.Module): def __init__(self): super(VGG19Normalizer,self).__init__() imagenet_mean = [0.40760392, 0.45795686, 0.48501961] self.mean = torch.as_tensor(imagenet_mean)[None, :, None, None].cuda() def normalize(self, image): image = image * 0.5 + 0.5 # (-1, 1) to (0, 1) image = image[:, [2, 1, 0], :, :] # turn to BGR image = image - self.mean # subtract imagenet mean image = image * 255.0 return image def denormalize(self, image): image = image / 255.0 image = image + self.mean # add imagenet mean image = image[:, [2, 1, 0], :, :] # turn to RGB image = image * 2.0 - 1.0 # (0, 1) to (-1, 1) return image
6,808
40.266667
109
py
DiffProxy
DiffProxy-main/loss.py
import torch.nn as nn import vgg from training.loss import * class ProxyLoss: def __init__(self, device, G, D, augment_pipe=None, r1_gamma=10, style_mixing_prob=0, pl_weight=0, pl_batch_shrink=2, pl_decay=0.01, pl_no_weight_grad=False, blur_init_sigma=0, blur_fade_kimg=0): super().__init__() self.device = device self.G = G # l1 loss self.l1_criterion = nn.L1Loss() # VGG loss self.vgg19 = vgg.get_vgg19() self.style_layers = ['r11', 'r21', 'r31', 'r41', 'r51'] self.style_weights = [1 / n ** 2 for n in [64, 128, 256, 512, 512]] self.feature_layers = ['r22', 'r32', 'r42'] self.feature_weights = [1e-3, 1e-3, 1e-3] self.criterion_feature = vgg.WeightedLoss(self.feature_weights, metric='l1') self.criterion_style = vgg.WeightedLoss(self.style_weights, metric='l1') self.feat_w = 10.0 self.style_w = 1.0 print(f"Loss Config: l1_w = 1, feat_w = {self.feat_w}, style_w = {self.style_w}") def run_G(self, z, c, update_emas=False): ws = self.G.mapping(z, c, update_emas=update_emas) img = self.G.synthesis(ws, update_emas=update_emas) return img, ws def accumulate_gradients(self, real_img, real_c, gen_z): with torch.autograd.profiler.record_function('Gmain_forward'): gen_img, _gen_ws = self.run_G(gen_z, real_c) # l1 loss loss_l1 = self.l1_criterion(gen_img, real_img) training_stats.report('Loss/L1/loss', loss_l1) # VGG loss real_feat, real_style = self.vgg19.extract_features(real_img, self.feature_layers, self.style_layers, detach_features=True, detach_styles=True) recon_feat, recon_style = self.vgg19.extract_features(gen_img, self.feature_layers, self.style_layers) feature_loss = self.criterion_feature(real_feat, recon_feat) * self.feat_w style_loss = self.criterion_style(real_style, recon_style) * self.style_w training_stats.report('Loss/Feat/loss', feature_loss) training_stats.report('Loss/Style/loss', style_loss) loss = loss_l1 + feature_loss + style_loss training_stats.report('Loss/G/loss', loss) with torch.autograd.profiler.record_function('Gmain_backward'): loss.backward() class ProxyGANLoss: def __init__(self, device, G, D, augment_pipe=None, r1_gamma=10, style_mixing_prob=0, pl_weight=0, pl_batch_shrink=2, pl_decay=0.01, pl_no_weight_grad=False, blur_init_sigma=0, blur_fade_kimg=0): super().__init__() self.device = device self.G = G self.D = D self.r1_gamma = r1_gamma self.pl_weight = 0 # do not include path length regularization self.blur_init_sigma = blur_init_sigma self.blur_fade_kimg = blur_fade_kimg # l1 loss self.l1_criterion = nn.L1Loss() # VGG loss self.vgg19 = vgg.get_vgg19() self.style_layers = ['r11', 'r21', 'r31', 'r41', 'r51'] self.style_weights = [1 / n ** 2 for n in [64, 128, 256, 512, 512]] self.feature_layers = ['r22', 'r32', 'r42'] self.feature_weights = [1e-3, 1e-3, 1e-3] self.criterion_feature = vgg.WeightedLoss(self.feature_weights, metric='l1') self.criterion_style = vgg.WeightedLoss(self.style_weights, metric='l1') self.l1_w = 1.0 self.feat_w = 1.0 self.style_w = 1.0 self.gan_w = 0.1 print(f"Loss Config: l1_w = {self.l1_w}, feat_w = {self.feat_w}, style_w = {self.style_w}, gan_w = {self.gan_w}") def run_G(self, z, c, update_emas=False): ws = self.G.mapping(z, c, update_emas=update_emas) img = self.G.synthesis(ws, update_emas=update_emas) return img, ws def run_D(self, img, c, blur_sigma=0, update_emas=False): blur_size = np.floor(blur_sigma * 3) if blur_size > 0: with torch.autograd.profiler.record_function('blur'): f = torch.arange(-blur_size, blur_size + 1, device=img.device).div(blur_sigma).square().neg().exp2() img = upfirdn2d.filter2d(img, f / f.sum()) logits = self.D(img, c, update_emas=update_emas) return logits def accumulate_gradients(self, phase, real_img, real_c, gen_z, gain, cur_nimg): assert phase in ['Gmain', 'Greg', 'Gboth', 'Dmain', 'Dreg', 'Dboth'] if self.pl_weight == 0: phase = {'Greg': 'none', 'Gboth': 'Gmain'}.get(phase, phase) if self.r1_gamma == 0: phase = {'Dreg': 'none', 'Dboth': 'Dmain'}.get(phase, phase) blur_sigma = max(1 - cur_nimg / (self.blur_fade_kimg * 1e3), 0) * self.blur_init_sigma if self.blur_fade_kimg > 0 else 0 # Gmain: Maximize logits for generated images. if phase in ['Gmain', 'Gboth']: with torch.autograd.profiler.record_function('Gmain_forward'): gen_img, _gen_ws = self.run_G(gen_z, real_c) gen_logits = self.run_D(gen_img, real_c, blur_sigma=blur_sigma) training_stats.report('Loss/scores/fake', gen_logits) training_stats.report('Loss/signs/fake', gen_logits.sign()) # l1 loss loss_l1 = self.l1_criterion(gen_img, real_img) * self.l1_w training_stats.report('Loss/L1/loss', loss_l1) # VGG loss real_feat, real_style = self.vgg19.extract_features(real_img, self.feature_layers, self.style_layers, detach_features=True, detach_styles=True) recon_feat, recon_style = self.vgg19.extract_features(gen_img, self.feature_layers, self.style_layers) feature_loss = self.criterion_feature(real_feat, recon_feat) * self.feat_w style_loss = self.criterion_style(real_style, recon_style) * self.style_w training_stats.report('Loss/Feat/loss', feature_loss) training_stats.report('Loss/Style/loss', style_loss) # GAN loss loss_G = torch.nn.functional.softplus(-gen_logits).mean()*self.gan_w # -log(sigmoid(gen_logits)) training_stats.report('Loss/GAN/loss', loss_G) loss_Gmain = loss_G + loss_l1 + feature_loss + style_loss training_stats.report('Loss/G/loss', loss_Gmain) with torch.autograd.profiler.record_function('Gmain_backward'): loss_Gmain.mul(gain).backward() # loss_Gmain.mean().mul(gain).backward() # Dmain: Minimize logits for generated images. loss_Dgen = 0 if phase in ['Dmain', 'Dboth']: with torch.autograd.profiler.record_function('Dgen_forward'): gen_img, _gen_ws = self.run_G(gen_z, real_c, update_emas=True) gen_logits = self.run_D(gen_img, real_c, blur_sigma=blur_sigma, update_emas=True) training_stats.report('Loss/scores/fake', gen_logits) training_stats.report('Loss/signs/fake', gen_logits.sign()) loss_Dgen = torch.nn.functional.softplus(gen_logits) # -log(1 - sigmoid(gen_logits)) with torch.autograd.profiler.record_function('Dgen_backward'): loss_Dgen.mean().mul(gain).backward() # Dmain: Maximize logits for real images. # Dr1: Apply R1 regularization. if phase in ['Dmain', 'Dreg', 'Dboth']: name = 'Dreal' if phase == 'Dmain' else 'Dr1' if phase == 'Dreg' else 'Dreal_Dr1' with torch.autograd.profiler.record_function(name + '_forward'): real_img_tmp = real_img.detach().requires_grad_(phase in ['Dreg', 'Dboth']) real_logits = self.run_D(real_img_tmp, real_c, blur_sigma=blur_sigma) training_stats.report('Loss/scores/real', real_logits) training_stats.report('Loss/signs/real', real_logits.sign()) loss_Dreal = 0 if phase in ['Dmain', 'Dboth']: loss_Dreal = torch.nn.functional.softplus(-real_logits) # -log(sigmoid(real_logits)) training_stats.report('Loss/D/loss', loss_Dgen + loss_Dreal) loss_Dr1 = 0 if phase in ['Dreg', 'Dboth']: with torch.autograd.profiler.record_function('r1_grads'), conv2d_gradfix.no_weight_gradients(): r1_grads = torch.autograd.grad(outputs=[real_logits.sum()], inputs=[real_img_tmp], create_graph=True, only_inputs=True)[0] r1_penalty = r1_grads.square().sum([1,2,3]) loss_Dr1 = r1_penalty * (self.r1_gamma / 2) training_stats.report('Loss/r1_penalty', r1_penalty) training_stats.report('Loss/D/reg', loss_Dr1) with torch.autograd.profiler.record_function(name + '_backward'): (loss_Dreal + loss_Dr1).mean().mul(gain).backward()
9,171
50.52809
199
py
DiffProxy
DiffProxy-main/differentiable_generator.py
import sys import os.path as pth import torch.nn as nn import torch import pickle import cv2 from utils import read_image from sbs_generators import generator_lookup_table, SBSGenerators sys.path.append('stylegan') sat_dir = 'C:/Program Files/Allegorithmic/Substance Automation Toolkit' sbs_lib = {'arc_pavement': './data/sbs/arc_pavement.sbs'} def get_real_generator(generator_name): if generator_name in sbs_lib: graph_filename = sbs_lib[generator_name] return generator_lookup_table[generator_name](graph_filename, 'generator', sat_dir, 256) else: raise RuntimeError(f'Cannot find {generator_name}\'s real signature') class DifferentiableGenerator(nn.Module): def output(self, *arg): return self.forward() def sample_params(self): return None def get_params(self): return None def set_params(self, params): pass def regularize(self): pass def get_optimizable(self): raise NotImplementedError class StyleGANCond(DifferentiableGenerator): @staticmethod def load_pretrained_model(model_path): with open(model_path, 'rb') as f: dat = pickle.load(f) if 'G_ema' in dat: G = dat['G_ema'] else: G = dat['G'] G = G.eval().requires_grad_(False).cuda() # torch.nn.Module return G def get_normalizer(self): params = generator_lookup_table[self.generator_name].get_params() normalizer = SBSGenerators.get_normalizer(params, batch_size=1, device=self.device) return normalizer def get_standarizer(self): params = generator_lookup_table[self.generator_name].get_params() normalizer = SBSGenerators.get_standarizer(params, batch_size=1, device=self.device) return normalizer def get_reguluarizer(self): params = generator_lookup_table[self.generator_name].get_params() regularizer = SBSGenerators.get_regularizer(params, batch_size=1, device=self.device) return regularizer def init_avg(self): assert self.normalization is not None if self.normalization == 'norm': self.set_params((self.normalizer.min_ + self.normalizer.max_)*0.5) else: self.set_params(self.normalizer.mean) def init_rand(self): params = self.sample_params() self.set_params(params) def sample_params(self): if self.normalization == 'norm': params = torch.rand_like(self.normalizer.min_) * (self.normalizer.max_ - self.normalizer.min_) + self.normalizer.min_ else: params = torch.randn_like(self.normalizer.mean) * self.normalizer.std + self.normalizer.mean self.regularizer.regularize_(params) return params def init_direct(self, init_params): params = torch.as_tensor(init_params, device=self.device, dtype=torch.float64).unsqueeze(0) assert params.ndim == 2 self.set_params(params) def __init__(self, generator_name, model_path, init, model_type='none'): super(StyleGANCond, self).__init__() self.model_path = model_path self.generator_name = generator_name self.init = init self.G = self.load_pretrained_model(model_path) self.img_resolution = self.G.img_resolution self.real_G = None self.params = None self.optimizable = None self.device = torch.device('cuda') self.model_type = model_type assert model_type in ['norm', 'std'] self.normalization = model_type if self.normalization == 'norm': self.normalizer = self.get_normalizer() else: self.normalizer = self.get_standarizer() self.regularizer = self.get_reguluarizer() init_method = init['method'] if init_method == 'avg': self.init_avg() elif init_method == 'rand': self.init_rand() elif init_method == 'direct': self.init_direct(init['init_params']) else: raise RuntimeError("Unknown initialization method") def get_optimizable(self): return self.optimizable def forward(self): ws = self.G.mapping(None, self.params) im = self.G.synthesis(ws) im = (im + 1.0) / 2.0 # (-1, 1) to (0, 1) return im.clamp(0, 1) def regularize(self): params = self.get_params() params = self.regularizer.regularize(params) params = self.normalizer.normalize(params) self.params.data = params.data def get_params(self): return self.normalizer.denormalize(self.params) def set_params(self, params, validate=True): if validate: # params are unnormalized self.regularizer.check_valid(params) self.params = self.normalizer.normalize(params).detach().clone() self.params.requires_grad = True self.optimizable = [self.params] # get real generator map def output(self, tmp_dir, tmp_image_filename): if self.real_G is None: self.real_G = get_real_generator(self.generator_name) params = self.get_params()[0].detach().cpu().tolist() json_file = [(tmp_image_filename, params)] self.real_G.sample_with_json(tmp_dir, json_file) # reload image from disk image_filename = pth.join(tmp_dir, tmp_image_filename) image_np = read_image(image_filename) image_np = cv2.resize(image_np, (self.G.img_resolution, self.G.img_resolution)) assert image_np.ndim == 2 image = torch.as_tensor(image_np, dtype=torch.float32, device=self.device).unsqueeze(0).unsqueeze(0) return image
5,747
32.811765
129
py
DiffProxy
DiffProxy-main/stylegan/legacy.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Converting legacy network pickle into the new format.""" import click import pickle import re import copy import numpy as np import torch import dnnlib from torch_utils import misc #---------------------------------------------------------------------------- def load_network_pkl(f, force_fp16=False): data = _LegacyUnpickler(f).load() # Legacy TensorFlow pickle => convert. if isinstance(data, tuple) and len(data) == 3 and all(isinstance(net, _TFNetworkStub) for net in data): tf_G, tf_D, tf_Gs = data G = convert_tf_generator(tf_G) D = convert_tf_discriminator(tf_D) G_ema = convert_tf_generator(tf_Gs) data = dict(G=G, D=D, G_ema=G_ema) # Add missing fields. if 'training_set_kwargs' not in data: data['training_set_kwargs'] = None if 'augment_pipe' not in data: data['augment_pipe'] = None # Validate contents. assert isinstance(data['G'], torch.nn.Module) assert isinstance(data['D'], torch.nn.Module) assert isinstance(data['G_ema'], torch.nn.Module) assert isinstance(data['training_set_kwargs'], (dict, type(None))) assert isinstance(data['augment_pipe'], (torch.nn.Module, type(None))) # Force FP16. if force_fp16: for key in ['G', 'D', 'G_ema']: old = data[key] kwargs = copy.deepcopy(old.init_kwargs) fp16_kwargs = kwargs.get('synthesis_kwargs', kwargs) fp16_kwargs.num_fp16_res = 4 fp16_kwargs.conv_clamp = 256 if kwargs != old.init_kwargs: new = type(old)(**kwargs).eval().requires_grad_(False) misc.copy_params_and_buffers(old, new, require_all=True) data[key] = new return data #---------------------------------------------------------------------------- class _TFNetworkStub(dnnlib.EasyDict): pass class _LegacyUnpickler(pickle.Unpickler): def find_class(self, module, name): if module == 'dnnlib.tflib.network' and name == 'Network': return _TFNetworkStub return super().find_class(module, name) #---------------------------------------------------------------------------- def _collect_tf_params(tf_net): # pylint: disable=protected-access tf_params = dict() def recurse(prefix, tf_net): for name, value in tf_net.variables: tf_params[prefix + name] = value for name, comp in tf_net.components.items(): recurse(prefix + name + '/', comp) recurse('', tf_net) return tf_params #---------------------------------------------------------------------------- def _populate_module_params(module, *patterns): for name, tensor in misc.named_params_and_buffers(module): found = False value = None for pattern, value_fn in zip(patterns[0::2], patterns[1::2]): match = re.fullmatch(pattern, name) if match: found = True if value_fn is not None: value = value_fn(*match.groups()) break try: assert found if value is not None: tensor.copy_(torch.from_numpy(np.array(value))) except: print(name, list(tensor.shape)) raise #---------------------------------------------------------------------------- def convert_tf_generator(tf_G): if tf_G.version < 4: raise ValueError('TensorFlow pickle version too low') # Collect kwargs. tf_kwargs = tf_G.static_kwargs known_kwargs = set() def kwarg(tf_name, default=None, none=None): known_kwargs.add(tf_name) val = tf_kwargs.get(tf_name, default) return val if val is not None else none # Convert kwargs. from training import networks_stylegan2 network_class = networks_stylegan2.Generator kwargs = dnnlib.EasyDict( z_dim = kwarg('latent_size', 512), c_dim = kwarg('label_size', 0), w_dim = kwarg('dlatent_size', 512), img_resolution = kwarg('resolution', 1024), img_channels = kwarg('num_channels', 3), channel_base = kwarg('fmap_base', 16384) * 2, channel_max = kwarg('fmap_max', 512), num_fp16_res = kwarg('num_fp16_res', 0), conv_clamp = kwarg('conv_clamp', None), architecture = kwarg('architecture', 'skip'), resample_filter = kwarg('resample_kernel', [1,3,3,1]), use_noise = kwarg('use_noise', True), activation = kwarg('nonlinearity', 'lrelu'), mapping_kwargs = dnnlib.EasyDict( num_layers = kwarg('mapping_layers', 8), embed_features = kwarg('label_fmaps', None), layer_features = kwarg('mapping_fmaps', None), activation = kwarg('mapping_nonlinearity', 'lrelu'), lr_multiplier = kwarg('mapping_lrmul', 0.01), w_avg_beta = kwarg('w_avg_beta', 0.995, none=1), ), ) # Check for unknown kwargs. kwarg('truncation_psi') kwarg('truncation_cutoff') kwarg('style_mixing_prob') kwarg('structure') kwarg('conditioning') kwarg('fused_modconv') unknown_kwargs = list(set(tf_kwargs.keys()) - known_kwargs) if len(unknown_kwargs) > 0: raise ValueError('Unknown TensorFlow kwarg', unknown_kwargs[0]) # Collect params. tf_params = _collect_tf_params(tf_G) for name, value in list(tf_params.items()): match = re.fullmatch(r'ToRGB_lod(\d+)/(.*)', name) if match: r = kwargs.img_resolution // (2 ** int(match.group(1))) tf_params[f'{r}x{r}/ToRGB/{match.group(2)}'] = value kwargs.synthesis.kwargs.architecture = 'orig' #for name, value in tf_params.items(): print(f'{name:<50s}{list(value.shape)}') # Convert params. G = network_class(**kwargs).eval().requires_grad_(False) # pylint: disable=unnecessary-lambda # pylint: disable=f-string-without-interpolation _populate_module_params(G, r'mapping\.w_avg', lambda: tf_params[f'dlatent_avg'], r'mapping\.embed\.weight', lambda: tf_params[f'mapping/LabelEmbed/weight'].transpose(), r'mapping\.embed\.bias', lambda: tf_params[f'mapping/LabelEmbed/bias'], r'mapping\.fc(\d+)\.weight', lambda i: tf_params[f'mapping/Dense{i}/weight'].transpose(), r'mapping\.fc(\d+)\.bias', lambda i: tf_params[f'mapping/Dense{i}/bias'], r'synthesis\.b4\.const', lambda: tf_params[f'synthesis/4x4/Const/const'][0], r'synthesis\.b4\.conv1\.weight', lambda: tf_params[f'synthesis/4x4/Conv/weight'].transpose(3, 2, 0, 1), r'synthesis\.b4\.conv1\.bias', lambda: tf_params[f'synthesis/4x4/Conv/bias'], r'synthesis\.b4\.conv1\.noise_const', lambda: tf_params[f'synthesis/noise0'][0, 0], r'synthesis\.b4\.conv1\.noise_strength', lambda: tf_params[f'synthesis/4x4/Conv/noise_strength'], r'synthesis\.b4\.conv1\.affine\.weight', lambda: tf_params[f'synthesis/4x4/Conv/mod_weight'].transpose(), r'synthesis\.b4\.conv1\.affine\.bias', lambda: tf_params[f'synthesis/4x4/Conv/mod_bias'] + 1, r'synthesis\.b(\d+)\.conv0\.weight', lambda r: tf_params[f'synthesis/{r}x{r}/Conv0_up/weight'][::-1, ::-1].transpose(3, 2, 0, 1), r'synthesis\.b(\d+)\.conv0\.bias', lambda r: tf_params[f'synthesis/{r}x{r}/Conv0_up/bias'], r'synthesis\.b(\d+)\.conv0\.noise_const', lambda r: tf_params[f'synthesis/noise{int(np.log2(int(r)))*2-5}'][0, 0], r'synthesis\.b(\d+)\.conv0\.noise_strength', lambda r: tf_params[f'synthesis/{r}x{r}/Conv0_up/noise_strength'], r'synthesis\.b(\d+)\.conv0\.affine\.weight', lambda r: tf_params[f'synthesis/{r}x{r}/Conv0_up/mod_weight'].transpose(), r'synthesis\.b(\d+)\.conv0\.affine\.bias', lambda r: tf_params[f'synthesis/{r}x{r}/Conv0_up/mod_bias'] + 1, r'synthesis\.b(\d+)\.conv1\.weight', lambda r: tf_params[f'synthesis/{r}x{r}/Conv1/weight'].transpose(3, 2, 0, 1), r'synthesis\.b(\d+)\.conv1\.bias', lambda r: tf_params[f'synthesis/{r}x{r}/Conv1/bias'], r'synthesis\.b(\d+)\.conv1\.noise_const', lambda r: tf_params[f'synthesis/noise{int(np.log2(int(r)))*2-4}'][0, 0], r'synthesis\.b(\d+)\.conv1\.noise_strength', lambda r: tf_params[f'synthesis/{r}x{r}/Conv1/noise_strength'], r'synthesis\.b(\d+)\.conv1\.affine\.weight', lambda r: tf_params[f'synthesis/{r}x{r}/Conv1/mod_weight'].transpose(), r'synthesis\.b(\d+)\.conv1\.affine\.bias', lambda r: tf_params[f'synthesis/{r}x{r}/Conv1/mod_bias'] + 1, r'synthesis\.b(\d+)\.torgb\.weight', lambda r: tf_params[f'synthesis/{r}x{r}/ToRGB/weight'].transpose(3, 2, 0, 1), r'synthesis\.b(\d+)\.torgb\.bias', lambda r: tf_params[f'synthesis/{r}x{r}/ToRGB/bias'], r'synthesis\.b(\d+)\.torgb\.affine\.weight', lambda r: tf_params[f'synthesis/{r}x{r}/ToRGB/mod_weight'].transpose(), r'synthesis\.b(\d+)\.torgb\.affine\.bias', lambda r: tf_params[f'synthesis/{r}x{r}/ToRGB/mod_bias'] + 1, r'synthesis\.b(\d+)\.skip\.weight', lambda r: tf_params[f'synthesis/{r}x{r}/Skip/weight'][::-1, ::-1].transpose(3, 2, 0, 1), r'.*\.resample_filter', None, r'.*\.act_filter', None, ) return G #---------------------------------------------------------------------------- def convert_tf_discriminator(tf_D): if tf_D.version < 4: raise ValueError('TensorFlow pickle version too low') # Collect kwargs. tf_kwargs = tf_D.static_kwargs known_kwargs = set() def kwarg(tf_name, default=None): known_kwargs.add(tf_name) return tf_kwargs.get(tf_name, default) # Convert kwargs. kwargs = dnnlib.EasyDict( c_dim = kwarg('label_size', 0), img_resolution = kwarg('resolution', 1024), img_channels = kwarg('num_channels', 3), architecture = kwarg('architecture', 'resnet'), channel_base = kwarg('fmap_base', 16384) * 2, channel_max = kwarg('fmap_max', 512), num_fp16_res = kwarg('num_fp16_res', 0), conv_clamp = kwarg('conv_clamp', None), cmap_dim = kwarg('mapping_fmaps', None), block_kwargs = dnnlib.EasyDict( activation = kwarg('nonlinearity', 'lrelu'), resample_filter = kwarg('resample_kernel', [1,3,3,1]), freeze_layers = kwarg('freeze_layers', 0), ), mapping_kwargs = dnnlib.EasyDict( num_layers = kwarg('mapping_layers', 0), embed_features = kwarg('mapping_fmaps', None), layer_features = kwarg('mapping_fmaps', None), activation = kwarg('nonlinearity', 'lrelu'), lr_multiplier = kwarg('mapping_lrmul', 0.1), ), epilogue_kwargs = dnnlib.EasyDict( mbstd_group_size = kwarg('mbstd_group_size', None), mbstd_num_channels = kwarg('mbstd_num_features', 1), activation = kwarg('nonlinearity', 'lrelu'), ), ) # Check for unknown kwargs. kwarg('structure') kwarg('conditioning') unknown_kwargs = list(set(tf_kwargs.keys()) - known_kwargs) if len(unknown_kwargs) > 0: raise ValueError('Unknown TensorFlow kwarg', unknown_kwargs[0]) # Collect params. tf_params = _collect_tf_params(tf_D) for name, value in list(tf_params.items()): match = re.fullmatch(r'FromRGB_lod(\d+)/(.*)', name) if match: r = kwargs.img_resolution // (2 ** int(match.group(1))) tf_params[f'{r}x{r}/FromRGB/{match.group(2)}'] = value kwargs.architecture = 'orig' #for name, value in tf_params.items(): print(f'{name:<50s}{list(value.shape)}') # Convert params. from training import networks_stylegan2 D = networks_stylegan2.Discriminator(**kwargs).eval().requires_grad_(False) # pylint: disable=unnecessary-lambda # pylint: disable=f-string-without-interpolation _populate_module_params(D, r'b(\d+)\.fromrgb\.weight', lambda r: tf_params[f'{r}x{r}/FromRGB/weight'].transpose(3, 2, 0, 1), r'b(\d+)\.fromrgb\.bias', lambda r: tf_params[f'{r}x{r}/FromRGB/bias'], r'b(\d+)\.conv(\d+)\.weight', lambda r, i: tf_params[f'{r}x{r}/Conv{i}{["","_down"][int(i)]}/weight'].transpose(3, 2, 0, 1), r'b(\d+)\.conv(\d+)\.bias', lambda r, i: tf_params[f'{r}x{r}/Conv{i}{["","_down"][int(i)]}/bias'], r'b(\d+)\.skip\.weight', lambda r: tf_params[f'{r}x{r}/Skip/weight'].transpose(3, 2, 0, 1), r'mapping\.embed\.weight', lambda: tf_params[f'LabelEmbed/weight'].transpose(), r'mapping\.embed\.bias', lambda: tf_params[f'LabelEmbed/bias'], r'mapping\.fc(\d+)\.weight', lambda i: tf_params[f'Mapping{i}/weight'].transpose(), r'mapping\.fc(\d+)\.bias', lambda i: tf_params[f'Mapping{i}/bias'], r'b4\.conv\.weight', lambda: tf_params[f'4x4/Conv/weight'].transpose(3, 2, 0, 1), r'b4\.conv\.bias', lambda: tf_params[f'4x4/Conv/bias'], r'b4\.fc\.weight', lambda: tf_params[f'4x4/Dense0/weight'].transpose(), r'b4\.fc\.bias', lambda: tf_params[f'4x4/Dense0/bias'], r'b4\.out\.weight', lambda: tf_params[f'Output/weight'].transpose(), r'b4\.out\.bias', lambda: tf_params[f'Output/bias'], r'.*\.resample_filter', None, ) return D #---------------------------------------------------------------------------- @click.command() @click.option('--source', help='Input pickle', required=True, metavar='PATH') @click.option('--dest', help='Output pickle', required=True, metavar='PATH') @click.option('--force-fp16', help='Force the networks to use FP16', type=bool, default=False, metavar='BOOL', show_default=True) def convert_network_pickle(source, dest, force_fp16): """Convert legacy network pickle into the native PyTorch format. The tool is able to load the main network configurations exported using the TensorFlow version of StyleGAN2 or StyleGAN2-ADA. It does not support e.g. StyleGAN2-ADA comparison methods, StyleGAN2 configs A-D, or StyleGAN1 networks. Example: \b python legacy.py \\ --source=https://nvlabs-fi-cdn.nvidia.com/stylegan2/networks/stylegan2-cat-config-f.pkl \\ --dest=stylegan2-cat-config-f.pkl """ print(f'Loading "{source}"...') with dnnlib.util.open_url(source) as f: data = load_network_pkl(f, force_fp16=force_fp16) print(f'Saving "{dest}"...') with open(dest, 'wb') as f: pickle.dump(data, f) print('Done.') #---------------------------------------------------------------------------- if __name__ == "__main__": convert_network_pickle() # pylint: disable=no-value-for-parameter #----------------------------------------------------------------------------
16,561
50.117284
154
py
DiffProxy
DiffProxy-main/stylegan/train.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Train a GAN using the techniques described in the paper "Alias-Free Generative Adversarial Networks".""" import os import click import re import json import tempfile import torch import dnnlib from training import training_loop_gan, training_loop_nogan from metrics import metric_main from torch_utils import training_stats from torch_utils import custom_ops #---------------------------------------------------------------------------- def subprocess_fn(rank, c, temp_dir, no_gan): dnnlib.util.Logger(file_name=os.path.join(c.run_dir, 'log.txt'), file_mode='a', should_flush=True) # Init torch.distributed. if c.num_gpus > 1: init_file = os.path.abspath(os.path.join(temp_dir, '.torch_distributed_init')) if os.name == 'nt': init_method = 'file:///' + init_file.replace('\\', '/') torch.distributed.init_process_group(backend='gloo', init_method=init_method, rank=rank, world_size=c.num_gpus) else: init_method = f'file://{init_file}' torch.distributed.init_process_group(backend='nccl', init_method=init_method, rank=rank, world_size=c.num_gpus) # Init torch_utils. sync_device = torch.device('cuda', rank) if c.num_gpus > 1 else None training_stats.init_multiprocessing(rank=rank, sync_device=sync_device) if rank != 0: custom_ops.verbosity = 'none' # Execute training loop. if no_gan: training_loop_nogan.training_loop(rank=rank, **c) else: training_loop_gan.training_loop(rank=rank, **c) #---------------------------------------------------------------------------- def launch_training(c, desc, outdir, dry_run, no_gan): dnnlib.util.Logger(should_flush=True) # Pick output directory. prev_run_dirs = [] if os.path.isdir(outdir): prev_run_dirs = [x for x in os.listdir(outdir) if os.path.isdir(os.path.join(outdir, x))] prev_run_ids = [re.match(r'^\d+', x) for x in prev_run_dirs] prev_run_ids = [int(x.group()) for x in prev_run_ids if x is not None] cur_run_id = max(prev_run_ids, default=-1) + 1 c.run_dir = os.path.join(outdir, f'{cur_run_id:05d}-{desc}') assert not os.path.exists(c.run_dir) # Print options. print() print('Training options:') print(json.dumps(c, indent=2)) print() print(f'Output directory: {c.run_dir}') print(f'Number of GPUs: {c.num_gpus}') print(f'Batch size: {c.batch_size} images') print(f'Training duration: {c.total_kimg} kimg') print(f'Dataset path: {c.training_set_kwargs.path}') print(f'Dataset size: {c.training_set_kwargs.max_size} images') print(f'Dataset resolution: {c.training_set_kwargs.resolution}') print(f'Dataset labels: {c.training_set_kwargs.use_labels}') print(f'Dataset x-flips: {c.training_set_kwargs.xflip}') print(f'Training by GAN loss: {not no_gan}') print() # Dry run? if dry_run: print('Dry run; exiting.') return # Create output directory. print('Creating output directory...') os.makedirs(c.run_dir) with open(os.path.join(c.run_dir, 'training_options.json'), 'wt') as f: json.dump(c, f, indent=2) # Launch processes. print('Launching processes...') torch.multiprocessing.set_start_method('spawn') with tempfile.TemporaryDirectory() as temp_dir: if c.num_gpus == 1: subprocess_fn(rank=0, c=c, temp_dir=temp_dir, no_gan=no_gan) else: torch.multiprocessing.spawn(fn=subprocess_fn, args=(c, temp_dir, no_gan), nprocs=c.num_gpus) #---------------------------------------------------------------------------- def init_dataset_kwargs(data): try: dataset_kwargs = dnnlib.EasyDict(class_name='training.dataset.ImageFolderDataset', path=data, use_labels=True, max_size=None, xflip=False) dataset_obj = dnnlib.util.construct_class_by_name(**dataset_kwargs) # Subclass of training.dataset.Dataset. dataset_kwargs.resolution = dataset_obj.resolution # Be explicit about resolution. dataset_kwargs.use_labels = dataset_obj.has_labels # Be explicit about labels. dataset_kwargs.max_size = len(dataset_obj) # Be explicit about dataset size. return dataset_kwargs, dataset_obj.name except IOError as err: raise click.ClickException(f'--data: {err}') #---------------------------------------------------------------------------- def parse_comma_separated_list(s): if isinstance(s, list): return s if s is None or s.lower() == 'none' or s == '': return [] return s.split(',') #---------------------------------------------------------------------------- @click.command() # Required. @click.option('--outdir', help='Where to save the results', metavar='DIR', required=True) @click.option('--cfg', help='Base configuration', type=click.Choice(['stylegan3-t', 'stylegan3-r', 'stylegan2']), required=True) @click.option('--data', help='Training data', metavar='[ZIP|DIR]', type=str, required=True) @click.option('--gpus', help='Number of GPUs to use', metavar='INT', type=click.IntRange(min=1), required=True) @click.option('--batch', help='Total batch size', metavar='INT', type=click.IntRange(min=1), required=True) @click.option('--gamma', help='R1 regularization weight', metavar='FLOAT', type=click.FloatRange(min=0), required=True) # Optional features. @click.option('--cond', help='Train conditional model', metavar='BOOL', type=bool, default=False, show_default=True) @click.option('--mirror', help='Enable dataset x-flips', metavar='BOOL', type=bool, default=False, show_default=True) @click.option('--aug', help='Augmentation mode', type=click.Choice(['noaug', 'ada', 'fixed']), default='ada', show_default=True) @click.option('--resume', help='Resume from given network pickle', metavar='[PATH|URL]', type=str) @click.option('--freezed', help='Freeze first layers of D', metavar='INT', type=click.IntRange(min=0), default=0, show_default=True) # Misc hyperparameters. @click.option('--p', help='Probability for --aug=fixed', metavar='FLOAT', type=click.FloatRange(min=0, max=1), default=0.2, show_default=True) @click.option('--target', help='Target value for --aug=ada', metavar='FLOAT', type=click.FloatRange(min=0, max=1), default=0.6, show_default=True) @click.option('--batch-gpu', help='Limit batch size per GPU', metavar='INT', type=click.IntRange(min=1)) @click.option('--cbase', help='Capacity multiplier', metavar='INT', type=click.IntRange(min=1), default=32768, show_default=True) @click.option('--cmax', help='Max. feature maps', metavar='INT', type=click.IntRange(min=1), default=512, show_default=True) @click.option('--glr', help='G learning rate [default: varies]', metavar='FLOAT', type=click.FloatRange(min=0)) @click.option('--dlr', help='D learning rate', metavar='FLOAT', type=click.FloatRange(min=0), default=0.002, show_default=True) @click.option('--map-depth', help='Mapping network depth [default: varies]', metavar='INT', type=click.IntRange(min=1)) @click.option('--mbstd-group', help='Minibatch std group size', metavar='INT', type=click.IntRange(min=1), default=4, show_default=True) # Misc settings. @click.option('--desc', help='String to include in result dir name', metavar='STR', type=str) @click.option('--metrics', help='Quality metrics', metavar='[NAME|A,B,C|none]', type=parse_comma_separated_list, default='fid50k_full', show_default=True) @click.option('--kimg', help='Total training duration', metavar='KIMG', type=click.IntRange(min=1), default=25000, show_default=True) @click.option('--tick', help='How often to print progress', metavar='KIMG', type=click.IntRange(min=1), default=4, show_default=True) @click.option('--snap', help='How often to save snapshots', metavar='TICKS', type=click.IntRange(min=1), default=50, show_default=True) @click.option('--seed', help='Random seed', metavar='INT', type=click.IntRange(min=0), default=0, show_default=True) @click.option('--fp32', help='Disable mixed-precision', metavar='BOOL', type=bool, default=False, show_default=True) @click.option('--nobench', help='Disable cuDNN benchmarking', metavar='BOOL', type=bool, default=False, show_default=True) @click.option('--workers', help='DataLoader worker processes', metavar='INT', type=click.IntRange(min=1), default=3, show_default=True) @click.option('-n','--dry-run', help='Print training options and exit', is_flag=True) # additional @click.option('--generator_name', help='Specify the name of generator', metavar='STR', type=str, required=True) @click.option('--norm_type', help='Normalization performed on parameters', metavar='STR', type=str, required=True) @click.option('--no_gan', help='If not use GAN loss', metavar='BOOL', type=bool, required=True) @click.option('--cond_d', help='If use Conditional Discriminator', metavar='BOOL', type=bool, required=True) def main(**kwargs): """Train a GAN using the techniques described in the paper "Alias-Free Generative Adversarial Networks". Examples: \b # Train StyleGAN3-T for AFHQv2 using 8 GPUs. python train.py --outdir=~/training-runs --cfg=stylegan3-t --data=~/datasets/afhqv2-512x512.zip \\ --gpus=8 --batch=32 --gamma=8.2 --mirror=1 \b # Fine-tune StyleGAN3-R for MetFaces-U using 1 GPU, starting from the pre-trained FFHQ-U pickle. python train.py --outdir=~/training-runs --cfg=stylegan3-r --data=~/datasets/metfacesu-1024x1024.zip \\ --gpus=8 --batch=32 --gamma=6.6 --mirror=1 --kimg=5000 --snap=5 \\ --resume=https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-ffhqu-1024x1024.pkl \b # Train StyleGAN2 for FFHQ at 1024x1024 resolution using 8 GPUs. python train.py --outdir=~/training-runs --cfg=stylegan2 --data=~/datasets/ffhq-1024x1024.zip \\ --gpus=8 --batch=32 --gamma=10 --mirror=1 --aug=noaug """ # size of z z_dim = 0 # 512 print(f"z_dim = {z_dim}") # Initialize config. opts = dnnlib.EasyDict(kwargs) # Command line arguments. c = dnnlib.EasyDict() # Main config dict. c.G_kwargs = dnnlib.EasyDict(class_name=None, z_dim=z_dim, w_dim=512, mapping_kwargs=dnnlib.EasyDict()) c.D_kwargs = dnnlib.EasyDict(class_name='training.networks_stylegan2.Discriminator', block_kwargs=dnnlib.EasyDict(), mapping_kwargs=dnnlib.EasyDict(), epilogue_kwargs=dnnlib.EasyDict()) c.G_opt_kwargs = dnnlib.EasyDict(class_name='torch.optim.Adam', betas=[0,0.99], eps=1e-8) c.D_opt_kwargs = dnnlib.EasyDict(class_name='torch.optim.Adam', betas=[0,0.99], eps=1e-8) c.loss_kwargs = dnnlib.EasyDict(class_name='loss.ProxyLoss') if opts.no_gan else dnnlib.EasyDict(class_name='loss.ProxyGANLoss') c.data_loader_kwargs = dnnlib.EasyDict(pin_memory=True, prefetch_factor=2) # customized configs c.generator_name = opts.generator_name c.norm_type = opts.norm_type if not opts.no_gan: c.cond_D = opts.cond_d # Training set. c.training_set_kwargs, dataset_name = init_dataset_kwargs(data=opts.data) if opts.cond and not c.training_set_kwargs.use_labels: raise click.ClickException('--cond=True requires labels specified in dataset.json') c.training_set_kwargs.use_labels = opts.cond c.training_set_kwargs.xflip = opts.mirror # Hyperparameters & settings. c.num_gpus = opts.gpus c.batch_size = opts.batch c.batch_gpu = opts.batch_gpu or opts.batch // opts.gpus c.G_kwargs.channel_base = c.D_kwargs.channel_base = opts.cbase c.G_kwargs.channel_max = c.D_kwargs.channel_max = opts.cmax c.G_kwargs.mapping_kwargs.num_layers = (8 if opts.cfg == 'stylegan2' else 2) if opts.map_depth is None else opts.map_depth c.D_kwargs.block_kwargs.freeze_layers = opts.freezed c.D_kwargs.epilogue_kwargs.mbstd_group_size = opts.mbstd_group c.loss_kwargs.r1_gamma = opts.gamma c.G_opt_kwargs.lr = (0.002 if opts.cfg == 'stylegan2' else 0.0025) if opts.glr is None else opts.glr c.D_opt_kwargs.lr = opts.dlr c.metrics = opts.metrics c.total_kimg = opts.kimg c.kimg_per_tick = opts.tick c.image_snapshot_ticks = c.network_snapshot_ticks = opts.snap c.random_seed = c.training_set_kwargs.random_seed = opts.seed c.data_loader_kwargs.num_workers = opts.workers # Sanity checks. if c.batch_size % c.num_gpus != 0: raise click.ClickException('--batch must be a multiple of --gpus') if c.batch_size % (c.num_gpus * c.batch_gpu) != 0: raise click.ClickException('--batch must be a multiple of --gpus times --batch-gpu') if c.batch_gpu < c.D_kwargs.epilogue_kwargs.mbstd_group_size: raise click.ClickException('--batch-gpu cannot be smaller than --mbstd') if any(not metric_main.is_valid_metric(metric) for metric in c.metrics): raise click.ClickException('\n'.join(['--metrics can only contain the following values:'] + metric_main.list_valid_metrics())) # Base configuration. c.ema_kimg = c.batch_size * 10 / 32 if opts.cfg == 'stylegan2': c.G_kwargs.class_name = 'training.networks_stylegan2.Generator' c.loss_kwargs.style_mixing_prob = 0.9 # Enable style mixing regularization. c.loss_kwargs.pl_weight = 2 # Enable path length regularization. c.G_reg_interval = 4 # Enable lazy regularization for G. c.G_kwargs.fused_modconv_default = 'inference_only' # Speed up training by using regular convolutions instead of grouped convolutions. c.loss_kwargs.pl_no_weight_grad = True # Speed up path length regularization by skipping gradient computation wrt. conv2d weights. else: c.G_kwargs.class_name = 'training.networks_stylegan3.Generator' c.G_kwargs.magnitude_ema_beta = 0.5 ** (c.batch_size / (20 * 1e3)) if opts.cfg == 'stylegan3-r': c.G_kwargs.conv_kernel = 1 # Use 1x1 convolutions. c.G_kwargs.channel_base *= 2 # Double the number of feature maps. c.G_kwargs.channel_max *= 2 c.G_kwargs.use_radial_filters = True # Use radially symmetric downsampling filters. c.loss_kwargs.blur_init_sigma = 10 # Blur the images seen by the discriminator. c.loss_kwargs.blur_fade_kimg = c.batch_size * 200 / 32 # Fade out the blur during the first N kimg. # Augmentation. if opts.aug != 'noaug': c.augment_kwargs = dnnlib.EasyDict(class_name='training.augment.AugmentPipe', xflip=1, rotate90=1, xint=1, scale=1, rotate=1, aniso=1, xfrac=1, brightness=1, contrast=1, lumaflip=1, hue=1, saturation=1) if opts.aug == 'ada': c.ada_target = opts.target if opts.aug == 'fixed': c.augment_p = opts.p # Resume. if opts.resume is not None: c.resume_pkl = opts.resume c.ada_kimg = 100 # Make ADA react faster at the beginning. c.ema_rampup = None # Disable EMA rampup. c.loss_kwargs.blur_init_sigma = 0 # Disable blur rampup. # Performance-related toggles. if opts.fp32: c.G_kwargs.num_fp16_res = c.D_kwargs.num_fp16_res = 0 c.G_kwargs.conv_clamp = c.D_kwargs.conv_clamp = None if opts.nobench: c.cudnn_benchmark = False # Description string. if opts.no_gan: desc = f'{opts.cfg:s}-{dataset_name:s}-gpus{c.num_gpus:d}-batch{c.batch_size:d}-no_gan' else: desc = f'{opts.cfg:s}-{dataset_name:s}-gpus{c.num_gpus:d}-batch{c.batch_size:d}-gamma{c.loss_kwargs.r1_gamma:g}' if opts.desc is not None: desc += f'-{opts.desc}' # Launch. launch_training(c=c, desc=desc, outdir=opts.outdir, dry_run=opts.dry_run, no_gan=opts.no_gan) #---------------------------------------------------------------------------- if __name__ == "__main__": main() # pylint: disable=no-value-for-parameter #----------------------------------------------------------------------------
16,998
53.483974
210
py
DiffProxy
DiffProxy-main/stylegan/training/loss.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Loss functions.""" import numpy as np import torch from torch_utils import training_stats from torch_utils.ops import conv2d_gradfix from torch_utils.ops import upfirdn2d #---------------------------------------------------------------------------- class Loss: def accumulate_gradients(self, phase, real_img, real_c, gen_z, gen_c, gain, cur_nimg): # to be overridden by subclass raise NotImplementedError() #---------------------------------------------------------------------------- class StyleGAN2Loss(Loss): def __init__(self, device, G, D, augment_pipe=None, r1_gamma=10, style_mixing_prob=0, pl_weight=0, pl_batch_shrink=2, pl_decay=0.01, pl_no_weight_grad=False, blur_init_sigma=0, blur_fade_kimg=0): super().__init__() self.device = device self.G = G self.D = D self.augment_pipe = augment_pipe self.r1_gamma = r1_gamma self.style_mixing_prob = style_mixing_prob self.pl_weight = pl_weight self.pl_batch_shrink = pl_batch_shrink self.pl_decay = pl_decay self.pl_no_weight_grad = pl_no_weight_grad self.pl_mean = torch.zeros([], device=device) self.blur_init_sigma = blur_init_sigma self.blur_fade_kimg = blur_fade_kimg def run_G(self, z, c, update_emas=False): ws = self.G.mapping(z, c, update_emas=update_emas) if self.style_mixing_prob > 0: with torch.autograd.profiler.record_function('style_mixing'): cutoff = torch.empty([], dtype=torch.int64, device=ws.device).random_(1, ws.shape[1]) cutoff = torch.where(torch.rand([], device=ws.device) < self.style_mixing_prob, cutoff, torch.full_like(cutoff, ws.shape[1])) ws[:, cutoff:] = self.G.mapping(torch.randn_like(z), c, update_emas=False)[:, cutoff:] img = self.G.synthesis(ws, update_emas=update_emas) return img, ws def run_D(self, img, c, blur_sigma=0, update_emas=False): blur_size = np.floor(blur_sigma * 3) if blur_size > 0: with torch.autograd.profiler.record_function('blur'): f = torch.arange(-blur_size, blur_size + 1, device=img.device).div(blur_sigma).square().neg().exp2() img = upfirdn2d.filter2d(img, f / f.sum()) if self.augment_pipe is not None: img = self.augment_pipe(img) logits = self.D(img, c, update_emas=update_emas) return logits def accumulate_gradients(self, phase, real_img, real_c, gen_z, gen_c, gain, cur_nimg): assert phase in ['Gmain', 'Greg', 'Gboth', 'Dmain', 'Dreg', 'Dboth'] if self.pl_weight == 0: phase = {'Greg': 'none', 'Gboth': 'Gmain'}.get(phase, phase) if self.r1_gamma == 0: phase = {'Dreg': 'none', 'Dboth': 'Dmain'}.get(phase, phase) blur_sigma = max(1 - cur_nimg / (self.blur_fade_kimg * 1e3), 0) * self.blur_init_sigma if self.blur_fade_kimg > 0 else 0 # Gmain: Maximize logits for generated images. if phase in ['Gmain', 'Gboth']: with torch.autograd.profiler.record_function('Gmain_forward'): gen_img, _gen_ws = self.run_G(gen_z, gen_c) gen_logits = self.run_D(gen_img, gen_c, blur_sigma=blur_sigma) training_stats.report('Loss/scores/fake', gen_logits) training_stats.report('Loss/signs/fake', gen_logits.sign()) loss_Gmain = torch.nn.functional.softplus(-gen_logits) # -log(sigmoid(gen_logits)) training_stats.report('Loss/G/loss', loss_Gmain) with torch.autograd.profiler.record_function('Gmain_backward'): loss_Gmain.mean().mul(gain).backward() # Gpl: Apply path length regularization. if phase in ['Greg', 'Gboth']: with torch.autograd.profiler.record_function('Gpl_forward'): batch_size = gen_z.shape[0] // self.pl_batch_shrink gen_img, gen_ws = self.run_G(gen_z[:batch_size], gen_c[:batch_size]) pl_noise = torch.randn_like(gen_img) / np.sqrt(gen_img.shape[2] * gen_img.shape[3]) with torch.autograd.profiler.record_function('pl_grads'), conv2d_gradfix.no_weight_gradients(self.pl_no_weight_grad): pl_grads = torch.autograd.grad(outputs=[(gen_img * pl_noise).sum()], inputs=[gen_ws], create_graph=True, only_inputs=True)[0] pl_lengths = pl_grads.square().sum(2).mean(1).sqrt() pl_mean = self.pl_mean.lerp(pl_lengths.mean(), self.pl_decay) self.pl_mean.copy_(pl_mean.detach()) pl_penalty = (pl_lengths - pl_mean).square() training_stats.report('Loss/pl_penalty', pl_penalty) loss_Gpl = pl_penalty * self.pl_weight training_stats.report('Loss/G/reg', loss_Gpl) with torch.autograd.profiler.record_function('Gpl_backward'): loss_Gpl.mean().mul(gain).backward() # Dmain: Minimize logits for generated images. loss_Dgen = 0 if phase in ['Dmain', 'Dboth']: with torch.autograd.profiler.record_function('Dgen_forward'): gen_img, _gen_ws = self.run_G(gen_z, gen_c, update_emas=True) gen_logits = self.run_D(gen_img, gen_c, blur_sigma=blur_sigma, update_emas=True) training_stats.report('Loss/scores/fake', gen_logits) training_stats.report('Loss/signs/fake', gen_logits.sign()) loss_Dgen = torch.nn.functional.softplus(gen_logits) # -log(1 - sigmoid(gen_logits)) with torch.autograd.profiler.record_function('Dgen_backward'): loss_Dgen.mean().mul(gain).backward() # Dmain: Maximize logits for real images. # Dr1: Apply R1 regularization. if phase in ['Dmain', 'Dreg', 'Dboth']: name = 'Dreal' if phase == 'Dmain' else 'Dr1' if phase == 'Dreg' else 'Dreal_Dr1' with torch.autograd.profiler.record_function(name + '_forward'): real_img_tmp = real_img.detach().requires_grad_(phase in ['Dreg', 'Dboth']) real_logits = self.run_D(real_img_tmp, real_c, blur_sigma=blur_sigma) training_stats.report('Loss/scores/real', real_logits) training_stats.report('Loss/signs/real', real_logits.sign()) loss_Dreal = 0 if phase in ['Dmain', 'Dboth']: loss_Dreal = torch.nn.functional.softplus(-real_logits) # -log(sigmoid(real_logits)) training_stats.report('Loss/D/loss', loss_Dgen + loss_Dreal) loss_Dr1 = 0 if phase in ['Dreg', 'Dboth']: with torch.autograd.profiler.record_function('r1_grads'), conv2d_gradfix.no_weight_gradients(): r1_grads = torch.autograd.grad(outputs=[real_logits.sum()], inputs=[real_img_tmp], create_graph=True, only_inputs=True)[0] r1_penalty = r1_grads.square().sum([1,2,3]) loss_Dr1 = r1_penalty * (self.r1_gamma / 2) training_stats.report('Loss/r1_penalty', r1_penalty) training_stats.report('Loss/D/reg', loss_Dr1) with torch.autograd.profiler.record_function(name + '_backward'): (loss_Dreal + loss_Dr1).mean().mul(gain).backward() #----------------------------------------------------------------------------
7,998
56.135714
199
py
DiffProxy
DiffProxy-main/stylegan/training/augment.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Augmentation pipeline from the paper "Training Generative Adversarial Networks with Limited Data". Matches the original implementation by Karras et al. at https://github.com/NVlabs/stylegan2-ada/blob/main/training/augment.py""" import numpy as np import scipy.signal import torch from torch_utils import persistence from torch_utils import misc from torch_utils.ops import upfirdn2d from torch_utils.ops import grid_sample_gradfix from torch_utils.ops import conv2d_gradfix #---------------------------------------------------------------------------- # Coefficients of various wavelet decomposition low-pass filters. wavelets = { 'haar': [0.7071067811865476, 0.7071067811865476], 'db1': [0.7071067811865476, 0.7071067811865476], 'db2': [-0.12940952255092145, 0.22414386804185735, 0.836516303737469, 0.48296291314469025], 'db3': [0.035226291882100656, -0.08544127388224149, -0.13501102001039084, 0.4598775021193313, 0.8068915093133388, 0.3326705529509569], 'db4': [-0.010597401784997278, 0.032883011666982945, 0.030841381835986965, -0.18703481171888114, -0.02798376941698385, 0.6308807679295904, 0.7148465705525415, 0.23037781330885523], 'db5': [0.003335725285001549, -0.012580751999015526, -0.006241490213011705, 0.07757149384006515, -0.03224486958502952, -0.24229488706619015, 0.13842814590110342, 0.7243085284385744, 0.6038292697974729, 0.160102397974125], 'db6': [-0.00107730108499558, 0.004777257511010651, 0.0005538422009938016, -0.031582039318031156, 0.02752286553001629, 0.09750160558707936, -0.12976686756709563, -0.22626469396516913, 0.3152503517092432, 0.7511339080215775, 0.4946238903983854, 0.11154074335008017], 'db7': [0.0003537138000010399, -0.0018016407039998328, 0.00042957797300470274, 0.012550998556013784, -0.01657454163101562, -0.03802993693503463, 0.0806126091510659, 0.07130921926705004, -0.22403618499416572, -0.14390600392910627, 0.4697822874053586, 0.7291320908465551, 0.39653931948230575, 0.07785205408506236], 'db8': [-0.00011747678400228192, 0.0006754494059985568, -0.0003917403729959771, -0.00487035299301066, 0.008746094047015655, 0.013981027917015516, -0.04408825393106472, -0.01736930100202211, 0.128747426620186, 0.00047248457399797254, -0.2840155429624281, -0.015829105256023893, 0.5853546836548691, 0.6756307362980128, 0.3128715909144659, 0.05441584224308161], 'sym2': [-0.12940952255092145, 0.22414386804185735, 0.836516303737469, 0.48296291314469025], 'sym3': [0.035226291882100656, -0.08544127388224149, -0.13501102001039084, 0.4598775021193313, 0.8068915093133388, 0.3326705529509569], 'sym4': [-0.07576571478927333, -0.02963552764599851, 0.49761866763201545, 0.8037387518059161, 0.29785779560527736, -0.09921954357684722, -0.012603967262037833, 0.0322231006040427], 'sym5': [0.027333068345077982, 0.029519490925774643, -0.039134249302383094, 0.1993975339773936, 0.7234076904024206, 0.6339789634582119, 0.01660210576452232, -0.17532808990845047, -0.021101834024758855, 0.019538882735286728], 'sym6': [0.015404109327027373, 0.0034907120842174702, -0.11799011114819057, -0.048311742585633, 0.4910559419267466, 0.787641141030194, 0.3379294217276218, -0.07263752278646252, -0.021060292512300564, 0.04472490177066578, 0.0017677118642428036, -0.007800708325034148], 'sym7': [0.002681814568257878, -0.0010473848886829163, -0.01263630340325193, 0.03051551316596357, 0.0678926935013727, -0.049552834937127255, 0.017441255086855827, 0.5361019170917628, 0.767764317003164, 0.2886296317515146, -0.14004724044296152, -0.10780823770381774, 0.004010244871533663, 0.010268176708511255], 'sym8': [-0.0033824159510061256, -0.0005421323317911481, 0.03169508781149298, 0.007607487324917605, -0.1432942383508097, -0.061273359067658524, 0.4813596512583722, 0.7771857517005235, 0.3644418948353314, -0.05194583810770904, -0.027219029917056003, 0.049137179673607506, 0.003808752013890615, -0.01495225833704823, -0.0003029205147213668, 0.0018899503327594609], } #---------------------------------------------------------------------------- # Helpers for constructing transformation matrices. def matrix(*rows, device=None): assert all(len(row) == len(rows[0]) for row in rows) elems = [x for row in rows for x in row] ref = [x for x in elems if isinstance(x, torch.Tensor)] if len(ref) == 0: return misc.constant(np.asarray(rows), device=device) assert device is None or device == ref[0].device elems = [x if isinstance(x, torch.Tensor) else misc.constant(x, shape=ref[0].shape, device=ref[0].device) for x in elems] return torch.stack(elems, dim=-1).reshape(ref[0].shape + (len(rows), -1)) def translate2d(tx, ty, **kwargs): return matrix( [1, 0, tx], [0, 1, ty], [0, 0, 1], **kwargs) def translate3d(tx, ty, tz, **kwargs): return matrix( [1, 0, 0, tx], [0, 1, 0, ty], [0, 0, 1, tz], [0, 0, 0, 1], **kwargs) def scale2d(sx, sy, **kwargs): return matrix( [sx, 0, 0], [0, sy, 0], [0, 0, 1], **kwargs) def scale3d(sx, sy, sz, **kwargs): return matrix( [sx, 0, 0, 0], [0, sy, 0, 0], [0, 0, sz, 0], [0, 0, 0, 1], **kwargs) def rotate2d(theta, **kwargs): return matrix( [torch.cos(theta), torch.sin(-theta), 0], [torch.sin(theta), torch.cos(theta), 0], [0, 0, 1], **kwargs) def rotate3d(v, theta, **kwargs): vx = v[..., 0]; vy = v[..., 1]; vz = v[..., 2] s = torch.sin(theta); c = torch.cos(theta); cc = 1 - c return matrix( [vx*vx*cc+c, vx*vy*cc-vz*s, vx*vz*cc+vy*s, 0], [vy*vx*cc+vz*s, vy*vy*cc+c, vy*vz*cc-vx*s, 0], [vz*vx*cc-vy*s, vz*vy*cc+vx*s, vz*vz*cc+c, 0], [0, 0, 0, 1], **kwargs) def translate2d_inv(tx, ty, **kwargs): return translate2d(-tx, -ty, **kwargs) def scale2d_inv(sx, sy, **kwargs): return scale2d(1 / sx, 1 / sy, **kwargs) def rotate2d_inv(theta, **kwargs): return rotate2d(-theta, **kwargs) #---------------------------------------------------------------------------- # Versatile image augmentation pipeline from the paper # "Training Generative Adversarial Networks with Limited Data". # # All augmentations are disabled by default; individual augmentations can # be enabled by setting their probability multipliers to 1. @persistence.persistent_class class AugmentPipe(torch.nn.Module): def __init__(self, xflip=0, rotate90=0, xint=0, xint_max=0.125, scale=0, rotate=0, aniso=0, xfrac=0, scale_std=0.2, rotate_max=1, aniso_std=0.2, xfrac_std=0.125, brightness=0, contrast=0, lumaflip=0, hue=0, saturation=0, brightness_std=0.2, contrast_std=0.5, hue_max=1, saturation_std=1, imgfilter=0, imgfilter_bands=[1,1,1,1], imgfilter_std=1, noise=0, cutout=0, noise_std=0.1, cutout_size=0.5, ): super().__init__() self.register_buffer('p', torch.ones([])) # Overall multiplier for augmentation probability. # Pixel blitting. self.xflip = float(xflip) # Probability multiplier for x-flip. self.rotate90 = float(rotate90) # Probability multiplier for 90 degree rotations. self.xint = float(xint) # Probability multiplier for integer translation. self.xint_max = float(xint_max) # Range of integer translation, relative to image dimensions. # General geometric transformations. self.scale = float(scale) # Probability multiplier for isotropic scaling. self.rotate = float(rotate) # Probability multiplier for arbitrary rotation. self.aniso = float(aniso) # Probability multiplier for anisotropic scaling. self.xfrac = float(xfrac) # Probability multiplier for fractional translation. self.scale_std = float(scale_std) # Log2 standard deviation of isotropic scaling. self.rotate_max = float(rotate_max) # Range of arbitrary rotation, 1 = full circle. self.aniso_std = float(aniso_std) # Log2 standard deviation of anisotropic scaling. self.xfrac_std = float(xfrac_std) # Standard deviation of frational translation, relative to image dimensions. # Color transformations. self.brightness = float(brightness) # Probability multiplier for brightness. self.contrast = float(contrast) # Probability multiplier for contrast. self.lumaflip = float(lumaflip) # Probability multiplier for luma flip. self.hue = float(hue) # Probability multiplier for hue rotation. self.saturation = float(saturation) # Probability multiplier for saturation. self.brightness_std = float(brightness_std) # Standard deviation of brightness. self.contrast_std = float(contrast_std) # Log2 standard deviation of contrast. self.hue_max = float(hue_max) # Range of hue rotation, 1 = full circle. self.saturation_std = float(saturation_std) # Log2 standard deviation of saturation. # Image-space filtering. self.imgfilter = float(imgfilter) # Probability multiplier for image-space filtering. self.imgfilter_bands = list(imgfilter_bands) # Probability multipliers for individual frequency bands. self.imgfilter_std = float(imgfilter_std) # Log2 standard deviation of image-space filter amplification. # Image-space corruptions. self.noise = float(noise) # Probability multiplier for additive RGB noise. self.cutout = float(cutout) # Probability multiplier for cutout. self.noise_std = float(noise_std) # Standard deviation of additive RGB noise. self.cutout_size = float(cutout_size) # Size of the cutout rectangle, relative to image dimensions. # Setup orthogonal lowpass filter for geometric augmentations. self.register_buffer('Hz_geom', upfirdn2d.setup_filter(wavelets['sym6'])) # Construct filter bank for image-space filtering. Hz_lo = np.asarray(wavelets['sym2']) # H(z) Hz_hi = Hz_lo * ((-1) ** np.arange(Hz_lo.size)) # H(-z) Hz_lo2 = np.convolve(Hz_lo, Hz_lo[::-1]) / 2 # H(z) * H(z^-1) / 2 Hz_hi2 = np.convolve(Hz_hi, Hz_hi[::-1]) / 2 # H(-z) * H(-z^-1) / 2 Hz_fbank = np.eye(4, 1) # Bandpass(H(z), b_i) for i in range(1, Hz_fbank.shape[0]): Hz_fbank = np.dstack([Hz_fbank, np.zeros_like(Hz_fbank)]).reshape(Hz_fbank.shape[0], -1)[:, :-1] Hz_fbank = scipy.signal.convolve(Hz_fbank, [Hz_lo2]) Hz_fbank[i, (Hz_fbank.shape[1] - Hz_hi2.size) // 2 : (Hz_fbank.shape[1] + Hz_hi2.size) // 2] += Hz_hi2 self.register_buffer('Hz_fbank', torch.as_tensor(Hz_fbank, dtype=torch.float32)) def forward(self, images, debug_percentile=None): assert isinstance(images, torch.Tensor) and images.ndim == 4 batch_size, num_channels, height, width = images.shape device = images.device if debug_percentile is not None: debug_percentile = torch.as_tensor(debug_percentile, dtype=torch.float32, device=device) # ------------------------------------- # Select parameters for pixel blitting. # ------------------------------------- # Initialize inverse homogeneous 2D transform: G_inv @ pixel_out ==> pixel_in I_3 = torch.eye(3, device=device) G_inv = I_3 # Apply x-flip with probability (xflip * strength). if self.xflip > 0: i = torch.floor(torch.rand([batch_size], device=device) * 2) i = torch.where(torch.rand([batch_size], device=device) < self.xflip * self.p, i, torch.zeros_like(i)) if debug_percentile is not None: i = torch.full_like(i, torch.floor(debug_percentile * 2)) G_inv = G_inv @ scale2d_inv(1 - 2 * i, 1) # Apply 90 degree rotations with probability (rotate90 * strength). if self.rotate90 > 0: i = torch.floor(torch.rand([batch_size], device=device) * 4) i = torch.where(torch.rand([batch_size], device=device) < self.rotate90 * self.p, i, torch.zeros_like(i)) if debug_percentile is not None: i = torch.full_like(i, torch.floor(debug_percentile * 4)) G_inv = G_inv @ rotate2d_inv(-np.pi / 2 * i) # Apply integer translation with probability (xint * strength). if self.xint > 0: t = (torch.rand([batch_size, 2], device=device) * 2 - 1) * self.xint_max t = torch.where(torch.rand([batch_size, 1], device=device) < self.xint * self.p, t, torch.zeros_like(t)) if debug_percentile is not None: t = torch.full_like(t, (debug_percentile * 2 - 1) * self.xint_max) G_inv = G_inv @ translate2d_inv(torch.round(t[:,0] * width), torch.round(t[:,1] * height)) # -------------------------------------------------------- # Select parameters for general geometric transformations. # -------------------------------------------------------- # Apply isotropic scaling with probability (scale * strength). if self.scale > 0: s = torch.exp2(torch.randn([batch_size], device=device) * self.scale_std) s = torch.where(torch.rand([batch_size], device=device) < self.scale * self.p, s, torch.ones_like(s)) if debug_percentile is not None: s = torch.full_like(s, torch.exp2(torch.erfinv(debug_percentile * 2 - 1) * self.scale_std)) G_inv = G_inv @ scale2d_inv(s, s) # Apply pre-rotation with probability p_rot. p_rot = 1 - torch.sqrt((1 - self.rotate * self.p).clamp(0, 1)) # P(pre OR post) = p if self.rotate > 0: theta = (torch.rand([batch_size], device=device) * 2 - 1) * np.pi * self.rotate_max theta = torch.where(torch.rand([batch_size], device=device) < p_rot, theta, torch.zeros_like(theta)) if debug_percentile is not None: theta = torch.full_like(theta, (debug_percentile * 2 - 1) * np.pi * self.rotate_max) G_inv = G_inv @ rotate2d_inv(-theta) # Before anisotropic scaling. # Apply anisotropic scaling with probability (aniso * strength). if self.aniso > 0: s = torch.exp2(torch.randn([batch_size], device=device) * self.aniso_std) s = torch.where(torch.rand([batch_size], device=device) < self.aniso * self.p, s, torch.ones_like(s)) if debug_percentile is not None: s = torch.full_like(s, torch.exp2(torch.erfinv(debug_percentile * 2 - 1) * self.aniso_std)) G_inv = G_inv @ scale2d_inv(s, 1 / s) # Apply post-rotation with probability p_rot. if self.rotate > 0: theta = (torch.rand([batch_size], device=device) * 2 - 1) * np.pi * self.rotate_max theta = torch.where(torch.rand([batch_size], device=device) < p_rot, theta, torch.zeros_like(theta)) if debug_percentile is not None: theta = torch.zeros_like(theta) G_inv = G_inv @ rotate2d_inv(-theta) # After anisotropic scaling. # Apply fractional translation with probability (xfrac * strength). if self.xfrac > 0: t = torch.randn([batch_size, 2], device=device) * self.xfrac_std t = torch.where(torch.rand([batch_size, 1], device=device) < self.xfrac * self.p, t, torch.zeros_like(t)) if debug_percentile is not None: t = torch.full_like(t, torch.erfinv(debug_percentile * 2 - 1) * self.xfrac_std) G_inv = G_inv @ translate2d_inv(t[:,0] * width, t[:,1] * height) # ---------------------------------- # Execute geometric transformations. # ---------------------------------- # Execute if the transform is not identity. if G_inv is not I_3: # Calculate padding. cx = (width - 1) / 2 cy = (height - 1) / 2 cp = matrix([-cx, -cy, 1], [cx, -cy, 1], [cx, cy, 1], [-cx, cy, 1], device=device) # [idx, xyz] cp = G_inv @ cp.t() # [batch, xyz, idx] Hz_pad = self.Hz_geom.shape[0] // 4 margin = cp[:, :2, :].permute(1, 0, 2).flatten(1) # [xy, batch * idx] margin = torch.cat([-margin, margin]).max(dim=1).values # [x0, y0, x1, y1] margin = margin + misc.constant([Hz_pad * 2 - cx, Hz_pad * 2 - cy] * 2, device=device) margin = margin.max(misc.constant([0, 0] * 2, device=device)) margin = margin.min(misc.constant([width-1, height-1] * 2, device=device)) mx0, my0, mx1, my1 = margin.ceil().to(torch.int32) # Pad image and adjust origin. images = torch.nn.functional.pad(input=images, pad=[mx0,mx1,my0,my1], mode='reflect') G_inv = translate2d((mx0 - mx1) / 2, (my0 - my1) / 2) @ G_inv # Upsample. images = upfirdn2d.upsample2d(x=images, f=self.Hz_geom, up=2) G_inv = scale2d(2, 2, device=device) @ G_inv @ scale2d_inv(2, 2, device=device) G_inv = translate2d(-0.5, -0.5, device=device) @ G_inv @ translate2d_inv(-0.5, -0.5, device=device) # Execute transformation. shape = [batch_size, num_channels, (height + Hz_pad * 2) * 2, (width + Hz_pad * 2) * 2] G_inv = scale2d(2 / images.shape[3], 2 / images.shape[2], device=device) @ G_inv @ scale2d_inv(2 / shape[3], 2 / shape[2], device=device) grid = torch.nn.functional.affine_grid(theta=G_inv[:,:2,:], size=shape, align_corners=False) images = grid_sample_gradfix.grid_sample(images, grid) # Downsample and crop. images = upfirdn2d.downsample2d(x=images, f=self.Hz_geom, down=2, padding=-Hz_pad*2, flip_filter=True) # -------------------------------------------- # Select parameters for color transformations. # -------------------------------------------- # Initialize homogeneous 3D transformation matrix: C @ color_in ==> color_out I_4 = torch.eye(4, device=device) C = I_4 # Apply brightness with probability (brightness * strength). if self.brightness > 0: b = torch.randn([batch_size], device=device) * self.brightness_std b = torch.where(torch.rand([batch_size], device=device) < self.brightness * self.p, b, torch.zeros_like(b)) if debug_percentile is not None: b = torch.full_like(b, torch.erfinv(debug_percentile * 2 - 1) * self.brightness_std) C = translate3d(b, b, b) @ C # Apply contrast with probability (contrast * strength). if self.contrast > 0: c = torch.exp2(torch.randn([batch_size], device=device) * self.contrast_std) c = torch.where(torch.rand([batch_size], device=device) < self.contrast * self.p, c, torch.ones_like(c)) if debug_percentile is not None: c = torch.full_like(c, torch.exp2(torch.erfinv(debug_percentile * 2 - 1) * self.contrast_std)) C = scale3d(c, c, c) @ C # Apply luma flip with probability (lumaflip * strength). v = misc.constant(np.asarray([1, 1, 1, 0]) / np.sqrt(3), device=device) # Luma axis. if self.lumaflip > 0: i = torch.floor(torch.rand([batch_size, 1, 1], device=device) * 2) i = torch.where(torch.rand([batch_size, 1, 1], device=device) < self.lumaflip * self.p, i, torch.zeros_like(i)) if debug_percentile is not None: i = torch.full_like(i, torch.floor(debug_percentile * 2)) C = (I_4 - 2 * v.ger(v) * i) @ C # Householder reflection. # Apply hue rotation with probability (hue * strength). if self.hue > 0 and num_channels > 1: theta = (torch.rand([batch_size], device=device) * 2 - 1) * np.pi * self.hue_max theta = torch.where(torch.rand([batch_size], device=device) < self.hue * self.p, theta, torch.zeros_like(theta)) if debug_percentile is not None: theta = torch.full_like(theta, (debug_percentile * 2 - 1) * np.pi * self.hue_max) C = rotate3d(v, theta) @ C # Rotate around v. # Apply saturation with probability (saturation * strength). if self.saturation > 0 and num_channels > 1: s = torch.exp2(torch.randn([batch_size, 1, 1], device=device) * self.saturation_std) s = torch.where(torch.rand([batch_size, 1, 1], device=device) < self.saturation * self.p, s, torch.ones_like(s)) if debug_percentile is not None: s = torch.full_like(s, torch.exp2(torch.erfinv(debug_percentile * 2 - 1) * self.saturation_std)) C = (v.ger(v) + (I_4 - v.ger(v)) * s) @ C # ------------------------------ # Execute color transformations. # ------------------------------ # Execute if the transform is not identity. if C is not I_4: images = images.reshape([batch_size, num_channels, height * width]) if num_channels == 3: images = C[:, :3, :3] @ images + C[:, :3, 3:] elif num_channels == 1: C = C[:, :3, :].mean(dim=1, keepdims=True) images = images * C[:, :, :3].sum(dim=2, keepdims=True) + C[:, :, 3:] else: raise ValueError('Image must be RGB (3 channels) or L (1 channel)') images = images.reshape([batch_size, num_channels, height, width]) # ---------------------- # Image-space filtering. # ---------------------- if self.imgfilter > 0: num_bands = self.Hz_fbank.shape[0] assert len(self.imgfilter_bands) == num_bands expected_power = misc.constant(np.array([10, 1, 1, 1]) / 13, device=device) # Expected power spectrum (1/f). # Apply amplification for each band with probability (imgfilter * strength * band_strength). g = torch.ones([batch_size, num_bands], device=device) # Global gain vector (identity). for i, band_strength in enumerate(self.imgfilter_bands): t_i = torch.exp2(torch.randn([batch_size], device=device) * self.imgfilter_std) t_i = torch.where(torch.rand([batch_size], device=device) < self.imgfilter * self.p * band_strength, t_i, torch.ones_like(t_i)) if debug_percentile is not None: t_i = torch.full_like(t_i, torch.exp2(torch.erfinv(debug_percentile * 2 - 1) * self.imgfilter_std)) if band_strength > 0 else torch.ones_like(t_i) t = torch.ones([batch_size, num_bands], device=device) # Temporary gain vector. t[:, i] = t_i # Replace i'th element. t = t / (expected_power * t.square()).sum(dim=-1, keepdims=True).sqrt() # Normalize power. g = g * t # Accumulate into global gain. # Construct combined amplification filter. Hz_prime = g @ self.Hz_fbank # [batch, tap] Hz_prime = Hz_prime.unsqueeze(1).repeat([1, num_channels, 1]) # [batch, channels, tap] Hz_prime = Hz_prime.reshape([batch_size * num_channels, 1, -1]) # [batch * channels, 1, tap] # Apply filter. p = self.Hz_fbank.shape[1] // 2 images = images.reshape([1, batch_size * num_channels, height, width]) images = torch.nn.functional.pad(input=images, pad=[p,p,p,p], mode='reflect') images = conv2d_gradfix.conv2d(input=images, weight=Hz_prime.unsqueeze(2), groups=batch_size*num_channels) images = conv2d_gradfix.conv2d(input=images, weight=Hz_prime.unsqueeze(3), groups=batch_size*num_channels) images = images.reshape([batch_size, num_channels, height, width]) # ------------------------ # Image-space corruptions. # ------------------------ # Apply additive RGB noise with probability (noise * strength). if self.noise > 0: sigma = torch.randn([batch_size, 1, 1, 1], device=device).abs() * self.noise_std sigma = torch.where(torch.rand([batch_size, 1, 1, 1], device=device) < self.noise * self.p, sigma, torch.zeros_like(sigma)) if debug_percentile is not None: sigma = torch.full_like(sigma, torch.erfinv(debug_percentile) * self.noise_std) images = images + torch.randn([batch_size, num_channels, height, width], device=device) * sigma # Apply cutout with probability (cutout * strength). if self.cutout > 0: size = torch.full([batch_size, 2, 1, 1, 1], self.cutout_size, device=device) size = torch.where(torch.rand([batch_size, 1, 1, 1, 1], device=device) < self.cutout * self.p, size, torch.zeros_like(size)) center = torch.rand([batch_size, 2, 1, 1, 1], device=device) if debug_percentile is not None: size = torch.full_like(size, self.cutout_size) center = torch.full_like(center, debug_percentile) coord_x = torch.arange(width, device=device).reshape([1, 1, 1, -1]) coord_y = torch.arange(height, device=device).reshape([1, 1, -1, 1]) mask_x = (((coord_x + 0.5) / width - center[:, 0]).abs() >= size[:, 0] / 2) mask_y = (((coord_y + 0.5) / height - center[:, 1]).abs() >= size[:, 1] / 2) mask = torch.logical_or(mask_x, mask_y).to(torch.float32) images = images * mask return images #----------------------------------------------------------------------------
26,617
59.910755
366
py
DiffProxy
DiffProxy-main/stylegan/training/dataset.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Streaming images and labels from datasets created with dataset_tool.py.""" import os import numpy as np import zipfile import PIL.Image import json import torch import dnnlib try: import pyspng except ImportError: pyspng = None #---------------------------------------------------------------------------- class Dataset(torch.utils.data.Dataset): def __init__(self, name, # Name of the dataset. raw_shape, # Shape of the raw image data (NCHW). max_size = None, # Artificially limit the size of the dataset. None = no limit. Applied before xflip. use_labels = False, # Enable conditioning labels? False = label dimension is zero. xflip = False, # Artificially double the size of the dataset via x-flips. Applied after max_size. random_seed = 0, # Random seed to use when applying max_size. ): self._name = name self._raw_shape = list(raw_shape) self._use_labels = use_labels self._raw_labels = None self._label_shape = None # Apply max_size. self._raw_idx = np.arange(self._raw_shape[0], dtype=np.int64) if (max_size is not None) and (self._raw_idx.size > max_size): np.random.RandomState(random_seed).shuffle(self._raw_idx) self._raw_idx = np.sort(self._raw_idx[:max_size]) # Apply xflip. self._xflip = np.zeros(self._raw_idx.size, dtype=np.uint8) if xflip: self._raw_idx = np.tile(self._raw_idx, 2) self._xflip = np.concatenate([self._xflip, np.ones_like(self._xflip)]) def _get_raw_labels(self): if self._raw_labels is None: self._raw_labels = self._load_raw_labels() if self._use_labels else None if self._raw_labels is None: self._raw_labels = np.zeros([self._raw_shape[0], 0], dtype=np.float32) assert isinstance(self._raw_labels, np.ndarray) assert self._raw_labels.shape[0] == self._raw_shape[0] assert self._raw_labels.dtype in [np.float32, np.int64] if self._raw_labels.dtype == np.int64: assert self._raw_labels.ndim == 1 assert np.all(self._raw_labels >= 0) return self._raw_labels def close(self): # to be overridden by subclass pass def _load_raw_image(self, raw_idx): # to be overridden by subclass raise NotImplementedError def _load_raw_labels(self): # to be overridden by subclass raise NotImplementedError def __getstate__(self): return dict(self.__dict__, _raw_labels=None) def __del__(self): try: self.close() except: pass def __len__(self): return self._raw_idx.size def __getitem__(self, idx): image = self._load_raw_image(self._raw_idx[idx]) assert isinstance(image, np.ndarray) assert list(image.shape) == self.image_shape assert image.dtype == np.uint8 if self._xflip[idx]: assert image.ndim == 3 # CHW image = image[:, :, ::-1] return image.copy(), self.get_label(idx) def get_label(self, idx): label = self._get_raw_labels()[self._raw_idx[idx]] if label.dtype == np.int64: onehot = np.zeros(self.label_shape, dtype=np.float32) onehot[label] = 1 label = onehot return label.copy() def get_details(self, idx): d = dnnlib.EasyDict() d.raw_idx = int(self._raw_idx[idx]) d.xflip = (int(self._xflip[idx]) != 0) d.raw_label = self._get_raw_labels()[d.raw_idx].copy() return d @property def name(self): return self._name @property def image_shape(self): return list(self._raw_shape[1:]) @property def num_channels(self): assert len(self.image_shape) == 3 # CHW return self.image_shape[0] @property def resolution(self): assert len(self.image_shape) == 3 # CHW assert self.image_shape[1] == self.image_shape[2] return self.image_shape[1] @property def label_shape(self): if self._label_shape is None: raw_labels = self._get_raw_labels() if raw_labels.dtype == np.int64: self._label_shape = [int(np.max(raw_labels)) + 1] else: self._label_shape = raw_labels.shape[1:] return list(self._label_shape) @property def label_dim(self): assert len(self.label_shape) == 1 return self.label_shape[0] @property def has_labels(self): return any(x != 0 for x in self.label_shape) @property def has_onehot_labels(self): return self._get_raw_labels().dtype == np.int64 #---------------------------------------------------------------------------- class ImageFolderDataset(Dataset): def __init__(self, path, # Path to directory or zip. resolution = None, # Ensure specific resolution, None = highest available. **super_kwargs, # Additional arguments for the Dataset base class. ): self._path = path self._zipfile = None if os.path.isdir(self._path): self._type = 'dir' self._all_fnames = {os.path.relpath(os.path.join(root, fname), start=self._path) for root, _dirs, files in os.walk(self._path) for fname in files} elif self._file_ext(self._path) == '.zip': self._type = 'zip' self._all_fnames = set(self._get_zipfile().namelist()) else: raise IOError('Path must point to a directory or zip') PIL.Image.init() self._image_fnames = sorted(fname for fname in self._all_fnames if self._file_ext(fname) in PIL.Image.EXTENSION) if len(self._image_fnames) == 0: raise IOError('No image files found in the specified path') name = os.path.splitext(os.path.basename(self._path))[0] raw_shape = [len(self._image_fnames)] + list(self._load_raw_image(0).shape) if resolution is not None and (raw_shape[2] != resolution or raw_shape[3] != resolution): raise IOError('Image files do not match the specified resolution') super().__init__(name=name, raw_shape=raw_shape, **super_kwargs) @staticmethod def _file_ext(fname): return os.path.splitext(fname)[1].lower() def _get_zipfile(self): assert self._type == 'zip' if self._zipfile is None: self._zipfile = zipfile.ZipFile(self._path) return self._zipfile def _open_file(self, fname): if self._type == 'dir': return open(os.path.join(self._path, fname), 'rb') if self._type == 'zip': return self._get_zipfile().open(fname, 'r') return None def close(self): try: if self._zipfile is not None: self._zipfile.close() finally: self._zipfile = None def __getstate__(self): return dict(super().__getstate__(), _zipfile=None) def _load_raw_image(self, raw_idx): fname = self._image_fnames[raw_idx] with self._open_file(fname) as f: if pyspng is not None and self._file_ext(fname) == '.png': image = pyspng.load(f.read()) else: image = np.array(PIL.Image.open(f)) if image.ndim == 2: image = image[:, :, np.newaxis] # HW => HWC image = image.transpose(2, 0, 1) # HWC => CHW return image def _load_raw_labels(self): fname = 'dataset.json' if fname not in self._all_fnames: return None with self._open_file(fname) as f: labels = json.load(f)['labels'] if labels is None: return None labels = dict(labels) labels = [labels[fname.replace('\\', '/')] for fname in self._image_fnames] labels = np.array(labels) labels = labels.astype({1: np.int64, 2: np.float32}[labels.ndim]) return labels #----------------------------------------------------------------------------
8,642
35.16318
158
py
DiffProxy
DiffProxy-main/stylegan/training/training_loop_gan.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Main training loop.""" import os import time import copy import json import pickle import psutil import PIL.Image import numpy as np import torch import dnnlib from torch_utils import misc from torch_utils import training_stats from torch_utils.ops import conv2d_gradfix from torch_utils.ops import grid_sample_gradfix import legacy from metrics import metric_main from sbs_generators import get_normalizer #---------------------------------------------------------------------------- def setup_snapshot_image_grid(training_set, random_seed=0): rnd = np.random.RandomState(random_seed) gw = np.clip(7680 // training_set.image_shape[2], 7, 32) gh = np.clip(4320 // training_set.image_shape[1], 4, 32) # No labels => show random subset of training samples. all_indices = list(range(len(training_set))) rnd.shuffle(all_indices) grid_indices = [all_indices[i % len(all_indices)] for i in range(gw * gh)] # Load data. images, labels = zip(*[training_set[i] for i in grid_indices]) return (gw, gh), np.stack(images), np.stack(labels) #---------------------------------------------------------------------------- def save_image_grid(img, fname, drange, grid_size): lo, hi = drange img = np.asarray(img, dtype=np.float32) img = (img - lo) * (255 / (hi - lo)) img = np.rint(img).clip(0, 255).astype(np.uint8) gw, gh = grid_size _N, C, H, W = img.shape img = img.reshape([gh, gw, C, H, W]) img = img.transpose(0, 3, 1, 4, 2) img = img.reshape([gh * H, gw * W, C]) assert C in [1, 3] if C == 1: PIL.Image.fromarray(img[:, :, 0], 'L').save(fname) if C == 3: PIL.Image.fromarray(img, 'RGB').save(fname) #---------------------------------------------------------------------------- def training_loop( run_dir = '.', # Output directory. training_set_kwargs = {}, # Options for training set. data_loader_kwargs = {}, # Options for torch.utils.data.DataLoader. G_kwargs = {}, # Options for generator network. D_kwargs = {}, # Options for discriminator network. G_opt_kwargs = {}, # Options for generator optimizer. D_opt_kwargs = {}, # Options for discriminator optimizer. augment_kwargs = None, # Options for augmentation pipeline. None = disable. loss_kwargs = {}, # Options for loss function. metrics = [], # Metrics to evaluate during training. random_seed = 0, # Global random seed. num_gpus = 1, # Number of GPUs participating in the training. rank = 0, # Rank of the current process in [0, num_gpus[. batch_size = 4, # Total batch size for one training iteration. Can be larger than batch_gpu * num_gpus. batch_gpu = 4, # Number of samples processed at a time by one GPU. ema_kimg = 10, # Half-life of the exponential moving average (EMA) of generator weights. ema_rampup = 0.05, # EMA ramp-up coefficient. None = no rampup. G_reg_interval = None, # How often to perform regularization for G? None = disable lazy regularization. D_reg_interval = 16, # How often to perform regularization for D? None = disable lazy regularization. augment_p = 0, # Initial value of augmentation probability. ada_target = None, # ADA target value. None = fixed p. ada_interval = 4, # How often to perform ADA adjustment? ada_kimg = 500, # ADA adjustment speed, measured in how many kimg it takes for p to increase/decrease by one unit. total_kimg = 25000, # Total length of the training, measured in thousands of real images. kimg_per_tick = 4, # Progress snapshot interval. image_snapshot_ticks = 50, # How often to save image snapshots? None = disable. network_snapshot_ticks = 50, # How often to save network snapshots? None = disable. resume_pkl = None, # Network pickle to resume training from. resume_kimg = 0, # First kimg to report when resuming training. cudnn_benchmark = True, # Enable torch.backends.cudnn.benchmark? abort_fn = None, # Callback function for determining whether to abort training. Must return consistent results across ranks. progress_fn = None, # Callback function for updating training progress. Called for all ranks. generator_name = None, # Generator name used to get parameter normalizer norm_type = 'none', # Normalization types cond_D = True, # Conditional on Discriminator ): # Initialize. start_time = time.time() device = torch.device('cuda', rank) np.random.seed(random_seed * num_gpus + rank) torch.manual_seed(random_seed * num_gpus + rank) torch.backends.cudnn.benchmark = cudnn_benchmark # Improves training speed. torch.backends.cuda.matmul.allow_tf32 = False # Improves numerical accuracy. torch.backends.cudnn.allow_tf32 = False # Improves numerical accuracy. conv2d_gradfix.enabled = True # Improves training speed. grid_sample_gradfix.enabled = True # Avoids errors with the augmentation pipe. # Load training set. if rank == 0: print('Loading training set...') training_set = dnnlib.util.construct_class_by_name(**training_set_kwargs) # subclass of training.dataset.Dataset training_set_sampler = misc.InfiniteSampler(dataset=training_set, rank=rank, num_replicas=num_gpus, seed=random_seed) training_set_iterator = iter(torch.utils.data.DataLoader(dataset=training_set, sampler=training_set_sampler, batch_size=batch_size//num_gpus, **data_loader_kwargs)) if rank == 0: print() print('Num images: ', len(training_set)) print('Image shape:', training_set.image_shape) print('Label shape:', training_set.label_shape) print('Batch Size:', batch_size) print('# of GPU: ', num_gpus) print('Batch per GPU:', batch_gpu) print() augment_pipe = None # Construct networks. if rank == 0: print('Constructing networks...') common_kwargs = dict(c_dim=training_set.label_dim, img_resolution=training_set.resolution, img_channels=training_set.num_channels) G = dnnlib.util.construct_class_by_name(**G_kwargs, **common_kwargs).train().requires_grad_(False).to(device) # subclass of torch.nn.Module d_common_kwargs = common_kwargs.copy() if not cond_D: d_common_kwargs['c_dim'] = 0 D = dnnlib.util.construct_class_by_name(**D_kwargs, **d_common_kwargs).train().requires_grad_(False).to(device) # subclass of torch.nn.Module G_ema = copy.deepcopy(G).eval() if rank == 0: print("D.c_dim = ", D.c_dim) # Resume from existing pickle. if (resume_pkl is not None) and (rank == 0): print(f'Resuming from "{resume_pkl}"') with dnnlib.util.open_url(resume_pkl) as f: resume_data = legacy.load_network_pkl(f) for name, module in [('G', G), ('D', D), ('G_ema', G_ema)]: misc.copy_params_and_buffers(resume_data[name], module, require_all=False) # Print network summary tables. if rank == 0: z = torch.empty([batch_gpu, G.z_dim], device=device) c = torch.empty([batch_gpu, G.c_dim], device=device) img = misc.print_module_summary(G, [z, c]) misc.print_module_summary(D, [img, c]) # Distribute across GPUs. if rank == 0: print(f'Distributing across {num_gpus} GPUs...') for module in [G, D, G_ema, augment_pipe]: if module is not None and num_gpus > 1: for param in misc.params_and_buffers(module): torch.distributed.broadcast(param, src=0) # Setup training phases. if rank == 0: print('Setting up training phases...') loss = dnnlib.util.construct_class_by_name(device=device, G=G, D=D, augment_pipe=augment_pipe, **loss_kwargs) # subclass of training.loss.Loss phases = [] for name, module, opt_kwargs, reg_interval in [('G', G, G_opt_kwargs, G_reg_interval), ('D', D, D_opt_kwargs, D_reg_interval)]: if reg_interval is None: opt = dnnlib.util.construct_class_by_name(params=module.parameters(), **opt_kwargs) # subclass of torch.optim.Optimizer phases += [dnnlib.EasyDict(name=name+'both', module=module, opt=opt, interval=1)] else: # Lazy regularization. mb_ratio = reg_interval / (reg_interval + 1) opt_kwargs = dnnlib.EasyDict(opt_kwargs) opt_kwargs.lr = opt_kwargs.lr * mb_ratio opt_kwargs.betas = [beta ** mb_ratio for beta in opt_kwargs.betas] opt = dnnlib.util.construct_class_by_name(module.parameters(), **opt_kwargs) # subclass of torch.optim.Optimizer phases += [dnnlib.EasyDict(name=name+'main', module=module, opt=opt, interval=1)] phases += [dnnlib.EasyDict(name=name+'reg', module=module, opt=opt, interval=reg_interval)] for phase in phases: phase.start_event = None phase.end_event = None if rank == 0: phase.start_event = torch.cuda.Event(enable_timing=True) phase.end_event = torch.cuda.Event(enable_timing=True) # initialize normalizer assert norm_type in ['none', 'norm', 'std'] normalizer = get_normalizer(generator_name, norm_type, batch_size, device) if rank == 0: print(f'Normalization type: {norm_type}') print(f'Normalizer: {normalizer}') # Export sample images. grid_size = None grid_z = None grid_c = None if rank == 0: print('Exporting sample images...') grid_size, images, labels = setup_snapshot_image_grid(training_set=training_set) save_image_grid(images, os.path.join(run_dir, 'reals.png'), drange=[0,255], grid_size=grid_size) grid_z = torch.randn([labels.shape[0], G.z_dim], device=device).split(batch_gpu) # grid_c = torch.from_numpy(labels).to(device).split(batch_gpu) grid_c = torch.from_numpy(labels).to(device) if normalizer is not None: normalizer_grid = get_normalizer(generator_name, norm_type, grid_c.shape[0], device) grid_c = normalizer_grid.normalize(grid_c) grid_c = grid_c.split(batch_gpu) images = torch.cat([G_ema(z=z, c=c, noise_mode='const').cpu() for z, c in zip(grid_z, grid_c)]).numpy() save_image_grid(images, os.path.join(run_dir, 'fakes_init.png'), drange=[-1,1], grid_size=grid_size) # Initialize logs. if rank == 0: print('Initializing logs...') stats_collector = training_stats.Collector(regex='.*') stats_metrics = dict() stats_jsonl = None stats_tfevents = None if rank == 0: stats_jsonl = open(os.path.join(run_dir, 'stats.jsonl'), 'wt') try: import torch.utils.tensorboard as tensorboard stats_tfevents = tensorboard.SummaryWriter(run_dir) except ImportError as err: print('Skipping tfevents export:', err) # Train. if rank == 0: print(f'Training for {total_kimg} kimg...') print() cur_nimg = resume_kimg * 1000 cur_tick = 0 tick_start_nimg = cur_nimg tick_start_time = time.time() maintenance_time = tick_start_time - start_time batch_idx = 0 if progress_fn is not None: progress_fn(0, total_kimg) while True: # Fetch training data. with torch.autograd.profiler.record_function('data_fetch'): phase_real_img, phase_real_c = next(training_set_iterator) phase_real_img = (phase_real_img.to(device).to(torch.float32) / 127.5 - 1).split(batch_gpu) # phase_real_c = phase_real_c.to(device).split(batch_gpu) phase_real_c = phase_real_c.to(device) if normalizer is not None: phase_real_c = normalizer.normalize(phase_real_c) phase_real_c = phase_real_c.split(batch_gpu) all_gen_z = torch.randn([len(phases) * batch_size, G.z_dim], device=device) all_gen_z = [phase_gen_z.split(batch_gpu) for phase_gen_z in all_gen_z.split(batch_size)] # Execute training phases. for phase, phase_gen_z in zip(phases, all_gen_z): # (Gmain 1); (Greg 4); (Dmain, 1); (Dreg, 16); if batch_idx % phase.interval != 0: continue if phase.start_event is not None: phase.start_event.record(torch.cuda.current_stream(device)) # Accumulate gradients. phase.opt.zero_grad(set_to_none=True) phase.module.requires_grad_(True) for real_img, real_c, gen_z in zip(phase_real_img, phase_real_c, phase_gen_z): loss.accumulate_gradients(phase=phase.name, real_img=real_img, real_c=real_c, gen_z=gen_z, gain=phase.interval, cur_nimg=cur_nimg) phase.module.requires_grad_(False) # Update weights. with torch.autograd.profiler.record_function(phase.name + '_opt'): params = [param for param in phase.module.parameters() if param.grad is not None] if len(params) > 0: flat = torch.cat([param.grad.flatten() for param in params]) if num_gpus > 1: torch.distributed.all_reduce(flat) flat /= num_gpus misc.nan_to_num(flat, nan=0, posinf=1e5, neginf=-1e5, out=flat) grads = flat.split([param.numel() for param in params]) for param, grad in zip(params, grads): param.grad = grad.reshape(param.shape) phase.opt.step() # Phase done. if phase.end_event is not None: phase.end_event.record(torch.cuda.current_stream(device)) # Update G_ema. with torch.autograd.profiler.record_function('Gema'): ema_nimg = ema_kimg * 1000 if ema_rampup is not None: ema_nimg = min(ema_nimg, cur_nimg * ema_rampup) ema_beta = 0.5 ** (batch_size / max(ema_nimg, 1e-8)) for p_ema, p in zip(G_ema.parameters(), G.parameters()): p_ema.copy_(p.lerp(p_ema, ema_beta)) for b_ema, b in zip(G_ema.buffers(), G.buffers()): b_ema.copy_(b) # Update state. cur_nimg += batch_size batch_idx += 1 # Perform maintenance tasks once per tick. done = (cur_nimg >= total_kimg * 1000) if (not done) and (cur_tick != 0) and (cur_nimg < tick_start_nimg + kimg_per_tick * 1000): continue # Print status line, accumulating the same information in training_stats. tick_end_time = time.time() fields = [] fields += [f"tick {training_stats.report0('Progress/tick', cur_tick):<5d}"] fields += [f"kimg {training_stats.report0('Progress/kimg', cur_nimg / 1e3):<8.1f}"] fields += [f"time {dnnlib.util.format_time(training_stats.report0('Timing/total_sec', tick_end_time - start_time)):<12s}"] fields += [f"sec/tick {training_stats.report0('Timing/sec_per_tick', tick_end_time - tick_start_time):<7.1f}"] fields += [f"sec/kimg {training_stats.report0('Timing/sec_per_kimg', (tick_end_time - tick_start_time) / (cur_nimg - tick_start_nimg) * 1e3):<7.2f}"] fields += [f"maintenance {training_stats.report0('Timing/maintenance_sec', maintenance_time):<6.1f}"] fields += [f"cpumem {training_stats.report0('Resources/cpu_mem_gb', psutil.Process(os.getpid()).memory_info().rss / 2**30):<6.2f}"] fields += [f"gpumem {training_stats.report0('Resources/peak_gpu_mem_gb', torch.cuda.max_memory_allocated(device) / 2**30):<6.2f}"] fields += [f"reserved {training_stats.report0('Resources/peak_gpu_mem_reserved_gb', torch.cuda.max_memory_reserved(device) / 2**30):<6.2f}"] torch.cuda.reset_peak_memory_stats() fields += [f"augment {training_stats.report0('Progress/augment', float(augment_pipe.p.cpu()) if augment_pipe is not None else 0):.3f}"] training_stats.report0('Timing/total_hours', (tick_end_time - start_time) / (60 * 60)) training_stats.report0('Timing/total_days', (tick_end_time - start_time) / (24 * 60 * 60)) if rank == 0: print(' '.join(fields)) # Check for abort. if (not done) and (abort_fn is not None) and abort_fn(): done = True if rank == 0: print() print('Aborting...') # Save image snapshot. if (rank == 0) and (image_snapshot_ticks is not None) and (done or cur_tick % image_snapshot_ticks == 0): images = torch.cat([G_ema(z=z, c=c, noise_mode='const').cpu() for z, c in zip(grid_z, grid_c)]).numpy() save_image_grid(images, os.path.join(run_dir, f'fakes{cur_nimg//1000:06d}.png'), drange=[-1,1], grid_size=grid_size) # Save network snapshot. snapshot_pkl = None snapshot_data = None if (network_snapshot_ticks is not None) and (done or cur_tick % network_snapshot_ticks == 0): snapshot_data = dict(G=G, D=D, G_ema=G_ema, augment_pipe=augment_pipe, training_set_kwargs=dict(training_set_kwargs)) for key, value in snapshot_data.items(): if isinstance(value, torch.nn.Module): value = copy.deepcopy(value).eval().requires_grad_(False) if num_gpus > 1: misc.check_ddp_consistency(value, ignore_regex=r'.*\.[^.]+_(avg|ema)') for param in misc.params_and_buffers(value): torch.distributed.broadcast(param, src=0) snapshot_data[key] = value.cpu() del value # conserve memory snapshot_pkl = os.path.join(run_dir, f'network-snapshot-{cur_nimg//1000:06d}.pkl') if rank == 0: with open(snapshot_pkl, 'wb') as f: pickle.dump(snapshot_data, f) # Evaluate metrics. if (snapshot_data is not None) and (len(metrics) > 0): if rank == 0: print('Evaluating metrics...') for metric in metrics: result_dict = metric_main.calc_metric(metric=metric, G=snapshot_data['G_ema'], dataset_kwargs=training_set_kwargs, num_gpus=num_gpus, rank=rank, device=device) if rank == 0: metric_main.report_metric(result_dict, run_dir=run_dir, snapshot_pkl=snapshot_pkl) stats_metrics.update(result_dict.results) del snapshot_data # conserve memory # Collect statistics. for phase in phases: value = [] if (phase.start_event is not None) and (phase.end_event is not None): phase.end_event.synchronize() value = phase.start_event.elapsed_time(phase.end_event) training_stats.report0('Timing/' + phase.name, value) stats_collector.update() stats_dict = stats_collector.as_dict() # Update logs. timestamp = time.time() if stats_jsonl is not None: fields = dict(stats_dict, timestamp=timestamp) stats_jsonl.write(json.dumps(fields) + '\n') stats_jsonl.flush() if stats_tfevents is not None: global_step = int(cur_nimg / 1e3) walltime = timestamp - start_time for name, value in stats_dict.items(): stats_tfevents.add_scalar(name, value.mean, global_step=global_step, walltime=walltime) for name, value in stats_metrics.items(): stats_tfevents.add_scalar(f'Metrics/{name}', value, global_step=global_step, walltime=walltime) stats_tfevents.flush() if progress_fn is not None: progress_fn(cur_nimg // 1000, total_kimg) # Update state. cur_tick += 1 tick_start_nimg = cur_nimg tick_start_time = time.time() maintenance_time = tick_start_time - tick_end_time if done: break # Done. if rank == 0: print() print('Exiting...') #----------------------------------------------------------------------------
21,150
49.239905
168
py
DiffProxy
DiffProxy-main/stylegan/training/training_loop_nogan.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Main training loop.""" import os import time import copy import json import pickle import psutil import PIL.Image import numpy as np import torch import dnnlib from torch_utils import misc from torch_utils import training_stats from torch_utils.ops import conv2d_gradfix from torch_utils.ops import grid_sample_gradfix import legacy from metrics import metric_main from sbs_generators import get_normalizer #---------------------------------------------------------------------------- def setup_snapshot_image_grid(training_set, random_seed=0): rnd = np.random.RandomState(random_seed) gw = np.clip(7680 // training_set.image_shape[2], 7, 32) gh = np.clip(4320 // training_set.image_shape[1], 4, 32) # No labels => show random subset of training samples. all_indices = list(range(len(training_set))) rnd.shuffle(all_indices) grid_indices = [all_indices[i % len(all_indices)] for i in range(gw * gh)] # Load data. images, labels = zip(*[training_set[i] for i in grid_indices]) return (gw, gh), np.stack(images), np.stack(labels) #---------------------------------------------------------------------------- def save_image_grid(img, fname, drange, grid_size): lo, hi = drange img = np.asarray(img, dtype=np.float32) img = (img - lo) * (255 / (hi - lo)) img = np.rint(img).clip(0, 255).astype(np.uint8) gw, gh = grid_size _N, C, H, W = img.shape img = img.reshape([gh, gw, C, H, W]) img = img.transpose(0, 3, 1, 4, 2) img = img.reshape([gh * H, gw * W, C]) assert C in [1, 3] if C == 1: PIL.Image.fromarray(img[:, :, 0], 'L').save(fname) if C == 3: PIL.Image.fromarray(img, 'RGB').save(fname) # ---------------------------------------------------------------------------- def training_loop( run_dir = '.', # Output directory. training_set_kwargs = {}, # Options for training set. data_loader_kwargs = {}, # Options for torch.utils.data.DataLoader. G_kwargs = {}, # Options for generator network. D_kwargs = {}, # Options for discriminator network. G_opt_kwargs = {}, # Options for generator optimizer. D_opt_kwargs = {}, # Options for discriminator optimizer. augment_kwargs = None, # Options for augmentation pipeline. None = disable. loss_kwargs = {}, # Options for loss function. metrics = [], # Metrics to evaluate during training. random_seed = 0, # Global random seed. num_gpus = 1, # Number of GPUs participating in the training. rank = 0, # Rank of the current process in [0, num_gpus[. batch_size = 4, # Total batch size for one training iteration. Can be larger than batch_gpu * num_gpus. batch_gpu = 4, # Number of samples processed at a time by one GPU. ema_kimg = 10, # Half-life of the exponential moving average (EMA) of generator weights. ema_rampup = 0.05, # EMA ramp-up coefficient. None = no rampup. G_reg_interval = None, # How often to perform regularization for G? None = disable lazy regularization. D_reg_interval = 16, # How often to perform regularization for D? None = disable lazy regularization. augment_p = 0, # Initial value of augmentation probability. ada_target = None, # ADA target value. None = fixed p. ada_interval = 4, # How often to perform ADA adjustment? ada_kimg = 500, # ADA adjustment speed, measured in how many kimg it takes for p to increase/decrease by one unit. total_kimg = 25000, # Total length of the training, measured in thousands of real images. kimg_per_tick = 4, # Progress snapshot interval. image_snapshot_ticks = 50, # How often to save image snapshots? None = disable. network_snapshot_ticks = 50, # How often to save network snapshots? None = disable. resume_pkl = None, # Network pickle to resume training from. resume_kimg = 0, # First kimg to report when resuming training. cudnn_benchmark = True, # Enable torch.backends.cudnn.benchmark? abort_fn = None, # Callback function for determining whether to abort training. Must return consistent results across ranks. progress_fn = None, # Callback function for updating training progress. Called for all ranks. generator_name = None, # Generator name used to get parameter normalizer norm_type = 'none', # Normalization types ): # Initialize. start_time = time.time() device = torch.device('cuda', rank) np.random.seed(random_seed * num_gpus + rank) torch.manual_seed(random_seed * num_gpus + rank) torch.backends.cudnn.benchmark = cudnn_benchmark # Improves training speed. torch.backends.cuda.matmul.allow_tf32 = False # Improves numerical accuracy. torch.backends.cudnn.allow_tf32 = False # Improves numerical accuracy. conv2d_gradfix.enabled = True # Improves training speed. grid_sample_gradfix.enabled = True # Avoids errors with the augmentation pipe. # Load training set. if rank == 0: print('Loading training set...') training_set = dnnlib.util.construct_class_by_name(**training_set_kwargs) # subclass of training.dataset.Dataset training_set_sampler = misc.InfiniteSampler(dataset=training_set, rank=rank, num_replicas=num_gpus, seed=random_seed) training_set_iterator = iter(torch.utils.data.DataLoader(dataset=training_set, sampler=training_set_sampler, batch_size=batch_size//num_gpus, **data_loader_kwargs)) if rank == 0: print() print('Num images: ', len(training_set)) print('Image shape:', training_set.image_shape) print('Label shape:', training_set.label_shape) print('Batch Size:', batch_size) print('# of GPU: ', num_gpus) print('Batch per GPU:', batch_gpu) print() augment_pipe= None # Construct networks. if rank == 0: print('Constructing networks...') common_kwargs = dict(c_dim=training_set.label_dim, img_resolution=training_set.resolution, img_channels=training_set.num_channels) G = dnnlib.util.construct_class_by_name(**G_kwargs, **common_kwargs).train().requires_grad_(False).to(device) # subclass of torch.nn.Module # Resume from existing pickle. if (resume_pkl is not None) and (rank == 0): print(f'Resuming from "{resume_pkl}"') with dnnlib.util.open_url(resume_pkl) as f: resume_data = pickle.load(f) for name, module in [('G', G)]: misc.copy_params_and_buffers(resume_data[name], module, require_all=False) # Print network summary tables. if rank == 0: z = torch.empty([batch_gpu, G.z_dim], device=device) c = torch.empty([batch_gpu, G.c_dim], device=device) misc.print_module_summary(G, [z, c]) # Setup training phases. if rank == 0: print('Setting up training phases...') # Distribute across GPUs. if rank == 0: print(f'Distributing across {num_gpus} GPUs...') for module in [G]: if module is not None and num_gpus > 1: for param in misc.params_and_buffers(module): torch.distributed.broadcast(param, src=0) loss = dnnlib.util.construct_class_by_name(device=device, G=G, D=None, augment_pipe=augment_pipe, **loss_kwargs) # subclass of training.loss.Loss phases = [] opt = dnnlib.util.construct_class_by_name(params=G.parameters(), **G_opt_kwargs) # subclass of torch.optim.Optimizer phases += [dnnlib.EasyDict(name='G', module=G, opt=opt)] for phase in phases: phase.start_event = None phase.end_event = None if rank == 0: phase.start_event = torch.cuda.Event(enable_timing=True) phase.end_event = torch.cuda.Event(enable_timing=True) # initialize normalizer assert norm_type in ['none', 'norm', 'std'] normalizer = get_normalizer(generator_name, norm_type, batch_size, device) if rank == 0: print(f'Normalization type: {norm_type}') print(f'Normalizer: {normalizer}') # Export sample images. grid_size = None grid_z = None grid_c = None if rank == 0: print('Exporting sample images...') grid_size, images, labels = setup_snapshot_image_grid(training_set=training_set) save_image_grid(images, os.path.join(run_dir, 'reals.png'), drange=[0,255], grid_size=grid_size) grid_z = torch.randn([labels.shape[0], G.z_dim], device=device).split(batch_gpu) # grid_c = torch.from_numpy(labels).to(device).split(batch_gpu) grid_c = torch.from_numpy(labels).to(device) if normalizer is not None: normalizer_grid = get_normalizer(generator_name, norm_type, grid_c.shape[0], device) grid_c = normalizer_grid.normalize(grid_c) grid_c = grid_c.split(batch_gpu) images = torch.cat([G(z=z, c=c, noise_mode='const').cpu() for z, c in zip(grid_z, grid_c)]).numpy() save_image_grid(images, os.path.join(run_dir, 'fakes_init.png'), drange=[-1,1], grid_size=grid_size) # Initialize logs. if rank == 0: print('Initializing logs...') stats_collector = training_stats.Collector(regex='.*') stats_metrics = dict() stats_jsonl = None stats_tfevents = None if rank == 0: stats_jsonl = open(os.path.join(run_dir, 'stats.jsonl'), 'wt') try: import torch.utils.tensorboard as tensorboard stats_tfevents = tensorboard.SummaryWriter(run_dir) except ImportError as err: print('Skipping tfevents export:', err) # Train. if rank == 0: print(f'Training for {total_kimg} kimg...') print() cur_nimg = resume_kimg * 1000 cur_tick = 0 tick_start_nimg = cur_nimg tick_start_time = time.time() maintenance_time = tick_start_time - start_time batch_idx = 0 if progress_fn is not None: progress_fn(0, total_kimg) while True: # Fetch training data. with torch.autograd.profiler.record_function('data_fetch'): phase_real_img, phase_real_c = next(training_set_iterator) phase_real_img = (phase_real_img.to(device).to(torch.float32) / 127.5 - 1).split(batch_gpu) # phase_real_c = phase_real_c.to(device).split(batch_gpu) phase_real_c = phase_real_c.to(device) if normalizer is not None: phase_real_c = normalizer.normalize(phase_real_c) phase_real_c = phase_real_c.split(batch_gpu) all_gen_z = torch.randn([len(phases) * batch_size, G.z_dim], device=device) all_gen_z = [phase_gen_z.split(batch_gpu) for phase_gen_z in all_gen_z.split(batch_size)] # Execute training phases. for phase, phase_gen_z in zip(phases, all_gen_z): if phase.start_event is not None: phase.start_event.record(torch.cuda.current_stream(device)) # Accumulate gradients. phase.opt.zero_grad(set_to_none=True) phase.module.requires_grad_(True) for real_img, real_c, gen_z in zip(phase_real_img, phase_real_c, phase_gen_z): loss.accumulate_gradients(real_img=real_img, real_c=real_c, gen_z=gen_z) phase.module.requires_grad_(False) # Update weights. with torch.autograd.profiler.record_function(phase.name + '_opt'): params = [param for param in phase.module.parameters() if param.grad is not None] if len(params) > 0: flat = torch.cat([param.grad.flatten() for param in params]) if num_gpus > 1: torch.distributed.all_reduce(flat) flat /= num_gpus misc.nan_to_num(flat, nan=0, posinf=1e5, neginf=-1e5, out=flat) grads = flat.split([param.numel() for param in params]) for param, grad in zip(params, grads): param.grad = grad.reshape(param.shape) phase.opt.step() # Phase done. if phase.end_event is not None: phase.end_event.record(torch.cuda.current_stream(device)) # Update state. cur_nimg += batch_size batch_idx += 1 # Perform maintenance tasks once per tick. done = (cur_nimg >= total_kimg * 1000) if (not done) and (cur_tick != 0) and (cur_nimg < tick_start_nimg + kimg_per_tick * 1000): continue # Print status line, accumulating the same information in training_stats. tick_end_time = time.time() fields = [] fields += [f"tick {training_stats.report0('Progress/tick', cur_tick):<5d}"] fields += [f"kimg {training_stats.report0('Progress/kimg', cur_nimg / 1e3):<8.1f}"] fields += [f"time {dnnlib.util.format_time(training_stats.report0('Timing/total_sec', tick_end_time - start_time)):<12s}"] fields += [f"sec/tick {training_stats.report0('Timing/sec_per_tick', tick_end_time - tick_start_time):<7.1f}"] fields += [f"sec/kimg {training_stats.report0('Timing/sec_per_kimg', (tick_end_time - tick_start_time) / (cur_nimg - tick_start_nimg) * 1e3):<7.2f}"] fields += [f"maintenance {training_stats.report0('Timing/maintenance_sec', maintenance_time):<6.1f}"] fields += [f"cpumem {training_stats.report0('Resources/cpu_mem_gb', psutil.Process(os.getpid()).memory_info().rss / 2**30):<6.2f}"] fields += [f"gpumem {training_stats.report0('Resources/peak_gpu_mem_gb', torch.cuda.max_memory_allocated(device) / 2**30):<6.2f}"] fields += [f"reserved {training_stats.report0('Resources/peak_gpu_mem_reserved_gb', torch.cuda.max_memory_reserved(device) / 2**30):<6.2f}"] torch.cuda.reset_peak_memory_stats() fields += [f"augment {training_stats.report0('Progress/augment', float(augment_pipe.p.cpu()) if augment_pipe is not None else 0):.3f}"] training_stats.report0('Timing/total_hours', (tick_end_time - start_time) / (60 * 60)) training_stats.report0('Timing/total_days', (tick_end_time - start_time) / (24 * 60 * 60)) if rank == 0: print(' '.join(fields)) # Check for abort. if (not done) and (abort_fn is not None) and abort_fn(): done = True if rank == 0: print() print('Aborting...') # Save image snapshot. if (rank == 0) and (image_snapshot_ticks is not None) and (done or cur_tick % image_snapshot_ticks == 0): images = torch.cat([G(z=z, c=c, noise_mode='const').cpu() for z, c in zip(grid_z, grid_c)]).numpy() save_image_grid(images, os.path.join(run_dir, f'fakes{cur_nimg//1000:06d}.png'), drange=[-1,1], grid_size=grid_size) # Save network snapshot. snapshot_pkl = None snapshot_data = None if (network_snapshot_ticks is not None) and (done or cur_tick % network_snapshot_ticks == 0): snapshot_data = dict(G=G, training_set_kwargs=dict(training_set_kwargs)) for key, value in snapshot_data.items(): if isinstance(value, torch.nn.Module): value = copy.deepcopy(value).eval().requires_grad_(False) if num_gpus > 1: misc.check_ddp_consistency(value, ignore_regex=r'.*\.[^.]+_(avg|ema)') for param in misc.params_and_buffers(value): torch.distributed.broadcast(param, src=0) snapshot_data[key] = value.cpu() del value # conserve memory snapshot_pkl = os.path.join(run_dir, f'network-snapshot-{cur_nimg//1000:06d}.pkl') if rank == 0: with open(snapshot_pkl, 'wb') as f: pickle.dump(snapshot_data, f) # Evaluate metrics. if (snapshot_data is not None) and (len(metrics) > 0): if rank == 0: print('Evaluating metrics...') for metric in metrics: result_dict = metric_main.calc_metric(metric=metric, G=snapshot_data['G_ema'], dataset_kwargs=training_set_kwargs, num_gpus=num_gpus, rank=rank, device=device) if rank == 0: metric_main.report_metric(result_dict, run_dir=run_dir, snapshot_pkl=snapshot_pkl) stats_metrics.update(result_dict.results) del snapshot_data # conserve memory # Collect statistics. for phase in phases: value = [] if (phase.start_event is not None) and (phase.end_event is not None): phase.end_event.synchronize() value = phase.start_event.elapsed_time(phase.end_event) training_stats.report0('Timing/' + phase.name, value) stats_collector.update() stats_dict = stats_collector.as_dict() # Update logs. timestamp = time.time() if stats_jsonl is not None: fields = dict(stats_dict, timestamp=timestamp) stats_jsonl.write(json.dumps(fields) + '\n') stats_jsonl.flush() if stats_tfevents is not None: global_step = int(cur_nimg / 1e3) walltime = timestamp - start_time for name, value in stats_dict.items(): stats_tfevents.add_scalar(name, value.mean, global_step=global_step, walltime=walltime) for name, value in stats_metrics.items(): stats_tfevents.add_scalar(f'Metrics/{name}', value, global_step=global_step, walltime=walltime) stats_tfevents.flush() if progress_fn is not None: progress_fn(cur_nimg // 1000, total_kimg) # Update state. cur_tick += 1 tick_start_nimg = cur_nimg tick_start_time = time.time() maintenance_time = tick_start_time - tick_end_time if done: break # Done. if rank == 0: print() print('Exiting...') #----------------------------------------------------------------------------
19,024
48.160207
168
py
DiffProxy
DiffProxy-main/stylegan/training/networks_stylegan2.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Network architectures from the paper "Analyzing and Improving the Image Quality of StyleGAN". Matches the original implementation of configs E-F by Karras et al. at https://github.com/NVlabs/stylegan2/blob/master/training/networks_stylegan2.py""" import numpy as np import torch from torch_utils import misc from torch_utils import persistence from torch_utils.ops import conv2d_resample from torch_utils.ops import upfirdn2d from torch_utils.ops import bias_act from torch_utils.ops import fma #---------------------------------------------------------------------------- @misc.profiled_function def normalize_2nd_moment(x, dim=1, eps=1e-8): return x * (x.square().mean(dim=dim, keepdim=True) + eps).rsqrt() #---------------------------------------------------------------------------- @misc.profiled_function def modulated_conv2d( x, # Input tensor of shape [batch_size, in_channels, in_height, in_width]. weight, # Weight tensor of shape [out_channels, in_channels, kernel_height, kernel_width]. styles, # Modulation coefficients of shape [batch_size, in_channels]. noise = None, # Optional noise tensor to add to the output activations. up = 1, # Integer upsampling factor. down = 1, # Integer downsampling factor. padding = 0, # Padding with respect to the upsampled image. resample_filter = None, # Low-pass filter to apply when resampling activations. Must be prepared beforehand by calling upfirdn2d.setup_filter(). demodulate = True, # Apply weight demodulation? flip_weight = True, # False = convolution, True = correlation (matches torch.nn.functional.conv2d). fused_modconv = True, # Perform modulation, convolution, and demodulation as a single fused operation? ): batch_size = x.shape[0] out_channels, in_channels, kh, kw = weight.shape misc.assert_shape(weight, [out_channels, in_channels, kh, kw]) # [OIkk] misc.assert_shape(x, [batch_size, in_channels, None, None]) # [NIHW] misc.assert_shape(styles, [batch_size, in_channels]) # [NI] # Pre-normalize inputs to avoid FP16 overflow. if x.dtype == torch.float16 and demodulate: weight = weight * (1 / np.sqrt(in_channels * kh * kw) / weight.norm(float('inf'), dim=[1,2,3], keepdim=True)) # max_Ikk styles = styles / styles.norm(float('inf'), dim=1, keepdim=True) # max_I # Calculate per-sample weights and demodulation coefficients. w = None dcoefs = None if demodulate or fused_modconv: w = weight.unsqueeze(0) # [NOIkk] w = w * styles.reshape(batch_size, 1, -1, 1, 1) # [NOIkk] if demodulate: dcoefs = (w.square().sum(dim=[2,3,4]) + 1e-8).rsqrt() # [NO] if demodulate and fused_modconv: w = w * dcoefs.reshape(batch_size, -1, 1, 1, 1) # [NOIkk] # Execute by scaling the activations before and after the convolution. if not fused_modconv: x = x * styles.to(x.dtype).reshape(batch_size, -1, 1, 1) x = conv2d_resample.conv2d_resample(x=x, w=weight.to(x.dtype), f=resample_filter, up=up, down=down, padding=padding, flip_weight=flip_weight) if demodulate and noise is not None: x = fma.fma(x, dcoefs.to(x.dtype).reshape(batch_size, -1, 1, 1), noise.to(x.dtype)) elif demodulate: x = x * dcoefs.to(x.dtype).reshape(batch_size, -1, 1, 1) elif noise is not None: x = x.add_(noise.to(x.dtype)) return x # Execute as one fused op using grouped convolution. with misc.suppress_tracer_warnings(): # this value will be treated as a constant batch_size = int(batch_size) misc.assert_shape(x, [batch_size, in_channels, None, None]) x = x.reshape(1, -1, *x.shape[2:]) w = w.reshape(-1, in_channels, kh, kw) x = conv2d_resample.conv2d_resample(x=x, w=w.to(x.dtype), f=resample_filter, up=up, down=down, padding=padding, groups=batch_size, flip_weight=flip_weight) x = x.reshape(batch_size, -1, *x.shape[2:]) if noise is not None: x = x.add_(noise) return x #---------------------------------------------------------------------------- @persistence.persistent_class class FullyConnectedLayer(torch.nn.Module): def __init__(self, in_features, # Number of input features. out_features, # Number of output features. bias = True, # Apply additive bias before the activation function? activation = 'linear', # Activation function: 'relu', 'lrelu', etc. lr_multiplier = 1, # Learning rate multiplier. bias_init = 0, # Initial value for the additive bias. ): super().__init__() self.in_features = in_features self.out_features = out_features self.activation = activation self.weight = torch.nn.Parameter(torch.randn([out_features, in_features]) / lr_multiplier) self.bias = torch.nn.Parameter(torch.full([out_features], np.float32(bias_init))) if bias else None self.weight_gain = lr_multiplier / np.sqrt(in_features) self.bias_gain = lr_multiplier def forward(self, x): w = self.weight.to(x.dtype) * self.weight_gain b = self.bias if b is not None: b = b.to(x.dtype) if self.bias_gain != 1: b = b * self.bias_gain if self.activation == 'linear' and b is not None: x = torch.addmm(b.unsqueeze(0), x, w.t()) else: x = x.matmul(w.t()) x = bias_act.bias_act(x, b, act=self.activation) return x def extra_repr(self): return f'in_features={self.in_features:d}, out_features={self.out_features:d}, activation={self.activation:s}' #---------------------------------------------------------------------------- @persistence.persistent_class class Conv2dLayer(torch.nn.Module): def __init__(self, in_channels, # Number of input channels. out_channels, # Number of output channels. kernel_size, # Width and height of the convolution kernel. bias = True, # Apply additive bias before the activation function? activation = 'linear', # Activation function: 'relu', 'lrelu', etc. up = 1, # Integer upsampling factor. down = 1, # Integer downsampling factor. resample_filter = [1,3,3,1], # Low-pass filter to apply when resampling activations. conv_clamp = None, # Clamp the output to +-X, None = disable clamping. channels_last = False, # Expect the input to have memory_format=channels_last? trainable = True, # Update the weights of this layer during training? ): super().__init__() self.in_channels = in_channels self.out_channels = out_channels self.activation = activation self.up = up self.down = down self.conv_clamp = conv_clamp self.register_buffer('resample_filter', upfirdn2d.setup_filter(resample_filter)) self.padding = kernel_size // 2 self.weight_gain = 1 / np.sqrt(in_channels * (kernel_size ** 2)) self.act_gain = bias_act.activation_funcs[activation].def_gain memory_format = torch.channels_last if channels_last else torch.contiguous_format weight = torch.randn([out_channels, in_channels, kernel_size, kernel_size]).to(memory_format=memory_format) bias = torch.zeros([out_channels]) if bias else None if trainable: self.weight = torch.nn.Parameter(weight) self.bias = torch.nn.Parameter(bias) if bias is not None else None else: self.register_buffer('weight', weight) if bias is not None: self.register_buffer('bias', bias) else: self.bias = None def forward(self, x, gain=1): w = self.weight * self.weight_gain b = self.bias.to(x.dtype) if self.bias is not None else None flip_weight = (self.up == 1) # slightly faster x = conv2d_resample.conv2d_resample(x=x, w=w.to(x.dtype), f=self.resample_filter, up=self.up, down=self.down, padding=self.padding, flip_weight=flip_weight) act_gain = self.act_gain * gain act_clamp = self.conv_clamp * gain if self.conv_clamp is not None else None x = bias_act.bias_act(x, b, act=self.activation, gain=act_gain, clamp=act_clamp) return x def extra_repr(self): return ' '.join([ f'in_channels={self.in_channels:d}, out_channels={self.out_channels:d}, activation={self.activation:s},', f'up={self.up}, down={self.down}']) #---------------------------------------------------------------------------- @persistence.persistent_class class MappingNetwork(torch.nn.Module): def __init__(self, z_dim, # Input latent (Z) dimensionality, 0 = no latent. c_dim, # Conditioning label (C) dimensionality, 0 = no label. w_dim, # Intermediate latent (W) dimensionality. num_ws, # Number of intermediate latents to output, None = do not broadcast. num_layers = 8, # Number of mapping layers. embed_features = None, # Label embedding dimensionality, None = same as w_dim. layer_features = None, # Number of intermediate features in the mapping layers, None = same as w_dim. activation = 'lrelu', # Activation function: 'relu', 'lrelu', etc. lr_multiplier = 0.01, # Learning rate multiplier for the mapping layers. w_avg_beta = 0.998, # Decay for tracking the moving average of W during training, None = do not track. ): super().__init__() self.z_dim = z_dim self.c_dim = c_dim self.w_dim = w_dim self.num_ws = num_ws self.num_layers = num_layers self.w_avg_beta = w_avg_beta if embed_features is None: embed_features = w_dim if c_dim == 0: embed_features = 0 if layer_features is None: layer_features = w_dim features_list = [z_dim + embed_features] + [layer_features] * (num_layers - 1) + [w_dim] if c_dim > 0: self.embed = FullyConnectedLayer(c_dim, embed_features) for idx in range(num_layers): in_features = features_list[idx] out_features = features_list[idx + 1] layer = FullyConnectedLayer(in_features, out_features, activation=activation, lr_multiplier=lr_multiplier) setattr(self, f'fc{idx}', layer) if num_ws is not None and w_avg_beta is not None: self.register_buffer('w_avg', torch.zeros([w_dim])) def forward(self, z, c, truncation_psi=1, truncation_cutoff=None, update_emas=False): # Embed, normalize, and concat inputs. x = None with torch.autograd.profiler.record_function('input'): if self.z_dim > 0: misc.assert_shape(z, [None, self.z_dim]) x = normalize_2nd_moment(z.to(torch.float32)) if self.c_dim > 0: misc.assert_shape(c, [None, self.c_dim]) y = normalize_2nd_moment(self.embed(c.to(torch.float32))) x = torch.cat([x, y], dim=1) if x is not None else y # Main layers. for idx in range(self.num_layers): layer = getattr(self, f'fc{idx}') x = layer(x) # Update moving average of W. if update_emas and self.w_avg_beta is not None: with torch.autograd.profiler.record_function('update_w_avg'): self.w_avg.copy_(x.detach().mean(dim=0).lerp(self.w_avg, self.w_avg_beta)) # Broadcast. if self.num_ws is not None: with torch.autograd.profiler.record_function('broadcast'): x = x.unsqueeze(1).repeat([1, self.num_ws, 1]) # Apply truncation. if truncation_psi != 1: with torch.autograd.profiler.record_function('truncate'): assert self.w_avg_beta is not None if self.num_ws is None or truncation_cutoff is None: x = self.w_avg.lerp(x, truncation_psi) else: x[:, :truncation_cutoff] = self.w_avg.lerp(x[:, :truncation_cutoff], truncation_psi) return x def extra_repr(self): return f'z_dim={self.z_dim:d}, c_dim={self.c_dim:d}, w_dim={self.w_dim:d}, num_ws={self.num_ws:d}' #---------------------------------------------------------------------------- @persistence.persistent_class class SynthesisLayer(torch.nn.Module): def __init__(self, in_channels, # Number of input channels. out_channels, # Number of output channels. w_dim, # Intermediate latent (W) dimensionality. resolution, # Resolution of this layer. kernel_size = 3, # Convolution kernel size. up = 1, # Integer upsampling factor. use_noise = False, # Enable noise input? activation = 'lrelu', # Activation function: 'relu', 'lrelu', etc. resample_filter = [1,3,3,1], # Low-pass filter to apply when resampling activations. conv_clamp = None, # Clamp the output of convolution layers to +-X, None = disable clamping. channels_last = False, # Use channels_last format for the weights? ): super().__init__() self.in_channels = in_channels self.out_channels = out_channels self.w_dim = w_dim self.resolution = resolution self.up = up self.use_noise = use_noise self.activation = activation self.conv_clamp = conv_clamp self.register_buffer('resample_filter', upfirdn2d.setup_filter(resample_filter)) self.padding = kernel_size // 2 self.act_gain = bias_act.activation_funcs[activation].def_gain self.affine = FullyConnectedLayer(w_dim, in_channels, bias_init=1) memory_format = torch.channels_last if channels_last else torch.contiguous_format self.weight = torch.nn.Parameter(torch.randn([out_channels, in_channels, kernel_size, kernel_size]).to(memory_format=memory_format)) if use_noise: self.register_buffer('noise_const', torch.randn([resolution, resolution])) self.noise_strength = torch.nn.Parameter(torch.zeros([])) self.bias = torch.nn.Parameter(torch.zeros([out_channels])) def forward(self, x, w, noise_mode='random', fused_modconv=True, gain=1): assert noise_mode in ['random', 'const', 'none'] in_resolution = self.resolution // self.up misc.assert_shape(x, [None, self.in_channels, in_resolution, in_resolution]) styles = self.affine(w) noise = None if self.use_noise and noise_mode == 'random': noise = torch.randn([x.shape[0], 1, self.resolution, self.resolution], device=x.device) * self.noise_strength if self.use_noise and noise_mode == 'const': noise = self.noise_const * self.noise_strength flip_weight = (self.up == 1) # slightly faster x = modulated_conv2d(x=x, weight=self.weight, styles=styles, noise=noise, up=self.up, padding=self.padding, resample_filter=self.resample_filter, flip_weight=flip_weight, fused_modconv=fused_modconv) act_gain = self.act_gain * gain act_clamp = self.conv_clamp * gain if self.conv_clamp is not None else None x = bias_act.bias_act(x, self.bias.to(x.dtype), act=self.activation, gain=act_gain, clamp=act_clamp) return x def extra_repr(self): return ' '.join([ f'in_channels={self.in_channels:d}, out_channels={self.out_channels:d}, w_dim={self.w_dim:d},', f'resolution={self.resolution:d}, up={self.up}, activation={self.activation:s}']) #---------------------------------------------------------------------------- @persistence.persistent_class class ToRGBLayer(torch.nn.Module): def __init__(self, in_channels, out_channels, w_dim, kernel_size=1, conv_clamp=None, channels_last=False): super().__init__() self.in_channels = in_channels self.out_channels = out_channels self.w_dim = w_dim self.conv_clamp = conv_clamp self.affine = FullyConnectedLayer(w_dim, in_channels, bias_init=1) memory_format = torch.channels_last if channels_last else torch.contiguous_format self.weight = torch.nn.Parameter(torch.randn([out_channels, in_channels, kernel_size, kernel_size]).to(memory_format=memory_format)) self.bias = torch.nn.Parameter(torch.zeros([out_channels])) self.weight_gain = 1 / np.sqrt(in_channels * (kernel_size ** 2)) def forward(self, x, w, fused_modconv=True): styles = self.affine(w) * self.weight_gain x = modulated_conv2d(x=x, weight=self.weight, styles=styles, demodulate=False, fused_modconv=fused_modconv) x = bias_act.bias_act(x, self.bias.to(x.dtype), clamp=self.conv_clamp) return x def extra_repr(self): return f'in_channels={self.in_channels:d}, out_channels={self.out_channels:d}, w_dim={self.w_dim:d}' #---------------------------------------------------------------------------- @persistence.persistent_class class SynthesisBlock(torch.nn.Module): def __init__(self, in_channels, # Number of input channels, 0 = first block. out_channels, # Number of output channels. w_dim, # Intermediate latent (W) dimensionality. resolution, # Resolution of this block. img_channels, # Number of output color channels. is_last, # Is this the last block? architecture = 'skip', # Architecture: 'orig', 'skip', 'resnet'. resample_filter = [1,3,3,1], # Low-pass filter to apply when resampling activations. conv_clamp = 256, # Clamp the output of convolution layers to +-X, None = disable clamping. use_fp16 = False, # Use FP16 for this block? fp16_channels_last = False, # Use channels-last memory format with FP16? fused_modconv_default = True, # Default value of fused_modconv. 'inference_only' = True for inference, False for training. **layer_kwargs, # Arguments for SynthesisLayer. ): assert architecture in ['orig', 'skip', 'resnet'] super().__init__() self.in_channels = in_channels self.w_dim = w_dim self.resolution = resolution self.img_channels = img_channels self.is_last = is_last self.architecture = architecture self.use_fp16 = use_fp16 self.channels_last = (use_fp16 and fp16_channels_last) self.fused_modconv_default = fused_modconv_default self.register_buffer('resample_filter', upfirdn2d.setup_filter(resample_filter)) self.num_conv = 0 self.num_torgb = 0 if in_channels == 0: self.const = torch.nn.Parameter(torch.randn([out_channels, resolution, resolution])) if in_channels != 0: self.conv0 = SynthesisLayer(in_channels, out_channels, w_dim=w_dim, resolution=resolution, up=2, resample_filter=resample_filter, conv_clamp=conv_clamp, channels_last=self.channels_last, **layer_kwargs) self.num_conv += 1 self.conv1 = SynthesisLayer(out_channels, out_channels, w_dim=w_dim, resolution=resolution, conv_clamp=conv_clamp, channels_last=self.channels_last, **layer_kwargs) self.num_conv += 1 if is_last or architecture == 'skip': self.torgb = ToRGBLayer(out_channels, img_channels, w_dim=w_dim, conv_clamp=conv_clamp, channels_last=self.channels_last) self.num_torgb += 1 if in_channels != 0 and architecture == 'resnet': self.skip = Conv2dLayer(in_channels, out_channels, kernel_size=1, bias=False, up=2, resample_filter=resample_filter, channels_last=self.channels_last) def forward(self, x, img, ws, force_fp32=False, fused_modconv=None, update_emas=False, **layer_kwargs): _ = update_emas # unused misc.assert_shape(ws, [None, self.num_conv + self.num_torgb, self.w_dim]) w_iter = iter(ws.unbind(dim=1)) if ws.device.type != 'cuda': force_fp32 = True dtype = torch.float16 if self.use_fp16 and not force_fp32 else torch.float32 memory_format = torch.channels_last if self.channels_last and not force_fp32 else torch.contiguous_format if fused_modconv is None: fused_modconv = self.fused_modconv_default if fused_modconv == 'inference_only': fused_modconv = (not self.training) # Input. if self.in_channels == 0: x = self.const.to(dtype=dtype, memory_format=memory_format) x = x.unsqueeze(0).repeat([ws.shape[0], 1, 1, 1]) else: misc.assert_shape(x, [None, self.in_channels, self.resolution // 2, self.resolution // 2]) x = x.to(dtype=dtype, memory_format=memory_format) # Main layers. if self.in_channels == 0: x = self.conv1(x, next(w_iter), fused_modconv=fused_modconv, **layer_kwargs) elif self.architecture == 'resnet': y = self.skip(x, gain=np.sqrt(0.5)) x = self.conv0(x, next(w_iter), fused_modconv=fused_modconv, **layer_kwargs) x = self.conv1(x, next(w_iter), fused_modconv=fused_modconv, gain=np.sqrt(0.5), **layer_kwargs) x = y.add_(x) else: x = self.conv0(x, next(w_iter), fused_modconv=fused_modconv, **layer_kwargs) x = self.conv1(x, next(w_iter), fused_modconv=fused_modconv, **layer_kwargs) # ToRGB. if img is not None: misc.assert_shape(img, [None, self.img_channels, self.resolution // 2, self.resolution // 2]) img = upfirdn2d.upsample2d(img, self.resample_filter) if self.is_last or self.architecture == 'skip': y = self.torgb(x, next(w_iter), fused_modconv=fused_modconv) y = y.to(dtype=torch.float32, memory_format=torch.contiguous_format) img = img.add_(y) if img is not None else y assert x.dtype == dtype assert img is None or img.dtype == torch.float32 return x, img def extra_repr(self): return f'resolution={self.resolution:d}, architecture={self.architecture:s}' #---------------------------------------------------------------------------- @persistence.persistent_class class SynthesisNetwork(torch.nn.Module): def __init__(self, w_dim, # Intermediate latent (W) dimensionality. img_resolution, # Output image resolution. img_channels, # Number of color channels. channel_base = 32768, # Overall multiplier for the number of channels. channel_max = 512, # Maximum number of channels in any layer. num_fp16_res = 4, # Use FP16 for the N highest resolutions. **block_kwargs, # Arguments for SynthesisBlock. ): assert img_resolution >= 4 and img_resolution & (img_resolution - 1) == 0 super().__init__() self.w_dim = w_dim self.img_resolution = img_resolution self.img_resolution_log2 = int(np.log2(img_resolution)) self.img_channels = img_channels self.num_fp16_res = num_fp16_res self.block_resolutions = [2 ** i for i in range(2, self.img_resolution_log2 + 1)] channels_dict = {res: min(channel_base // res, channel_max) for res in self.block_resolutions} fp16_resolution = max(2 ** (self.img_resolution_log2 + 1 - num_fp16_res), 8) self.num_ws = 0 for res in self.block_resolutions: in_channels = channels_dict[res // 2] if res > 4 else 0 out_channels = channels_dict[res] use_fp16 = (res >= fp16_resolution) is_last = (res == self.img_resolution) block = SynthesisBlock(in_channels, out_channels, w_dim=w_dim, resolution=res, img_channels=img_channels, is_last=is_last, use_fp16=use_fp16, **block_kwargs) self.num_ws += block.num_conv if is_last: self.num_ws += block.num_torgb setattr(self, f'b{res}', block) def forward(self, ws, **block_kwargs): block_ws = [] with torch.autograd.profiler.record_function('split_ws'): misc.assert_shape(ws, [None, self.num_ws, self.w_dim]) ws = ws.to(torch.float32) w_idx = 0 for res in self.block_resolutions: block = getattr(self, f'b{res}') block_ws.append(ws.narrow(1, w_idx, block.num_conv + block.num_torgb)) w_idx += block.num_conv x = img = None for res, cur_ws in zip(self.block_resolutions, block_ws): block = getattr(self, f'b{res}') x, img = block(x, img, cur_ws, **block_kwargs) return img def extra_repr(self): return ' '.join([ f'w_dim={self.w_dim:d}, num_ws={self.num_ws:d},', f'img_resolution={self.img_resolution:d}, img_channels={self.img_channels:d},', f'num_fp16_res={self.num_fp16_res:d}']) #---------------------------------------------------------------------------- @persistence.persistent_class class Generator(torch.nn.Module): def __init__(self, z_dim, # Input latent (Z) dimensionality. c_dim, # Conditioning label (C) dimensionality. w_dim, # Intermediate latent (W) dimensionality. img_resolution, # Output resolution. img_channels, # Number of output color channels. mapping_kwargs = {}, # Arguments for MappingNetwork. **synthesis_kwargs, # Arguments for SynthesisNetwork. ): super().__init__() self.z_dim = z_dim self.c_dim = c_dim self.w_dim = w_dim self.img_resolution = img_resolution self.img_channels = img_channels self.synthesis = SynthesisNetwork(w_dim=w_dim, img_resolution=img_resolution, img_channels=img_channels, **synthesis_kwargs) self.num_ws = self.synthesis.num_ws self.mapping = MappingNetwork(z_dim=z_dim, c_dim=c_dim, w_dim=w_dim, num_ws=self.num_ws, **mapping_kwargs) def forward(self, z, c, truncation_psi=1, truncation_cutoff=None, update_emas=False, **synthesis_kwargs): ws = self.mapping(z, c, truncation_psi=truncation_psi, truncation_cutoff=truncation_cutoff, update_emas=update_emas) img = self.synthesis(ws, update_emas=update_emas, **synthesis_kwargs) return img #---------------------------------------------------------------------------- @persistence.persistent_class class DiscriminatorBlock(torch.nn.Module): def __init__(self, in_channels, # Number of input channels, 0 = first block. tmp_channels, # Number of intermediate channels. out_channels, # Number of output channels. resolution, # Resolution of this block. img_channels, # Number of input color channels. first_layer_idx, # Index of the first layer. architecture = 'resnet', # Architecture: 'orig', 'skip', 'resnet'. activation = 'lrelu', # Activation function: 'relu', 'lrelu', etc. resample_filter = [1,3,3,1], # Low-pass filter to apply when resampling activations. conv_clamp = None, # Clamp the output of convolution layers to +-X, None = disable clamping. use_fp16 = False, # Use FP16 for this block? fp16_channels_last = False, # Use channels-last memory format with FP16? freeze_layers = 0, # Freeze-D: Number of layers to freeze. ): assert in_channels in [0, tmp_channels] assert architecture in ['orig', 'skip', 'resnet'] super().__init__() self.in_channels = in_channels self.resolution = resolution self.img_channels = img_channels self.first_layer_idx = first_layer_idx self.architecture = architecture self.use_fp16 = use_fp16 self.channels_last = (use_fp16 and fp16_channels_last) self.register_buffer('resample_filter', upfirdn2d.setup_filter(resample_filter)) self.num_layers = 0 def trainable_gen(): while True: layer_idx = self.first_layer_idx + self.num_layers trainable = (layer_idx >= freeze_layers) self.num_layers += 1 yield trainable trainable_iter = trainable_gen() if in_channels == 0 or architecture == 'skip': self.fromrgb = Conv2dLayer(img_channels, tmp_channels, kernel_size=1, activation=activation, trainable=next(trainable_iter), conv_clamp=conv_clamp, channels_last=self.channels_last) self.conv0 = Conv2dLayer(tmp_channels, tmp_channels, kernel_size=3, activation=activation, trainable=next(trainable_iter), conv_clamp=conv_clamp, channels_last=self.channels_last) self.conv1 = Conv2dLayer(tmp_channels, out_channels, kernel_size=3, activation=activation, down=2, trainable=next(trainable_iter), resample_filter=resample_filter, conv_clamp=conv_clamp, channels_last=self.channels_last) if architecture == 'resnet': self.skip = Conv2dLayer(tmp_channels, out_channels, kernel_size=1, bias=False, down=2, trainable=next(trainable_iter), resample_filter=resample_filter, channels_last=self.channels_last) def forward(self, x, img, force_fp32=False): if (x if x is not None else img).device.type != 'cuda': force_fp32 = True dtype = torch.float16 if self.use_fp16 and not force_fp32 else torch.float32 memory_format = torch.channels_last if self.channels_last and not force_fp32 else torch.contiguous_format # Input. if x is not None: misc.assert_shape(x, [None, self.in_channels, self.resolution, self.resolution]) x = x.to(dtype=dtype, memory_format=memory_format) # FromRGB. if self.in_channels == 0 or self.architecture == 'skip': misc.assert_shape(img, [None, self.img_channels, self.resolution, self.resolution]) img = img.to(dtype=dtype, memory_format=memory_format) y = self.fromrgb(img) x = x + y if x is not None else y img = upfirdn2d.downsample2d(img, self.resample_filter) if self.architecture == 'skip' else None # Main layers. if self.architecture == 'resnet': y = self.skip(x, gain=np.sqrt(0.5)) x = self.conv0(x) x = self.conv1(x, gain=np.sqrt(0.5)) x = y.add_(x) else: x = self.conv0(x) x = self.conv1(x) assert x.dtype == dtype return x, img def extra_repr(self): return f'resolution={self.resolution:d}, architecture={self.architecture:s}' #---------------------------------------------------------------------------- @persistence.persistent_class class MinibatchStdLayer(torch.nn.Module): def __init__(self, group_size, num_channels=1): super().__init__() self.group_size = group_size self.num_channels = num_channels def forward(self, x): N, C, H, W = x.shape with misc.suppress_tracer_warnings(): # as_tensor results are registered as constants G = torch.min(torch.as_tensor(self.group_size), torch.as_tensor(N)) if self.group_size is not None else N F = self.num_channels c = C // F y = x.reshape(G, -1, F, c, H, W) # [GnFcHW] Split minibatch N into n groups of size G, and channels C into F groups of size c. y = y - y.mean(dim=0) # [GnFcHW] Subtract mean over group. y = y.square().mean(dim=0) # [nFcHW] Calc variance over group. y = (y + 1e-8).sqrt() # [nFcHW] Calc stddev over group. y = y.mean(dim=[2,3,4]) # [nF] Take average over channels and pixels. y = y.reshape(-1, F, 1, 1) # [nF11] Add missing dimensions. y = y.repeat(G, 1, H, W) # [NFHW] Replicate over group and pixels. x = torch.cat([x, y], dim=1) # [NCHW] Append to input as new channels. return x def extra_repr(self): return f'group_size={self.group_size}, num_channels={self.num_channels:d}' #---------------------------------------------------------------------------- @persistence.persistent_class class DiscriminatorEpilogue(torch.nn.Module): def __init__(self, in_channels, # Number of input channels. cmap_dim, # Dimensionality of mapped conditioning label, 0 = no label. resolution, # Resolution of this block. img_channels, # Number of input color channels. architecture = 'resnet', # Architecture: 'orig', 'skip', 'resnet'. mbstd_group_size = 4, # Group size for the minibatch standard deviation layer, None = entire minibatch. mbstd_num_channels = 1, # Number of features for the minibatch standard deviation layer, 0 = disable. activation = 'lrelu', # Activation function: 'relu', 'lrelu', etc. conv_clamp = None, # Clamp the output of convolution layers to +-X, None = disable clamping. ): assert architecture in ['orig', 'skip', 'resnet'] super().__init__() self.in_channels = in_channels self.cmap_dim = cmap_dim self.resolution = resolution self.img_channels = img_channels self.architecture = architecture if architecture == 'skip': self.fromrgb = Conv2dLayer(img_channels, in_channels, kernel_size=1, activation=activation) self.mbstd = MinibatchStdLayer(group_size=mbstd_group_size, num_channels=mbstd_num_channels) if mbstd_num_channels > 0 else None self.conv = Conv2dLayer(in_channels + mbstd_num_channels, in_channels, kernel_size=3, activation=activation, conv_clamp=conv_clamp) self.fc = FullyConnectedLayer(in_channels * (resolution ** 2), in_channels, activation=activation) self.out = FullyConnectedLayer(in_channels, 1 if cmap_dim == 0 else cmap_dim) def forward(self, x, img, cmap, force_fp32=False): misc.assert_shape(x, [None, self.in_channels, self.resolution, self.resolution]) # [NCHW] _ = force_fp32 # unused dtype = torch.float32 memory_format = torch.contiguous_format # FromRGB. x = x.to(dtype=dtype, memory_format=memory_format) if self.architecture == 'skip': misc.assert_shape(img, [None, self.img_channels, self.resolution, self.resolution]) img = img.to(dtype=dtype, memory_format=memory_format) x = x + self.fromrgb(img) # Main layers. if self.mbstd is not None: x = self.mbstd(x) x = self.conv(x) x = self.fc(x.flatten(1)) x = self.out(x) # Conditioning. if self.cmap_dim > 0: misc.assert_shape(cmap, [None, self.cmap_dim]) x = (x * cmap).sum(dim=1, keepdim=True) * (1 / np.sqrt(self.cmap_dim)) assert x.dtype == dtype return x def extra_repr(self): return f'resolution={self.resolution:d}, architecture={self.architecture:s}' #---------------------------------------------------------------------------- @persistence.persistent_class class Discriminator(torch.nn.Module): def __init__(self, c_dim, # Conditioning label (C) dimensionality. img_resolution, # Input resolution. img_channels, # Number of input color channels. architecture = 'resnet', # Architecture: 'orig', 'skip', 'resnet'. channel_base = 32768, # Overall multiplier for the number of channels. channel_max = 512, # Maximum number of channels in any layer. num_fp16_res = 4, # Use FP16 for the N highest resolutions. conv_clamp = 256, # Clamp the output of convolution layers to +-X, None = disable clamping. cmap_dim = None, # Dimensionality of mapped conditioning label, None = default. block_kwargs = {}, # Arguments for DiscriminatorBlock. mapping_kwargs = {}, # Arguments for MappingNetwork. epilogue_kwargs = {}, # Arguments for DiscriminatorEpilogue. ): super().__init__() self.c_dim = c_dim self.img_resolution = img_resolution self.img_resolution_log2 = int(np.log2(img_resolution)) self.img_channels = img_channels self.block_resolutions = [2 ** i for i in range(self.img_resolution_log2, 2, -1)] channels_dict = {res: min(channel_base // res, channel_max) for res in self.block_resolutions + [4]} fp16_resolution = max(2 ** (self.img_resolution_log2 + 1 - num_fp16_res), 8) if cmap_dim is None: cmap_dim = channels_dict[4] if c_dim == 0: cmap_dim = 0 common_kwargs = dict(img_channels=img_channels, architecture=architecture, conv_clamp=conv_clamp) cur_layer_idx = 0 for res in self.block_resolutions: in_channels = channels_dict[res] if res < img_resolution else 0 tmp_channels = channels_dict[res] out_channels = channels_dict[res // 2] use_fp16 = (res >= fp16_resolution) block = DiscriminatorBlock(in_channels, tmp_channels, out_channels, resolution=res, first_layer_idx=cur_layer_idx, use_fp16=use_fp16, **block_kwargs, **common_kwargs) setattr(self, f'b{res}', block) cur_layer_idx += block.num_layers if c_dim > 0: self.mapping = MappingNetwork(z_dim=0, c_dim=c_dim, w_dim=cmap_dim, num_ws=None, w_avg_beta=None, **mapping_kwargs) self.b4 = DiscriminatorEpilogue(channels_dict[4], cmap_dim=cmap_dim, resolution=4, **epilogue_kwargs, **common_kwargs) def forward(self, img, c, update_emas=False, **block_kwargs): _ = update_emas # unused x = None for res in self.block_resolutions: block = getattr(self, f'b{res}') x, img = block(x, img, **block_kwargs) cmap = None if self.c_dim > 0: cmap = self.mapping(None, c) x = self.b4(x, img, cmap) return x def extra_repr(self): return f'c_dim={self.c_dim:d}, img_resolution={self.img_resolution:d}, img_channels={self.img_channels:d}' #----------------------------------------------------------------------------
40,302
49.695597
164
py
DiffProxy
DiffProxy-main/stylegan/training/networks_stylegan3.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Generator architecture from the paper "Alias-Free Generative Adversarial Networks".""" import numpy as np import scipy.signal import scipy.optimize import torch from torch_utils import misc from torch_utils import persistence from torch_utils.ops import conv2d_gradfix from torch_utils.ops import filtered_lrelu from torch_utils.ops import bias_act #---------------------------------------------------------------------------- @misc.profiled_function def modulated_conv2d( x, # Input tensor: [batch_size, in_channels, in_height, in_width] w, # Weight tensor: [out_channels, in_channels, kernel_height, kernel_width] s, # Style tensor: [batch_size, in_channels] demodulate = True, # Apply weight demodulation? padding = 0, # Padding: int or [padH, padW] input_gain = None, # Optional scale factors for the input channels: [], [in_channels], or [batch_size, in_channels] ): with misc.suppress_tracer_warnings(): # this value will be treated as a constant batch_size = int(x.shape[0]) out_channels, in_channels, kh, kw = w.shape misc.assert_shape(w, [out_channels, in_channels, kh, kw]) # [OIkk] misc.assert_shape(x, [batch_size, in_channels, None, None]) # [NIHW] misc.assert_shape(s, [batch_size, in_channels]) # [NI] # Pre-normalize inputs. if demodulate: w = w * w.square().mean([1,2,3], keepdim=True).rsqrt() s = s * s.square().mean().rsqrt() # Modulate weights. w = w.unsqueeze(0) # [NOIkk] w = w * s.unsqueeze(1).unsqueeze(3).unsqueeze(4) # [NOIkk] # Demodulate weights. if demodulate: dcoefs = (w.square().sum(dim=[2,3,4]) + 1e-8).rsqrt() # [NO] w = w * dcoefs.unsqueeze(2).unsqueeze(3).unsqueeze(4) # [NOIkk] # Apply input scaling. if input_gain is not None: input_gain = input_gain.expand(batch_size, in_channels) # [NI] w = w * input_gain.unsqueeze(1).unsqueeze(3).unsqueeze(4) # [NOIkk] # Execute as one fused op using grouped convolution. x = x.reshape(1, -1, *x.shape[2:]) w = w.reshape(-1, in_channels, kh, kw) x = conv2d_gradfix.conv2d(input=x, weight=w.to(x.dtype), padding=padding, groups=batch_size) x = x.reshape(batch_size, -1, *x.shape[2:]) return x #---------------------------------------------------------------------------- @persistence.persistent_class class FullyConnectedLayer(torch.nn.Module): def __init__(self, in_features, # Number of input features. out_features, # Number of output features. activation = 'linear', # Activation function: 'relu', 'lrelu', etc. bias = True, # Apply additive bias before the activation function? lr_multiplier = 1, # Learning rate multiplier. weight_init = 1, # Initial standard deviation of the weight tensor. bias_init = 0, # Initial value of the additive bias. ): super().__init__() self.in_features = in_features self.out_features = out_features self.activation = activation self.weight = torch.nn.Parameter(torch.randn([out_features, in_features]) * (weight_init / lr_multiplier)) bias_init = np.broadcast_to(np.asarray(bias_init, dtype=np.float32), [out_features]) self.bias = torch.nn.Parameter(torch.from_numpy(bias_init / lr_multiplier)) if bias else None self.weight_gain = lr_multiplier / np.sqrt(in_features) self.bias_gain = lr_multiplier def forward(self, x): w = self.weight.to(x.dtype) * self.weight_gain b = self.bias if b is not None: b = b.to(x.dtype) if self.bias_gain != 1: b = b * self.bias_gain if self.activation == 'linear' and b is not None: x = torch.addmm(b.unsqueeze(0), x, w.t()) else: x = x.matmul(w.t()) x = bias_act.bias_act(x, b, act=self.activation) return x def extra_repr(self): return f'in_features={self.in_features:d}, out_features={self.out_features:d}, activation={self.activation:s}' #---------------------------------------------------------------------------- @persistence.persistent_class class MappingNetwork(torch.nn.Module): def __init__(self, z_dim, # Input latent (Z) dimensionality. c_dim, # Conditioning label (C) dimensionality, 0 = no labels. w_dim, # Intermediate latent (W) dimensionality. num_ws, # Number of intermediate latents to output. num_layers = 2, # Number of mapping layers. lr_multiplier = 0.01, # Learning rate multiplier for the mapping layers. w_avg_beta = 0.998, # Decay for tracking the moving average of W during training. ): super().__init__() self.z_dim = z_dim self.c_dim = c_dim self.w_dim = w_dim self.num_ws = num_ws self.num_layers = num_layers self.w_avg_beta = w_avg_beta # Construct layers. self.embed = FullyConnectedLayer(self.c_dim, self.w_dim) if self.c_dim > 0 else None features = [self.z_dim + (self.w_dim if self.c_dim > 0 else 0)] + [self.w_dim] * self.num_layers for idx, in_features, out_features in zip(range(num_layers), features[:-1], features[1:]): layer = FullyConnectedLayer(in_features, out_features, activation='lrelu', lr_multiplier=lr_multiplier) setattr(self, f'fc{idx}', layer) self.register_buffer('w_avg', torch.zeros([w_dim])) def forward(self, z, c, truncation_psi=1, truncation_cutoff=None, update_emas=False): misc.assert_shape(z, [None, self.z_dim]) if truncation_cutoff is None: truncation_cutoff = self.num_ws # Embed, normalize, and concatenate inputs. x = z.to(torch.float32) x = x * (x.square().mean(1, keepdim=True) + 1e-8).rsqrt() if self.c_dim > 0: misc.assert_shape(c, [None, self.c_dim]) y = self.embed(c.to(torch.float32)) y = y * (y.square().mean(1, keepdim=True) + 1e-8).rsqrt() x = torch.cat([x, y], dim=1) if x is not None else y # Execute layers. for idx in range(self.num_layers): x = getattr(self, f'fc{idx}')(x) # Update moving average of W. if update_emas: self.w_avg.copy_(x.detach().mean(dim=0).lerp(self.w_avg, self.w_avg_beta)) # Broadcast and apply truncation. x = x.unsqueeze(1).repeat([1, self.num_ws, 1]) if truncation_psi != 1: x[:, :truncation_cutoff] = self.w_avg.lerp(x[:, :truncation_cutoff], truncation_psi) return x def extra_repr(self): return f'z_dim={self.z_dim:d}, c_dim={self.c_dim:d}, w_dim={self.w_dim:d}, num_ws={self.num_ws:d}' #---------------------------------------------------------------------------- @persistence.persistent_class class SynthesisInput(torch.nn.Module): def __init__(self, w_dim, # Intermediate latent (W) dimensionality. channels, # Number of output channels. size, # Output spatial size: int or [width, height]. sampling_rate, # Output sampling rate. bandwidth, # Output bandwidth. ): super().__init__() self.w_dim = w_dim self.channels = channels self.size = np.broadcast_to(np.asarray(size), [2]) self.sampling_rate = sampling_rate self.bandwidth = bandwidth # Draw random frequencies from uniform 2D disc. freqs = torch.randn([self.channels, 2]) radii = freqs.square().sum(dim=1, keepdim=True).sqrt() freqs /= radii * radii.square().exp().pow(0.25) freqs *= bandwidth phases = torch.rand([self.channels]) - 0.5 # Setup parameters and buffers. self.weight = torch.nn.Parameter(torch.randn([self.channels, self.channels])) self.affine = FullyConnectedLayer(w_dim, 4, weight_init=0, bias_init=[1,0,0,0]) self.register_buffer('transform', torch.eye(3, 3)) # User-specified inverse transform wrt. resulting image. self.register_buffer('freqs', freqs) self.register_buffer('phases', phases) def forward(self, w): # Introduce batch dimension. transforms = self.transform.unsqueeze(0) # [batch, row, col] freqs = self.freqs.unsqueeze(0) # [batch, channel, xy] phases = self.phases.unsqueeze(0) # [batch, channel] # Apply learned transformation. t = self.affine(w) # t = (r_c, r_s, t_x, t_y) t = t / t[:, :2].norm(dim=1, keepdim=True) # t' = (r'_c, r'_s, t'_x, t'_y) m_r = torch.eye(3, device=w.device).unsqueeze(0).repeat([w.shape[0], 1, 1]) # Inverse rotation wrt. resulting image. m_r[:, 0, 0] = t[:, 0] # r'_c m_r[:, 0, 1] = -t[:, 1] # r'_s m_r[:, 1, 0] = t[:, 1] # r'_s m_r[:, 1, 1] = t[:, 0] # r'_c m_t = torch.eye(3, device=w.device).unsqueeze(0).repeat([w.shape[0], 1, 1]) # Inverse translation wrt. resulting image. m_t[:, 0, 2] = -t[:, 2] # t'_x m_t[:, 1, 2] = -t[:, 3] # t'_y transforms = m_r @ m_t @ transforms # First rotate resulting image, then translate, and finally apply user-specified transform. # Transform frequencies. phases = phases + (freqs @ transforms[:, :2, 2:]).squeeze(2) freqs = freqs @ transforms[:, :2, :2] # Dampen out-of-band frequencies that may occur due to the user-specified transform. amplitudes = (1 - (freqs.norm(dim=2) - self.bandwidth) / (self.sampling_rate / 2 - self.bandwidth)).clamp(0, 1) # Construct sampling grid. theta = torch.eye(2, 3, device=w.device) theta[0, 0] = 0.5 * self.size[0] / self.sampling_rate theta[1, 1] = 0.5 * self.size[1] / self.sampling_rate grids = torch.nn.functional.affine_grid(theta.unsqueeze(0), [1, 1, self.size[1], self.size[0]], align_corners=False) # Compute Fourier features. x = (grids.unsqueeze(3) @ freqs.permute(0, 2, 1).unsqueeze(1).unsqueeze(2)).squeeze(3) # [batch, height, width, channel] x = x + phases.unsqueeze(1).unsqueeze(2) x = torch.sin(x * (np.pi * 2)) x = x * amplitudes.unsqueeze(1).unsqueeze(2) # Apply trainable mapping. weight = self.weight / np.sqrt(self.channels) x = x @ weight.t() # Ensure correct shape. x = x.permute(0, 3, 1, 2) # [batch, channel, height, width] misc.assert_shape(x, [w.shape[0], self.channels, int(self.size[1]), int(self.size[0])]) return x def extra_repr(self): return '\n'.join([ f'w_dim={self.w_dim:d}, channels={self.channels:d}, size={list(self.size)},', f'sampling_rate={self.sampling_rate:g}, bandwidth={self.bandwidth:g}']) #---------------------------------------------------------------------------- @persistence.persistent_class class SynthesisLayer(torch.nn.Module): def __init__(self, w_dim, # Intermediate latent (W) dimensionality. is_torgb, # Is this the final ToRGB layer? is_critically_sampled, # Does this layer use critical sampling? use_fp16, # Does this layer use FP16? # Input & output specifications. in_channels, # Number of input channels. out_channels, # Number of output channels. in_size, # Input spatial size: int or [width, height]. out_size, # Output spatial size: int or [width, height]. in_sampling_rate, # Input sampling rate (s). out_sampling_rate, # Output sampling rate (s). in_cutoff, # Input cutoff frequency (f_c). out_cutoff, # Output cutoff frequency (f_c). in_half_width, # Input transition band half-width (f_h). out_half_width, # Output Transition band half-width (f_h). # Hyperparameters. conv_kernel = 3, # Convolution kernel size. Ignored for final the ToRGB layer. filter_size = 6, # Low-pass filter size relative to the lower resolution when up/downsampling. lrelu_upsampling = 2, # Relative sampling rate for leaky ReLU. Ignored for final the ToRGB layer. use_radial_filters = False, # Use radially symmetric downsampling filter? Ignored for critically sampled layers. conv_clamp = 256, # Clamp the output to [-X, +X], None = disable clamping. magnitude_ema_beta = 0.999, # Decay rate for the moving average of input magnitudes. ): super().__init__() self.w_dim = w_dim self.is_torgb = is_torgb self.is_critically_sampled = is_critically_sampled self.use_fp16 = use_fp16 self.in_channels = in_channels self.out_channels = out_channels self.in_size = np.broadcast_to(np.asarray(in_size), [2]) self.out_size = np.broadcast_to(np.asarray(out_size), [2]) self.in_sampling_rate = in_sampling_rate self.out_sampling_rate = out_sampling_rate self.tmp_sampling_rate = max(in_sampling_rate, out_sampling_rate) * (1 if is_torgb else lrelu_upsampling) self.in_cutoff = in_cutoff self.out_cutoff = out_cutoff self.in_half_width = in_half_width self.out_half_width = out_half_width self.conv_kernel = 1 if is_torgb else conv_kernel self.conv_clamp = conv_clamp self.magnitude_ema_beta = magnitude_ema_beta # Setup parameters and buffers. self.affine = FullyConnectedLayer(self.w_dim, self.in_channels, bias_init=1) self.weight = torch.nn.Parameter(torch.randn([self.out_channels, self.in_channels, self.conv_kernel, self.conv_kernel])) self.bias = torch.nn.Parameter(torch.zeros([self.out_channels])) self.register_buffer('magnitude_ema', torch.ones([])) # Design upsampling filter. self.up_factor = int(np.rint(self.tmp_sampling_rate / self.in_sampling_rate)) assert self.in_sampling_rate * self.up_factor == self.tmp_sampling_rate self.up_taps = filter_size * self.up_factor if self.up_factor > 1 and not self.is_torgb else 1 self.register_buffer('up_filter', self.design_lowpass_filter( numtaps=self.up_taps, cutoff=self.in_cutoff, width=self.in_half_width*2, fs=self.tmp_sampling_rate)) # Design downsampling filter. self.down_factor = int(np.rint(self.tmp_sampling_rate / self.out_sampling_rate)) assert self.out_sampling_rate * self.down_factor == self.tmp_sampling_rate self.down_taps = filter_size * self.down_factor if self.down_factor > 1 and not self.is_torgb else 1 self.down_radial = use_radial_filters and not self.is_critically_sampled self.register_buffer('down_filter', self.design_lowpass_filter( numtaps=self.down_taps, cutoff=self.out_cutoff, width=self.out_half_width*2, fs=self.tmp_sampling_rate, radial=self.down_radial)) # Compute padding. pad_total = (self.out_size - 1) * self.down_factor + 1 # Desired output size before downsampling. pad_total -= (self.in_size + self.conv_kernel - 1) * self.up_factor # Input size after upsampling. pad_total += self.up_taps + self.down_taps - 2 # Size reduction caused by the filters. pad_lo = (pad_total + self.up_factor) // 2 # Shift sample locations according to the symmetric interpretation (Appendix C.3). pad_hi = pad_total - pad_lo self.padding = [int(pad_lo[0]), int(pad_hi[0]), int(pad_lo[1]), int(pad_hi[1])] def forward(self, x, w, noise_mode='random', force_fp32=False, update_emas=False): assert noise_mode in ['random', 'const', 'none'] # unused misc.assert_shape(x, [None, self.in_channels, int(self.in_size[1]), int(self.in_size[0])]) misc.assert_shape(w, [x.shape[0], self.w_dim]) # Track input magnitude. if update_emas: with torch.autograd.profiler.record_function('update_magnitude_ema'): magnitude_cur = x.detach().to(torch.float32).square().mean() self.magnitude_ema.copy_(magnitude_cur.lerp(self.magnitude_ema, self.magnitude_ema_beta)) input_gain = self.magnitude_ema.rsqrt() # Execute affine layer. styles = self.affine(w) if self.is_torgb: weight_gain = 1 / np.sqrt(self.in_channels * (self.conv_kernel ** 2)) styles = styles * weight_gain # Execute modulated conv2d. dtype = torch.float16 if (self.use_fp16 and not force_fp32 and x.device.type == 'cuda') else torch.float32 x = modulated_conv2d(x=x.to(dtype), w=self.weight, s=styles, padding=self.conv_kernel-1, demodulate=(not self.is_torgb), input_gain=input_gain) # Execute bias, filtered leaky ReLU, and clamping. gain = 1 if self.is_torgb else np.sqrt(2) slope = 1 if self.is_torgb else 0.2 x = filtered_lrelu.filtered_lrelu(x=x, fu=self.up_filter, fd=self.down_filter, b=self.bias.to(x.dtype), up=self.up_factor, down=self.down_factor, padding=self.padding, gain=gain, slope=slope, clamp=self.conv_clamp) # Ensure correct shape and dtype. misc.assert_shape(x, [None, self.out_channels, int(self.out_size[1]), int(self.out_size[0])]) assert x.dtype == dtype return x @staticmethod def design_lowpass_filter(numtaps, cutoff, width, fs, radial=False): assert numtaps >= 1 # Identity filter. if numtaps == 1: return None # Separable Kaiser low-pass filter. if not radial: f = scipy.signal.firwin(numtaps=numtaps, cutoff=cutoff, width=width, fs=fs) return torch.as_tensor(f, dtype=torch.float32) # Radially symmetric jinc-based filter. x = (np.arange(numtaps) - (numtaps - 1) / 2) / fs r = np.hypot(*np.meshgrid(x, x)) f = scipy.special.j1(2 * cutoff * (np.pi * r)) / (np.pi * r) beta = scipy.signal.kaiser_beta(scipy.signal.kaiser_atten(numtaps, width / (fs / 2))) w = np.kaiser(numtaps, beta) f *= np.outer(w, w) f /= np.sum(f) return torch.as_tensor(f, dtype=torch.float32) def extra_repr(self): return '\n'.join([ f'w_dim={self.w_dim:d}, is_torgb={self.is_torgb},', f'is_critically_sampled={self.is_critically_sampled}, use_fp16={self.use_fp16},', f'in_sampling_rate={self.in_sampling_rate:g}, out_sampling_rate={self.out_sampling_rate:g},', f'in_cutoff={self.in_cutoff:g}, out_cutoff={self.out_cutoff:g},', f'in_half_width={self.in_half_width:g}, out_half_width={self.out_half_width:g},', f'in_size={list(self.in_size)}, out_size={list(self.out_size)},', f'in_channels={self.in_channels:d}, out_channels={self.out_channels:d}']) #---------------------------------------------------------------------------- @persistence.persistent_class class SynthesisNetwork(torch.nn.Module): def __init__(self, w_dim, # Intermediate latent (W) dimensionality. img_resolution, # Output image resolution. img_channels, # Number of color channels. channel_base = 32768, # Overall multiplier for the number of channels. channel_max = 512, # Maximum number of channels in any layer. num_layers = 14, # Total number of layers, excluding Fourier features and ToRGB. num_critical = 2, # Number of critically sampled layers at the end. first_cutoff = 2, # Cutoff frequency of the first layer (f_{c,0}). first_stopband = 2**2.1, # Minimum stopband of the first layer (f_{t,0}). last_stopband_rel = 2**0.3, # Minimum stopband of the last layer, expressed relative to the cutoff. margin_size = 10, # Number of additional pixels outside the image. output_scale = 0.25, # Scale factor for the output image. num_fp16_res = 4, # Use FP16 for the N highest resolutions. **layer_kwargs, # Arguments for SynthesisLayer. ): super().__init__() self.w_dim = w_dim self.num_ws = num_layers + 2 self.img_resolution = img_resolution self.img_channels = img_channels self.num_layers = num_layers self.num_critical = num_critical self.margin_size = margin_size self.output_scale = output_scale self.num_fp16_res = num_fp16_res # Geometric progression of layer cutoffs and min. stopbands. last_cutoff = self.img_resolution / 2 # f_{c,N} last_stopband = last_cutoff * last_stopband_rel # f_{t,N} exponents = np.minimum(np.arange(self.num_layers + 1) / (self.num_layers - self.num_critical), 1) cutoffs = first_cutoff * (last_cutoff / first_cutoff) ** exponents # f_c[i] stopbands = first_stopband * (last_stopband / first_stopband) ** exponents # f_t[i] # Compute remaining layer parameters. sampling_rates = np.exp2(np.ceil(np.log2(np.minimum(stopbands * 2, self.img_resolution)))) # s[i] half_widths = np.maximum(stopbands, sampling_rates / 2) - cutoffs # f_h[i] sizes = sampling_rates + self.margin_size * 2 sizes[-2:] = self.img_resolution channels = np.rint(np.minimum((channel_base / 2) / cutoffs, channel_max)) channels[-1] = self.img_channels # Construct layers. self.input = SynthesisInput( w_dim=self.w_dim, channels=int(channels[0]), size=int(sizes[0]), sampling_rate=sampling_rates[0], bandwidth=cutoffs[0]) self.layer_names = [] for idx in range(self.num_layers + 1): prev = max(idx - 1, 0) is_torgb = (idx == self.num_layers) is_critically_sampled = (idx >= self.num_layers - self.num_critical) use_fp16 = (sampling_rates[idx] * (2 ** self.num_fp16_res) > self.img_resolution) layer = SynthesisLayer( w_dim=self.w_dim, is_torgb=is_torgb, is_critically_sampled=is_critically_sampled, use_fp16=use_fp16, in_channels=int(channels[prev]), out_channels= int(channels[idx]), in_size=int(sizes[prev]), out_size=int(sizes[idx]), in_sampling_rate=int(sampling_rates[prev]), out_sampling_rate=int(sampling_rates[idx]), in_cutoff=cutoffs[prev], out_cutoff=cutoffs[idx], in_half_width=half_widths[prev], out_half_width=half_widths[idx], **layer_kwargs) name = f'L{idx}_{layer.out_size[0]}_{layer.out_channels}' setattr(self, name, layer) self.layer_names.append(name) def forward(self, ws, **layer_kwargs): misc.assert_shape(ws, [None, self.num_ws, self.w_dim]) ws = ws.to(torch.float32).unbind(dim=1) # Execute layers. x = self.input(ws[0]) for name, w in zip(self.layer_names, ws[1:]): x = getattr(self, name)(x, w, **layer_kwargs) if self.output_scale != 1: x = x * self.output_scale # Ensure correct shape and dtype. misc.assert_shape(x, [None, self.img_channels, self.img_resolution, self.img_resolution]) x = x.to(torch.float32) return x def extra_repr(self): return '\n'.join([ f'w_dim={self.w_dim:d}, num_ws={self.num_ws:d},', f'img_resolution={self.img_resolution:d}, img_channels={self.img_channels:d},', f'num_layers={self.num_layers:d}, num_critical={self.num_critical:d},', f'margin_size={self.margin_size:d}, num_fp16_res={self.num_fp16_res:d}']) #---------------------------------------------------------------------------- @persistence.persistent_class class Generator(torch.nn.Module): def __init__(self, z_dim, # Input latent (Z) dimensionality. c_dim, # Conditioning label (C) dimensionality. w_dim, # Intermediate latent (W) dimensionality. img_resolution, # Output resolution. img_channels, # Number of output color channels. mapping_kwargs = {}, # Arguments for MappingNetwork. **synthesis_kwargs, # Arguments for SynthesisNetwork. ): super().__init__() self.z_dim = z_dim self.c_dim = c_dim self.w_dim = w_dim self.img_resolution = img_resolution self.img_channels = img_channels self.synthesis = SynthesisNetwork(w_dim=w_dim, img_resolution=img_resolution, img_channels=img_channels, **synthesis_kwargs) self.num_ws = self.synthesis.num_ws self.mapping = MappingNetwork(z_dim=z_dim, c_dim=c_dim, w_dim=w_dim, num_ws=self.num_ws, **mapping_kwargs) def forward(self, z, c, truncation_psi=1, truncation_cutoff=None, update_emas=False, **synthesis_kwargs): ws = self.mapping(z, c, truncation_psi=truncation_psi, truncation_cutoff=truncation_cutoff, update_emas=update_emas) img = self.synthesis(ws, update_emas=update_emas, **synthesis_kwargs) return img #----------------------------------------------------------------------------
26,208
49.792636
141
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/custom_ops.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. import glob import hashlib import importlib import os import re import shutil import uuid import torch import torch.utils.cpp_extension from torch.utils.file_baton import FileBaton #---------------------------------------------------------------------------- # Global options. verbosity = 'brief' # Verbosity level: 'none', 'brief', 'full' #---------------------------------------------------------------------------- # Internal helper funcs. def _find_compiler_bindir(): patterns = [ 'C:/Program Files (x86)/Microsoft Visual Studio/*/Professional/VC/Tools/MSVC/*/bin/Hostx64/x64', 'C:/Program Files (x86)/Microsoft Visual Studio/*/BuildTools/VC/Tools/MSVC/*/bin/Hostx64/x64', 'C:/Program Files (x86)/Microsoft Visual Studio/*/Community/VC/Tools/MSVC/*/bin/Hostx64/x64', 'C:/Program Files (x86)/Microsoft Visual Studio */vc/bin', ] for pattern in patterns: matches = sorted(glob.glob(pattern)) if len(matches): return matches[-1] return None #---------------------------------------------------------------------------- def _get_mangled_gpu_name(): name = torch.cuda.get_device_name().lower() out = [] for c in name: if re.match('[a-z0-9_-]+', c): out.append(c) else: out.append('-') return ''.join(out) #---------------------------------------------------------------------------- # Main entry point for compiling and loading C++/CUDA plugins. _cached_plugins = dict() def get_plugin(module_name, sources, headers=None, source_dir=None, **build_kwargs): assert verbosity in ['none', 'brief', 'full'] if headers is None: headers = [] if source_dir is not None: sources = [os.path.join(source_dir, fname) for fname in sources] headers = [os.path.join(source_dir, fname) for fname in headers] # Already cached? if module_name in _cached_plugins: return _cached_plugins[module_name] # Print status. if verbosity == 'full': print(f'Setting up PyTorch plugin "{module_name}"...') elif verbosity == 'brief': print(f'Setting up PyTorch plugin "{module_name}"... ', end='', flush=True) verbose_build = (verbosity == 'full') # Compile and load. try: # pylint: disable=too-many-nested-blocks # Make sure we can find the necessary compiler binaries. if os.name == 'nt' and os.system("where cl.exe >nul 2>nul") != 0: compiler_bindir = _find_compiler_bindir() if compiler_bindir is None: raise RuntimeError(f'Could not find MSVC/GCC/CLANG installation on this computer. Check _find_compiler_bindir() in "{__file__}".') os.environ['PATH'] += ';' + compiler_bindir # Some containers set TORCH_CUDA_ARCH_LIST to a list that can either # break the build or unnecessarily restrict what's available to nvcc. # Unset it to let nvcc decide based on what's available on the # machine. os.environ['TORCH_CUDA_ARCH_LIST'] = '' # Incremental build md5sum trickery. Copies all the input source files # into a cached build directory under a combined md5 digest of the input # source files. Copying is done only if the combined digest has changed. # This keeps input file timestamps and filenames the same as in previous # extension builds, allowing for fast incremental rebuilds. # # This optimization is done only in case all the source files reside in # a single directory (just for simplicity) and if the TORCH_EXTENSIONS_DIR # environment variable is set (we take this as a signal that the user # actually cares about this.) # # EDIT: We now do it regardless of TORCH_EXTENSIOS_DIR, in order to work # around the *.cu dependency bug in ninja config. # all_source_files = sorted(sources + headers) all_source_dirs = set(os.path.dirname(fname) for fname in all_source_files) if len(all_source_dirs) == 1: # and ('TORCH_EXTENSIONS_DIR' in os.environ): # Compute combined hash digest for all source files. hash_md5 = hashlib.md5() for src in all_source_files: with open(src, 'rb') as f: hash_md5.update(f.read()) # Select cached build directory name. source_digest = hash_md5.hexdigest() build_top_dir = torch.utils.cpp_extension._get_build_directory(module_name, verbose=verbose_build) # pylint: disable=protected-access cached_build_dir = os.path.join(build_top_dir, f'{source_digest}-{_get_mangled_gpu_name()}') if not os.path.isdir(cached_build_dir): tmpdir = f'{build_top_dir}/srctmp-{uuid.uuid4().hex}' os.makedirs(tmpdir) for src in all_source_files: shutil.copyfile(src, os.path.join(tmpdir, os.path.basename(src))) try: os.replace(tmpdir, cached_build_dir) # atomic except OSError: # source directory already exists, delete tmpdir and its contents. shutil.rmtree(tmpdir) if not os.path.isdir(cached_build_dir): raise # Compile. cached_sources = [os.path.join(cached_build_dir, os.path.basename(fname)) for fname in sources] torch.utils.cpp_extension.load(name=module_name, build_directory=cached_build_dir, verbose=verbose_build, sources=cached_sources, **build_kwargs) else: torch.utils.cpp_extension.load(name=module_name, verbose=verbose_build, sources=sources, **build_kwargs) # Load. module = importlib.import_module(module_name) except: if verbosity == 'brief': print('Failed!') raise # Print status and add to cache dict. if verbosity == 'full': print(f'Done setting up PyTorch plugin "{module_name}".') elif verbosity == 'brief': print('Done.') _cached_plugins[module_name] = module return module #----------------------------------------------------------------------------
6,666
41.196203
146
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/training_stats.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Facilities for reporting and collecting training statistics across multiple processes and devices. The interface is designed to minimize synchronization overhead as well as the amount of boilerplate in user code.""" import re import numpy as np import torch import dnnlib from . import misc #---------------------------------------------------------------------------- _num_moments = 3 # [num_scalars, sum_of_scalars, sum_of_squares] _reduce_dtype = torch.float32 # Data type to use for initial per-tensor reduction. _counter_dtype = torch.float64 # Data type to use for the internal counters. _rank = 0 # Rank of the current process. _sync_device = None # Device to use for multiprocess communication. None = single-process. _sync_called = False # Has _sync() been called yet? _counters = dict() # Running counters on each device, updated by report(): name => device => torch.Tensor _cumulative = dict() # Cumulative counters on the CPU, updated by _sync(): name => torch.Tensor #---------------------------------------------------------------------------- def init_multiprocessing(rank, sync_device): r"""Initializes `torch_utils.training_stats` for collecting statistics across multiple processes. This function must be called after `torch.distributed.init_process_group()` and before `Collector.update()`. The call is not necessary if multi-process collection is not needed. Args: rank: Rank of the current process. sync_device: PyTorch device to use for inter-process communication, or None to disable multi-process collection. Typically `torch.device('cuda', rank)`. """ global _rank, _sync_device assert not _sync_called _rank = rank _sync_device = sync_device #---------------------------------------------------------------------------- @misc.profiled_function def report(name, value): r"""Broadcasts the given set of scalars to all interested instances of `Collector`, across device and process boundaries. This function is expected to be extremely cheap and can be safely called from anywhere in the training loop, loss function, or inside a `torch.nn.Module`. Warning: The current implementation expects the set of unique names to be consistent across processes. Please make sure that `report()` is called at least once for each unique name by each process, and in the same order. If a given process has no scalars to broadcast, it can do `report(name, [])` (empty list). Args: name: Arbitrary string specifying the name of the statistic. Averages are accumulated separately for each unique name. value: Arbitrary set of scalars. Can be a list, tuple, NumPy array, PyTorch tensor, or Python scalar. Returns: The same `value` that was passed in. """ if name not in _counters: _counters[name] = dict() elems = torch.as_tensor(value) if elems.numel() == 0: return value elems = elems.detach().flatten().to(_reduce_dtype) moments = torch.stack([ torch.ones_like(elems).sum(), elems.sum(), elems.square().sum(), ]) assert moments.ndim == 1 and moments.shape[0] == _num_moments moments = moments.to(_counter_dtype) device = moments.device if device not in _counters[name]: _counters[name][device] = torch.zeros_like(moments) _counters[name][device].add_(moments) return value #---------------------------------------------------------------------------- def report0(name, value): r"""Broadcasts the given set of scalars by the first process (`rank = 0`), but ignores any scalars provided by the other processes. See `report()` for further details. """ report(name, value if _rank == 0 else []) return value #---------------------------------------------------------------------------- class Collector: r"""Collects the scalars broadcasted by `report()` and `report0()` and computes their long-term averages (mean and standard deviation) over user-defined periods of time. The averages are first collected into internal counters that are not directly visible to the user. They are then copied to the user-visible state as a result of calling `update()` and can then be queried using `mean()`, `std()`, `as_dict()`, etc. Calling `update()` also resets the internal counters for the next round, so that the user-visible state effectively reflects averages collected between the last two calls to `update()`. Args: regex: Regular expression defining which statistics to collect. The default is to collect everything. keep_previous: Whether to retain the previous averages if no scalars were collected on a given round (default: True). """ def __init__(self, regex='.*', keep_previous=True): self._regex = re.compile(regex) self._keep_previous = keep_previous self._cumulative = dict() self._moments = dict() self.update() self._moments.clear() def names(self): r"""Returns the names of all statistics broadcasted so far that match the regular expression specified at construction time. """ return [name for name in _counters if self._regex.fullmatch(name)] def update(self): r"""Copies current values of the internal counters to the user-visible state and resets them for the next round. If `keep_previous=True` was specified at construction time, the operation is skipped for statistics that have received no scalars since the last update, retaining their previous averages. This method performs a number of GPU-to-CPU transfers and one `torch.distributed.all_reduce()`. It is intended to be called periodically in the main training loop, typically once every N training steps. """ if not self._keep_previous: self._moments.clear() for name, cumulative in _sync(self.names()): if name not in self._cumulative: self._cumulative[name] = torch.zeros([_num_moments], dtype=_counter_dtype) delta = cumulative - self._cumulative[name] self._cumulative[name].copy_(cumulative) if float(delta[0]) != 0: self._moments[name] = delta def _get_delta(self, name): r"""Returns the raw moments that were accumulated for the given statistic between the last two calls to `update()`, or zero if no scalars were collected. """ assert self._regex.fullmatch(name) if name not in self._moments: self._moments[name] = torch.zeros([_num_moments], dtype=_counter_dtype) return self._moments[name] def num(self, name): r"""Returns the number of scalars that were accumulated for the given statistic between the last two calls to `update()`, or zero if no scalars were collected. """ delta = self._get_delta(name) return int(delta[0]) def mean(self, name): r"""Returns the mean of the scalars that were accumulated for the given statistic between the last two calls to `update()`, or NaN if no scalars were collected. """ delta = self._get_delta(name) if int(delta[0]) == 0: return float('nan') return float(delta[1] / delta[0]) def std(self, name): r"""Returns the standard deviation of the scalars that were accumulated for the given statistic between the last two calls to `update()`, or NaN if no scalars were collected. """ delta = self._get_delta(name) if int(delta[0]) == 0 or not np.isfinite(float(delta[1])): return float('nan') if int(delta[0]) == 1: return float(0) mean = float(delta[1] / delta[0]) raw_var = float(delta[2] / delta[0]) return np.sqrt(max(raw_var - np.square(mean), 0)) def as_dict(self): r"""Returns the averages accumulated between the last two calls to `update()` as an `dnnlib.EasyDict`. The contents are as follows: dnnlib.EasyDict( NAME = dnnlib.EasyDict(num=FLOAT, mean=FLOAT, std=FLOAT), ... ) """ stats = dnnlib.EasyDict() for name in self.names(): stats[name] = dnnlib.EasyDict(num=self.num(name), mean=self.mean(name), std=self.std(name)) return stats def __getitem__(self, name): r"""Convenience getter. `collector[name]` is a synonym for `collector.mean(name)`. """ return self.mean(name) #---------------------------------------------------------------------------- def _sync(names): r"""Synchronize the global cumulative counters across devices and processes. Called internally by `Collector.update()`. """ if len(names) == 0: return [] global _sync_called _sync_called = True # Collect deltas within current rank. deltas = [] device = _sync_device if _sync_device is not None else torch.device('cpu') for name in names: delta = torch.zeros([_num_moments], dtype=_counter_dtype, device=device) for counter in _counters[name].values(): delta.add_(counter.to(device)) counter.copy_(torch.zeros_like(counter)) deltas.append(delta) deltas = torch.stack(deltas) # Sum deltas across ranks. if _sync_device is not None: torch.distributed.all_reduce(deltas) # Update cumulative values. deltas = deltas.cpu() for idx, name in enumerate(names): if name not in _cumulative: _cumulative[name] = torch.zeros([_num_moments], dtype=_counter_dtype) _cumulative[name].add_(deltas[idx]) # Return name-value pairs. return [(name, _cumulative[name]) for name in names] #----------------------------------------------------------------------------
10,720
38.855019
118
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/persistence.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Facilities for pickling Python code alongside other data. The pickled code is automatically imported into a separate Python module during unpickling. This way, any previously exported pickles will remain usable even if the original code is no longer available, or if the current version of the code is not consistent with what was originally pickled.""" import sys import pickle import io import inspect import copy import uuid import types import dnnlib #---------------------------------------------------------------------------- _version = 6 # internal version number _decorators = set() # {decorator_class, ...} _import_hooks = [] # [hook_function, ...] _module_to_src_dict = dict() # {module: src, ...} _src_to_module_dict = dict() # {src: module, ...} #---------------------------------------------------------------------------- def persistent_class(orig_class): r"""Class decorator that extends a given class to save its source code when pickled. Example: from torch_utils import persistence @persistence.persistent_class class MyNetwork(torch.nn.Module): def __init__(self, num_inputs, num_outputs): super().__init__() self.fc = MyLayer(num_inputs, num_outputs) ... @persistence.persistent_class class MyLayer(torch.nn.Module): ... When pickled, any instance of `MyNetwork` and `MyLayer` will save its source code alongside other internal state (e.g., parameters, buffers, and submodules). This way, any previously exported pickle will remain usable even if the class definitions have been modified or are no longer available. The decorator saves the source code of the entire Python module containing the decorated class. It does *not* save the source code of any imported modules. Thus, the imported modules must be available during unpickling, also including `torch_utils.persistence` itself. It is ok to call functions defined in the same module from the decorated class. However, if the decorated class depends on other classes defined in the same module, they must be decorated as well. This is illustrated in the above example in the case of `MyLayer`. It is also possible to employ the decorator just-in-time before calling the constructor. For example: cls = MyLayer if want_to_make_it_persistent: cls = persistence.persistent_class(cls) layer = cls(num_inputs, num_outputs) As an additional feature, the decorator also keeps track of the arguments that were used to construct each instance of the decorated class. The arguments can be queried via `obj.init_args` and `obj.init_kwargs`, and they are automatically pickled alongside other object state. A typical use case is to first unpickle a previous instance of a persistent class, and then upgrade it to use the latest version of the source code: with open('old_pickle.pkl', 'rb') as f: old_net = pickle.load(f) new_net = MyNetwork(*old_obj.init_args, **old_obj.init_kwargs) misc.copy_params_and_buffers(old_net, new_net, require_all=True) """ assert isinstance(orig_class, type) if is_persistent(orig_class): return orig_class assert orig_class.__module__ in sys.modules orig_module = sys.modules[orig_class.__module__] orig_module_src = _module_to_src(orig_module) class Decorator(orig_class): _orig_module_src = orig_module_src _orig_class_name = orig_class.__name__ def __init__(self, *args, **kwargs): super().__init__(*args, **kwargs) self._init_args = copy.deepcopy(args) self._init_kwargs = copy.deepcopy(kwargs) assert orig_class.__name__ in orig_module.__dict__ _check_pickleable(self.__reduce__()) @property def init_args(self): return copy.deepcopy(self._init_args) @property def init_kwargs(self): return dnnlib.EasyDict(copy.deepcopy(self._init_kwargs)) def __reduce__(self): fields = list(super().__reduce__()) fields += [None] * max(3 - len(fields), 0) if fields[0] is not _reconstruct_persistent_obj: meta = dict(type='class', version=_version, module_src=self._orig_module_src, class_name=self._orig_class_name, state=fields[2]) fields[0] = _reconstruct_persistent_obj # reconstruct func fields[1] = (meta,) # reconstruct args fields[2] = None # state dict return tuple(fields) Decorator.__name__ = orig_class.__name__ _decorators.add(Decorator) return Decorator #---------------------------------------------------------------------------- def is_persistent(obj): r"""Test whether the given object or class is persistent, i.e., whether it will save its source code when pickled. """ try: if obj in _decorators: return True except TypeError: pass return type(obj) in _decorators # pylint: disable=unidiomatic-typecheck #---------------------------------------------------------------------------- def import_hook(hook): r"""Register an import hook that is called whenever a persistent object is being unpickled. A typical use case is to patch the pickled source code to avoid errors and inconsistencies when the API of some imported module has changed. The hook should have the following signature: hook(meta) -> modified meta `meta` is an instance of `dnnlib.EasyDict` with the following fields: type: Type of the persistent object, e.g. `'class'`. version: Internal version number of `torch_utils.persistence`. module_src Original source code of the Python module. class_name: Class name in the original Python module. state: Internal state of the object. Example: @persistence.import_hook def wreck_my_network(meta): if meta.class_name == 'MyNetwork': print('MyNetwork is being imported. I will wreck it!') meta.module_src = meta.module_src.replace("True", "False") return meta """ assert callable(hook) _import_hooks.append(hook) #---------------------------------------------------------------------------- def _reconstruct_persistent_obj(meta): r"""Hook that is called internally by the `pickle` module to unpickle a persistent object. """ meta = dnnlib.EasyDict(meta) meta.state = dnnlib.EasyDict(meta.state) for hook in _import_hooks: meta = hook(meta) assert meta is not None assert meta.version == _version module = _src_to_module(meta.module_src) assert meta.type == 'class' orig_class = module.__dict__[meta.class_name] decorator_class = persistent_class(orig_class) obj = decorator_class.__new__(decorator_class) setstate = getattr(obj, '__setstate__', None) if callable(setstate): setstate(meta.state) # pylint: disable=not-callable else: obj.__dict__.update(meta.state) return obj #---------------------------------------------------------------------------- def _module_to_src(module): r"""Query the source code of a given Python module. """ src = _module_to_src_dict.get(module, None) if src is None: src = inspect.getsource(module) _module_to_src_dict[module] = src _src_to_module_dict[src] = module return src def _src_to_module(src): r"""Get or create a Python module for the given source code. """ module = _src_to_module_dict.get(src, None) if module is None: module_name = "_imported_module_" + uuid.uuid4().hex module = types.ModuleType(module_name) sys.modules[module_name] = module _module_to_src_dict[module] = src _src_to_module_dict[src] = module exec(src, module.__dict__) # pylint: disable=exec-used return module #---------------------------------------------------------------------------- def _check_pickleable(obj): r"""Check that the given object is pickleable, raising an exception if it is not. This function is expected to be considerably more efficient than actually pickling the object. """ def recurse(obj): if isinstance(obj, (list, tuple, set)): return [recurse(x) for x in obj] if isinstance(obj, dict): return [[recurse(x), recurse(y)] for x, y in obj.items()] if isinstance(obj, (str, int, float, bool, bytes, bytearray)): return None # Python primitive types are pickleable. if f'{type(obj).__module__}.{type(obj).__name__}' in ['numpy.ndarray', 'torch.Tensor', 'torch.nn.parameter.Parameter']: return None # NumPy arrays and PyTorch tensors are pickleable. if is_persistent(obj): return None # Persistent objects are pickleable, by virtue of the constructor check. return obj with io.BytesIO() as f: pickle.dump(recurse(obj), f) #----------------------------------------------------------------------------
9,752
37.702381
144
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/misc.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. import re import contextlib import numpy as np import torch import warnings import dnnlib #---------------------------------------------------------------------------- # Cached construction of constant tensors. Avoids CPU=>GPU copy when the # same constant is used multiple times. _constant_cache = dict() def constant(value, shape=None, dtype=None, device=None, memory_format=None): value = np.asarray(value) if shape is not None: shape = tuple(shape) if dtype is None: dtype = torch.get_default_dtype() if device is None: device = torch.device('cpu') if memory_format is None: memory_format = torch.contiguous_format key = (value.shape, value.dtype, value.tobytes(), shape, dtype, device, memory_format) tensor = _constant_cache.get(key, None) if tensor is None: tensor = torch.as_tensor(value.copy(), dtype=dtype, device=device) if shape is not None: tensor, _ = torch.broadcast_tensors(tensor, torch.empty(shape)) tensor = tensor.contiguous(memory_format=memory_format) _constant_cache[key] = tensor return tensor #---------------------------------------------------------------------------- # Replace NaN/Inf with specified numerical values. try: nan_to_num = torch.nan_to_num # 1.8.0a0 except AttributeError: def nan_to_num(input, nan=0.0, posinf=None, neginf=None, *, out=None): # pylint: disable=redefined-builtin assert isinstance(input, torch.Tensor) if posinf is None: posinf = torch.finfo(input.dtype).max if neginf is None: neginf = torch.finfo(input.dtype).min assert nan == 0 return torch.clamp(input.unsqueeze(0).nansum(0), min=neginf, max=posinf, out=out) #---------------------------------------------------------------------------- # Symbolic assert. try: symbolic_assert = torch._assert # 1.8.0a0 # pylint: disable=protected-access except AttributeError: symbolic_assert = torch.Assert # 1.7.0 #---------------------------------------------------------------------------- # Context manager to temporarily suppress known warnings in torch.jit.trace(). # Note: Cannot use catch_warnings because of https://bugs.python.org/issue29672 @contextlib.contextmanager def suppress_tracer_warnings(): flt = ('ignore', None, torch.jit.TracerWarning, None, 0) warnings.filters.insert(0, flt) yield warnings.filters.remove(flt) #---------------------------------------------------------------------------- # Assert that the shape of a tensor matches the given list of integers. # None indicates that the size of a dimension is allowed to vary. # Performs symbolic assertion when used in torch.jit.trace(). def assert_shape(tensor, ref_shape): if tensor.ndim != len(ref_shape): raise AssertionError(f'Wrong number of dimensions: got {tensor.ndim}, expected {len(ref_shape)}') for idx, (size, ref_size) in enumerate(zip(tensor.shape, ref_shape)): if ref_size is None: pass elif isinstance(ref_size, torch.Tensor): with suppress_tracer_warnings(): # as_tensor results are registered as constants symbolic_assert(torch.equal(torch.as_tensor(size), ref_size), f'Wrong size for dimension {idx}') elif isinstance(size, torch.Tensor): with suppress_tracer_warnings(): # as_tensor results are registered as constants symbolic_assert(torch.equal(size, torch.as_tensor(ref_size)), f'Wrong size for dimension {idx}: expected {ref_size}') elif size != ref_size: raise AssertionError(f'Wrong size for dimension {idx}: got {size}, expected {ref_size}') #---------------------------------------------------------------------------- # Function decorator that calls torch.autograd.profiler.record_function(). def profiled_function(fn): def decorator(*args, **kwargs): with torch.autograd.profiler.record_function(fn.__name__): return fn(*args, **kwargs) decorator.__name__ = fn.__name__ return decorator #---------------------------------------------------------------------------- # Sampler for torch.utils.data.DataLoader that loops over the dataset # indefinitely, shuffling items as it goes. class InfiniteSampler(torch.utils.data.Sampler): def __init__(self, dataset, rank=0, num_replicas=1, shuffle=True, seed=0, window_size=0.5): assert len(dataset) > 0 assert num_replicas > 0 assert 0 <= rank < num_replicas assert 0 <= window_size <= 1 super().__init__(dataset) self.dataset = dataset self.rank = rank self.num_replicas = num_replicas self.shuffle = shuffle self.seed = seed self.window_size = window_size def __iter__(self): order = np.arange(len(self.dataset)) rnd = None window = 0 if self.shuffle: rnd = np.random.RandomState(self.seed) rnd.shuffle(order) window = int(np.rint(order.size * self.window_size)) idx = 0 while True: i = idx % order.size if idx % self.num_replicas == self.rank: yield order[i] if window >= 2: j = (i - rnd.randint(window)) % order.size order[i], order[j] = order[j], order[i] idx += 1 #---------------------------------------------------------------------------- # Utilities for operating with torch.nn.Module parameters and buffers. def params_and_buffers(module): assert isinstance(module, torch.nn.Module) return list(module.parameters()) + list(module.buffers()) def named_params_and_buffers(module): assert isinstance(module, torch.nn.Module) return list(module.named_parameters()) + list(module.named_buffers()) def copy_params_and_buffers(src_module, dst_module, require_all=False): assert isinstance(src_module, torch.nn.Module) assert isinstance(dst_module, torch.nn.Module) src_tensors = dict(named_params_and_buffers(src_module)) for name, tensor in named_params_and_buffers(dst_module): assert (name in src_tensors) or (not require_all) if name in src_tensors: tensor.copy_(src_tensors[name].detach()).requires_grad_(tensor.requires_grad) #---------------------------------------------------------------------------- # Context manager for easily enabling/disabling DistributedDataParallel # synchronization. @contextlib.contextmanager def ddp_sync(module, sync): assert isinstance(module, torch.nn.Module) if sync or not isinstance(module, torch.nn.parallel.DistributedDataParallel): yield else: with module.no_sync(): yield #---------------------------------------------------------------------------- # Check DistributedDataParallel consistency across processes. def check_ddp_consistency(module, ignore_regex=None): assert isinstance(module, torch.nn.Module) for name, tensor in named_params_and_buffers(module): fullname = type(module).__name__ + '.' + name if ignore_regex is not None and re.fullmatch(ignore_regex, fullname): continue tensor = tensor.detach() if tensor.is_floating_point(): tensor = nan_to_num(tensor) other = tensor.clone() torch.distributed.broadcast(tensor=other, src=0) assert (tensor == other).all(), fullname #---------------------------------------------------------------------------- # Print summary table of module hierarchy. def print_module_summary(module, inputs, max_nesting=3, skip_redundant=True): assert isinstance(module, torch.nn.Module) assert not isinstance(module, torch.jit.ScriptModule) assert isinstance(inputs, (tuple, list)) # Register hooks. entries = [] nesting = [0] def pre_hook(_mod, _inputs): nesting[0] += 1 def post_hook(mod, _inputs, outputs): nesting[0] -= 1 if nesting[0] <= max_nesting: outputs = list(outputs) if isinstance(outputs, (tuple, list)) else [outputs] outputs = [t for t in outputs if isinstance(t, torch.Tensor)] entries.append(dnnlib.EasyDict(mod=mod, outputs=outputs)) hooks = [mod.register_forward_pre_hook(pre_hook) for mod in module.modules()] hooks += [mod.register_forward_hook(post_hook) for mod in module.modules()] # Run module. outputs = module(*inputs) for hook in hooks: hook.remove() # Identify unique outputs, parameters, and buffers. tensors_seen = set() for e in entries: e.unique_params = [t for t in e.mod.parameters() if id(t) not in tensors_seen] e.unique_buffers = [t for t in e.mod.buffers() if id(t) not in tensors_seen] e.unique_outputs = [t for t in e.outputs if id(t) not in tensors_seen] tensors_seen |= {id(t) for t in e.unique_params + e.unique_buffers + e.unique_outputs} # Filter out redundant entries. if skip_redundant: entries = [e for e in entries if len(e.unique_params) or len(e.unique_buffers) or len(e.unique_outputs)] # Construct table. rows = [[type(module).__name__, 'Parameters', 'Buffers', 'Output shape', 'Datatype']] rows += [['---'] * len(rows[0])] param_total = 0 buffer_total = 0 submodule_names = {mod: name for name, mod in module.named_modules()} for e in entries: name = '<top-level>' if e.mod is module else submodule_names[e.mod] param_size = sum(t.numel() for t in e.unique_params) buffer_size = sum(t.numel() for t in e.unique_buffers) output_shapes = [str(list(t.shape)) for t in e.outputs] output_dtypes = [str(t.dtype).split('.')[-1] for t in e.outputs] rows += [[ name + (':0' if len(e.outputs) >= 2 else ''), str(param_size) if param_size else '-', str(buffer_size) if buffer_size else '-', (output_shapes + ['-'])[0], (output_dtypes + ['-'])[0], ]] for idx in range(1, len(e.outputs)): rows += [[name + f':{idx}', '-', '-', output_shapes[idx], output_dtypes[idx]]] param_total += param_size buffer_total += buffer_size rows += [['---'] * len(rows[0])] rows += [['Total', str(param_total), str(buffer_total), '-', '-']] # Print table. widths = [max(len(cell) for cell in column) for column in zip(*rows)] print() for row in rows: print(' '.join(cell + ' ' * (width - len(cell)) for cell, width in zip(row, widths))) print() return outputs #----------------------------------------------------------------------------
11,106
40.599251
133
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/ops/bias_act.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Custom PyTorch ops for efficient bias and activation.""" import os import numpy as np import torch import dnnlib from .. import custom_ops from .. import misc #---------------------------------------------------------------------------- activation_funcs = { 'linear': dnnlib.EasyDict(func=lambda x, **_: x, def_alpha=0, def_gain=1, cuda_idx=1, ref='', has_2nd_grad=False), 'relu': dnnlib.EasyDict(func=lambda x, **_: torch.nn.functional.relu(x), def_alpha=0, def_gain=np.sqrt(2), cuda_idx=2, ref='y', has_2nd_grad=False), 'lrelu': dnnlib.EasyDict(func=lambda x, alpha, **_: torch.nn.functional.leaky_relu(x, alpha), def_alpha=0.2, def_gain=np.sqrt(2), cuda_idx=3, ref='y', has_2nd_grad=False), 'tanh': dnnlib.EasyDict(func=lambda x, **_: torch.tanh(x), def_alpha=0, def_gain=1, cuda_idx=4, ref='y', has_2nd_grad=True), 'sigmoid': dnnlib.EasyDict(func=lambda x, **_: torch.sigmoid(x), def_alpha=0, def_gain=1, cuda_idx=5, ref='y', has_2nd_grad=True), 'elu': dnnlib.EasyDict(func=lambda x, **_: torch.nn.functional.elu(x), def_alpha=0, def_gain=1, cuda_idx=6, ref='y', has_2nd_grad=True), 'selu': dnnlib.EasyDict(func=lambda x, **_: torch.nn.functional.selu(x), def_alpha=0, def_gain=1, cuda_idx=7, ref='y', has_2nd_grad=True), 'softplus': dnnlib.EasyDict(func=lambda x, **_: torch.nn.functional.softplus(x), def_alpha=0, def_gain=1, cuda_idx=8, ref='y', has_2nd_grad=True), 'swish': dnnlib.EasyDict(func=lambda x, **_: torch.sigmoid(x) * x, def_alpha=0, def_gain=np.sqrt(2), cuda_idx=9, ref='x', has_2nd_grad=True), } #---------------------------------------------------------------------------- _plugin = None _null_tensor = torch.empty([0]) def _init(): global _plugin if _plugin is None: _plugin = custom_ops.get_plugin( module_name='bias_act_plugin', sources=['bias_act.cpp', 'bias_act.cu'], headers=['bias_act.h'], source_dir=os.path.dirname(__file__), extra_cuda_cflags=['--use_fast_math'], ) return True #---------------------------------------------------------------------------- def bias_act(x, b=None, dim=1, act='linear', alpha=None, gain=None, clamp=None, impl='cuda'): r"""Fused bias and activation function. Adds bias `b` to activation tensor `x`, evaluates activation function `act`, and scales the result by `gain`. Each of the steps is optional. In most cases, the fused op is considerably more efficient than performing the same calculation using standard PyTorch ops. It supports first and second order gradients, but not third order gradients. Args: x: Input activation tensor. Can be of any shape. b: Bias vector, or `None` to disable. Must be a 1D tensor of the same type as `x`. The shape must be known, and it must match the dimension of `x` corresponding to `dim`. dim: The dimension in `x` corresponding to the elements of `b`. The value of `dim` is ignored if `b` is not specified. act: Name of the activation function to evaluate, or `"linear"` to disable. Can be e.g. `"relu"`, `"lrelu"`, `"tanh"`, `"sigmoid"`, `"swish"`, etc. See `activation_funcs` for a full list. `None` is not allowed. alpha: Shape parameter for the activation function, or `None` to use the default. gain: Scaling factor for the output tensor, or `None` to use default. See `activation_funcs` for the default scaling of each activation function. If unsure, consider specifying 1. clamp: Clamp the output values to `[-clamp, +clamp]`, or `None` to disable the clamping (default). impl: Name of the implementation to use. Can be `"ref"` or `"cuda"` (default). Returns: Tensor of the same shape and datatype as `x`. """ assert isinstance(x, torch.Tensor) assert impl in ['ref', 'cuda'] if impl == 'cuda' and x.device.type == 'cuda' and _init(): return _bias_act_cuda(dim=dim, act=act, alpha=alpha, gain=gain, clamp=clamp).apply(x, b) return _bias_act_ref(x=x, b=b, dim=dim, act=act, alpha=alpha, gain=gain, clamp=clamp) #---------------------------------------------------------------------------- @misc.profiled_function def _bias_act_ref(x, b=None, dim=1, act='linear', alpha=None, gain=None, clamp=None): """Slow reference implementation of `bias_act()` using standard TensorFlow ops. """ assert isinstance(x, torch.Tensor) assert clamp is None or clamp >= 0 spec = activation_funcs[act] alpha = float(alpha if alpha is not None else spec.def_alpha) gain = float(gain if gain is not None else spec.def_gain) clamp = float(clamp if clamp is not None else -1) # Add bias. if b is not None: assert isinstance(b, torch.Tensor) and b.ndim == 1 assert 0 <= dim < x.ndim assert b.shape[0] == x.shape[dim] x = x + b.reshape([-1 if i == dim else 1 for i in range(x.ndim)]) # Evaluate activation function. alpha = float(alpha) x = spec.func(x, alpha=alpha) # Scale by gain. gain = float(gain) if gain != 1: x = x * gain # Clamp. if clamp >= 0: x = x.clamp(-clamp, clamp) # pylint: disable=invalid-unary-operand-type return x #---------------------------------------------------------------------------- _bias_act_cuda_cache = dict() def _bias_act_cuda(dim=1, act='linear', alpha=None, gain=None, clamp=None): """Fast CUDA implementation of `bias_act()` using custom ops. """ # Parse arguments. assert clamp is None or clamp >= 0 spec = activation_funcs[act] alpha = float(alpha if alpha is not None else spec.def_alpha) gain = float(gain if gain is not None else spec.def_gain) clamp = float(clamp if clamp is not None else -1) # Lookup from cache. key = (dim, act, alpha, gain, clamp) if key in _bias_act_cuda_cache: return _bias_act_cuda_cache[key] # Forward op. class BiasActCuda(torch.autograd.Function): @staticmethod def forward(ctx, x, b): # pylint: disable=arguments-differ ctx.memory_format = torch.channels_last if x.ndim > 2 and x.stride(1) == 1 else torch.contiguous_format x = x.contiguous(memory_format=ctx.memory_format) b = b.contiguous() if b is not None else _null_tensor y = x if act != 'linear' or gain != 1 or clamp >= 0 or b is not _null_tensor: y = _plugin.bias_act(x, b, _null_tensor, _null_tensor, _null_tensor, 0, dim, spec.cuda_idx, alpha, gain, clamp) ctx.save_for_backward( x if 'x' in spec.ref or spec.has_2nd_grad else _null_tensor, b if 'x' in spec.ref or spec.has_2nd_grad else _null_tensor, y if 'y' in spec.ref else _null_tensor) return y @staticmethod def backward(ctx, dy): # pylint: disable=arguments-differ dy = dy.contiguous(memory_format=ctx.memory_format) x, b, y = ctx.saved_tensors dx = None db = None if ctx.needs_input_grad[0] or ctx.needs_input_grad[1]: dx = dy if act != 'linear' or gain != 1 or clamp >= 0: dx = BiasActCudaGrad.apply(dy, x, b, y) if ctx.needs_input_grad[1]: db = dx.sum([i for i in range(dx.ndim) if i != dim]) return dx, db # Backward op. class BiasActCudaGrad(torch.autograd.Function): @staticmethod def forward(ctx, dy, x, b, y): # pylint: disable=arguments-differ ctx.memory_format = torch.channels_last if dy.ndim > 2 and dy.stride(1) == 1 else torch.contiguous_format dx = _plugin.bias_act(dy, b, x, y, _null_tensor, 1, dim, spec.cuda_idx, alpha, gain, clamp) ctx.save_for_backward( dy if spec.has_2nd_grad else _null_tensor, x, b, y) return dx @staticmethod def backward(ctx, d_dx): # pylint: disable=arguments-differ d_dx = d_dx.contiguous(memory_format=ctx.memory_format) dy, x, b, y = ctx.saved_tensors d_dy = None d_x = None d_b = None d_y = None if ctx.needs_input_grad[0]: d_dy = BiasActCudaGrad.apply(d_dx, x, b, y) if spec.has_2nd_grad and (ctx.needs_input_grad[1] or ctx.needs_input_grad[2]): d_x = _plugin.bias_act(d_dx, b, x, y, dy, 2, dim, spec.cuda_idx, alpha, gain, clamp) if spec.has_2nd_grad and ctx.needs_input_grad[2]: d_b = d_x.sum([i for i in range(d_x.ndim) if i != dim]) return d_dy, d_x, d_b, d_y # Add to cache. _bias_act_cuda_cache[key] = BiasActCuda return BiasActCuda #----------------------------------------------------------------------------
9,813
45.733333
185
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/ops/grid_sample_gradfix.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Custom replacement for `torch.nn.functional.grid_sample` that supports arbitrarily high order gradients between the input and output. Only works on 2D images and assumes `mode='bilinear'`, `padding_mode='zeros'`, `align_corners=False`.""" import torch # pylint: disable=redefined-builtin # pylint: disable=arguments-differ # pylint: disable=protected-access #---------------------------------------------------------------------------- enabled = False # Enable the custom op by setting this to true. #---------------------------------------------------------------------------- def grid_sample(input, grid): if _should_use_custom_op(): return _GridSample2dForward.apply(input, grid) return torch.nn.functional.grid_sample(input=input, grid=grid, mode='bilinear', padding_mode='zeros', align_corners=False) #---------------------------------------------------------------------------- def _should_use_custom_op(): return enabled #---------------------------------------------------------------------------- class _GridSample2dForward(torch.autograd.Function): @staticmethod def forward(ctx, input, grid): assert input.ndim == 4 assert grid.ndim == 4 output = torch.nn.functional.grid_sample(input=input, grid=grid, mode='bilinear', padding_mode='zeros', align_corners=False) ctx.save_for_backward(input, grid) return output @staticmethod def backward(ctx, grad_output): input, grid = ctx.saved_tensors grad_input, grad_grid = _GridSample2dBackward.apply(grad_output, input, grid) return grad_input, grad_grid #---------------------------------------------------------------------------- class _GridSample2dBackward(torch.autograd.Function): @staticmethod def forward(ctx, grad_output, input, grid): op = torch._C._jit_get_operation('aten::grid_sampler_2d_backward') grad_input, grad_grid = op(grad_output, input, grid, 0, 0, False) ctx.save_for_backward(grid) return grad_input, grad_grid @staticmethod def backward(ctx, grad2_grad_input, grad2_grad_grid): _ = grad2_grad_grid # unused grid, = ctx.saved_tensors grad2_grad_output = None grad2_input = None grad2_grid = None if ctx.needs_input_grad[0]: grad2_grad_output = _GridSample2dForward.apply(grad2_grad_input, grid) assert not ctx.needs_input_grad[2] return grad2_grad_output, grad2_input, grad2_grid #----------------------------------------------------------------------------
3,020
37.730769
132
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/ops/conv2d_gradfix.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Custom replacement for `torch.nn.functional.conv2d` that supports arbitrarily high order gradients with zero performance penalty.""" import contextlib import torch # pylint: disable=redefined-builtin # pylint: disable=arguments-differ # pylint: disable=protected-access #---------------------------------------------------------------------------- enabled = False # Enable the custom op by setting this to true. weight_gradients_disabled = False # Forcefully disable computation of gradients with respect to the weights. @contextlib.contextmanager def no_weight_gradients(disable=True): global weight_gradients_disabled old = weight_gradients_disabled if disable: weight_gradients_disabled = True yield weight_gradients_disabled = old #---------------------------------------------------------------------------- def conv2d(input, weight, bias=None, stride=1, padding=0, dilation=1, groups=1): if _should_use_custom_op(input): return _conv2d_gradfix(transpose=False, weight_shape=weight.shape, stride=stride, padding=padding, output_padding=0, dilation=dilation, groups=groups).apply(input, weight, bias) return torch.nn.functional.conv2d(input=input, weight=weight, bias=bias, stride=stride, padding=padding, dilation=dilation, groups=groups) def conv_transpose2d(input, weight, bias=None, stride=1, padding=0, output_padding=0, groups=1, dilation=1): if _should_use_custom_op(input): return _conv2d_gradfix(transpose=True, weight_shape=weight.shape, stride=stride, padding=padding, output_padding=output_padding, groups=groups, dilation=dilation).apply(input, weight, bias) return torch.nn.functional.conv_transpose2d(input=input, weight=weight, bias=bias, stride=stride, padding=padding, output_padding=output_padding, groups=groups, dilation=dilation) #---------------------------------------------------------------------------- def _should_use_custom_op(input): assert isinstance(input, torch.Tensor) if (not enabled) or (not torch.backends.cudnn.enabled): return False if input.device.type != 'cuda': return False return True def _tuple_of_ints(xs, ndim): xs = tuple(xs) if isinstance(xs, (tuple, list)) else (xs,) * ndim assert len(xs) == ndim assert all(isinstance(x, int) for x in xs) return xs #---------------------------------------------------------------------------- _conv2d_gradfix_cache = dict() _null_tensor = torch.empty([0]) def _conv2d_gradfix(transpose, weight_shape, stride, padding, output_padding, dilation, groups): # Parse arguments. ndim = 2 weight_shape = tuple(weight_shape) stride = _tuple_of_ints(stride, ndim) padding = _tuple_of_ints(padding, ndim) output_padding = _tuple_of_ints(output_padding, ndim) dilation = _tuple_of_ints(dilation, ndim) # Lookup from cache. key = (transpose, weight_shape, stride, padding, output_padding, dilation, groups) if key in _conv2d_gradfix_cache: return _conv2d_gradfix_cache[key] # Validate arguments. assert groups >= 1 assert len(weight_shape) == ndim + 2 assert all(stride[i] >= 1 for i in range(ndim)) assert all(padding[i] >= 0 for i in range(ndim)) assert all(dilation[i] >= 0 for i in range(ndim)) if not transpose: assert all(output_padding[i] == 0 for i in range(ndim)) else: # transpose assert all(0 <= output_padding[i] < max(stride[i], dilation[i]) for i in range(ndim)) # Helpers. common_kwargs = dict(stride=stride, padding=padding, dilation=dilation, groups=groups) def calc_output_padding(input_shape, output_shape): if transpose: return [0, 0] return [ input_shape[i + 2] - (output_shape[i + 2] - 1) * stride[i] - (1 - 2 * padding[i]) - dilation[i] * (weight_shape[i + 2] - 1) for i in range(ndim) ] # Forward & backward. class Conv2d(torch.autograd.Function): @staticmethod def forward(ctx, input, weight, bias): assert weight.shape == weight_shape ctx.save_for_backward( input if weight.requires_grad else _null_tensor, weight if input.requires_grad else _null_tensor, ) ctx.input_shape = input.shape # Simple 1x1 convolution => cuBLAS (only on Volta, not on Ampere). if weight_shape[2:] == stride == dilation == (1, 1) and padding == (0, 0) and torch.cuda.get_device_capability(input.device) < (8, 0): a = weight.reshape(groups, weight_shape[0] // groups, weight_shape[1]) b = input.reshape(input.shape[0], groups, input.shape[1] // groups, -1) c = (a.transpose(1, 2) if transpose else a) @ b.permute(1, 2, 0, 3).flatten(2) c = c.reshape(-1, input.shape[0], *input.shape[2:]).transpose(0, 1) c = c if bias is None else c + bias.unsqueeze(0).unsqueeze(2).unsqueeze(3) return c.contiguous(memory_format=(torch.channels_last if input.stride(1) == 1 else torch.contiguous_format)) # General case => cuDNN. if transpose: return torch.nn.functional.conv_transpose2d(input=input, weight=weight, bias=bias, output_padding=output_padding, **common_kwargs) return torch.nn.functional.conv2d(input=input, weight=weight, bias=bias, **common_kwargs) @staticmethod def backward(ctx, grad_output): input, weight = ctx.saved_tensors input_shape = ctx.input_shape grad_input = None grad_weight = None grad_bias = None if ctx.needs_input_grad[0]: p = calc_output_padding(input_shape=input_shape, output_shape=grad_output.shape) op = _conv2d_gradfix(transpose=(not transpose), weight_shape=weight_shape, output_padding=p, **common_kwargs) grad_input = op.apply(grad_output, weight, None) assert grad_input.shape == input_shape if ctx.needs_input_grad[1] and not weight_gradients_disabled: grad_weight = Conv2dGradWeight.apply(grad_output, input) assert grad_weight.shape == weight_shape if ctx.needs_input_grad[2]: grad_bias = grad_output.sum([0, 2, 3]) return grad_input, grad_weight, grad_bias # Gradient with respect to the weights. class Conv2dGradWeight(torch.autograd.Function): @staticmethod def forward(ctx, grad_output, input): ctx.save_for_backward( grad_output if input.requires_grad else _null_tensor, input if grad_output.requires_grad else _null_tensor, ) ctx.grad_output_shape = grad_output.shape ctx.input_shape = input.shape # Simple 1x1 convolution => cuBLAS (on both Volta and Ampere). if weight_shape[2:] == stride == dilation == (1, 1) and padding == (0, 0): a = grad_output.reshape(grad_output.shape[0], groups, grad_output.shape[1] // groups, -1).permute(1, 2, 0, 3).flatten(2) b = input.reshape(input.shape[0], groups, input.shape[1] // groups, -1).permute(1, 2, 0, 3).flatten(2) c = (b @ a.transpose(1, 2) if transpose else a @ b.transpose(1, 2)).reshape(weight_shape) return c.contiguous(memory_format=(torch.channels_last if input.stride(1) == 1 else torch.contiguous_format)) # General case => cuDNN. name = 'aten::cudnn_convolution_transpose_backward_weight' if transpose else 'aten::cudnn_convolution_backward_weight' flags = [torch.backends.cudnn.benchmark, torch.backends.cudnn.deterministic, torch.backends.cudnn.allow_tf32] return torch._C._jit_get_operation(name)(weight_shape, grad_output, input, padding, stride, dilation, groups, *flags) @staticmethod def backward(ctx, grad2_grad_weight): grad_output, input = ctx.saved_tensors grad_output_shape = ctx.grad_output_shape input_shape = ctx.input_shape grad2_grad_output = None grad2_input = None if ctx.needs_input_grad[0]: grad2_grad_output = Conv2d.apply(input, grad2_grad_weight, None) assert grad2_grad_output.shape == grad_output_shape if ctx.needs_input_grad[1]: p = calc_output_padding(input_shape=input_shape, output_shape=grad_output_shape) op = _conv2d_gradfix(transpose=(not transpose), weight_shape=weight_shape, output_padding=p, **common_kwargs) grad2_input = op.apply(grad_output, grad2_grad_weight, None) assert grad2_input.shape == input_shape return grad2_grad_output, grad2_input _conv2d_gradfix_cache[key] = Conv2d return Conv2d #----------------------------------------------------------------------------
9,465
46.567839
197
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/ops/upfirdn2d.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Custom PyTorch ops for efficient resampling of 2D images.""" import os import numpy as np import torch from .. import custom_ops from .. import misc from . import conv2d_gradfix #---------------------------------------------------------------------------- _plugin = None def _init(): global _plugin if _plugin is None: _plugin = custom_ops.get_plugin( module_name='upfirdn2d_plugin', sources=['upfirdn2d.cpp', 'upfirdn2d.cu'], headers=['upfirdn2d.h'], source_dir=os.path.dirname(__file__), extra_cuda_cflags=['--use_fast_math'], ) return True def _parse_scaling(scaling): if isinstance(scaling, int): scaling = [scaling, scaling] assert isinstance(scaling, (list, tuple)) assert all(isinstance(x, int) for x in scaling) sx, sy = scaling assert sx >= 1 and sy >= 1 return sx, sy def _parse_padding(padding): if isinstance(padding, int): padding = [padding, padding] assert isinstance(padding, (list, tuple)) assert all(isinstance(x, int) for x in padding) if len(padding) == 2: padx, pady = padding padding = [padx, padx, pady, pady] padx0, padx1, pady0, pady1 = padding return padx0, padx1, pady0, pady1 def _get_filter_size(f): if f is None: return 1, 1 assert isinstance(f, torch.Tensor) and f.ndim in [1, 2] fw = f.shape[-1] fh = f.shape[0] with misc.suppress_tracer_warnings(): fw = int(fw) fh = int(fh) misc.assert_shape(f, [fh, fw][:f.ndim]) assert fw >= 1 and fh >= 1 return fw, fh #---------------------------------------------------------------------------- def setup_filter(f, device=torch.device('cpu'), normalize=True, flip_filter=False, gain=1, separable=None): r"""Convenience function to setup 2D FIR filter for `upfirdn2d()`. Args: f: Torch tensor, numpy array, or python list of the shape `[filter_height, filter_width]` (non-separable), `[filter_taps]` (separable), `[]` (impulse), or `None` (identity). device: Result device (default: cpu). normalize: Normalize the filter so that it retains the magnitude for constant input signal (DC)? (default: True). flip_filter: Flip the filter? (default: False). gain: Overall scaling factor for signal magnitude (default: 1). separable: Return a separable filter? (default: select automatically). Returns: Float32 tensor of the shape `[filter_height, filter_width]` (non-separable) or `[filter_taps]` (separable). """ # Validate. if f is None: f = 1 f = torch.as_tensor(f, dtype=torch.float32) assert f.ndim in [0, 1, 2] assert f.numel() > 0 if f.ndim == 0: f = f[np.newaxis] # Separable? if separable is None: separable = (f.ndim == 1 and f.numel() >= 8) if f.ndim == 1 and not separable: f = f.ger(f) assert f.ndim == (1 if separable else 2) # Apply normalize, flip, gain, and device. if normalize: f /= f.sum() if flip_filter: f = f.flip(list(range(f.ndim))) f = f * (gain ** (f.ndim / 2)) f = f.to(device=device) return f #---------------------------------------------------------------------------- def upfirdn2d(x, f, up=1, down=1, padding=0, flip_filter=False, gain=1, impl='cuda'): r"""Pad, upsample, filter, and downsample a batch of 2D images. Performs the following sequence of operations for each channel: 1. Upsample the image by inserting N-1 zeros after each pixel (`up`). 2. Pad the image with the specified number of zeros on each side (`padding`). Negative padding corresponds to cropping the image. 3. Convolve the image with the specified 2D FIR filter (`f`), shrinking it so that the footprint of all output pixels lies within the input image. 4. Downsample the image by keeping every Nth pixel (`down`). This sequence of operations bears close resemblance to scipy.signal.upfirdn(). The fused op is considerably more efficient than performing the same calculation using standard PyTorch ops. It supports gradients of arbitrary order. Args: x: Float32/float64/float16 input tensor of the shape `[batch_size, num_channels, in_height, in_width]`. f: Float32 FIR filter of the shape `[filter_height, filter_width]` (non-separable), `[filter_taps]` (separable), or `None` (identity). up: Integer upsampling factor. Can be a single int or a list/tuple `[x, y]` (default: 1). down: Integer downsampling factor. Can be a single int or a list/tuple `[x, y]` (default: 1). padding: Padding with respect to the upsampled image. Can be a single number or a list/tuple `[x, y]` or `[x_before, x_after, y_before, y_after]` (default: 0). flip_filter: False = convolution, True = correlation (default: False). gain: Overall scaling factor for signal magnitude (default: 1). impl: Implementation to use. Can be `'ref'` or `'cuda'` (default: `'cuda'`). Returns: Tensor of the shape `[batch_size, num_channels, out_height, out_width]`. """ assert isinstance(x, torch.Tensor) assert impl in ['ref', 'cuda'] if impl == 'cuda' and x.device.type == 'cuda' and _init(): return _upfirdn2d_cuda(up=up, down=down, padding=padding, flip_filter=flip_filter, gain=gain).apply(x, f) return _upfirdn2d_ref(x, f, up=up, down=down, padding=padding, flip_filter=flip_filter, gain=gain) #---------------------------------------------------------------------------- @misc.profiled_function def _upfirdn2d_ref(x, f, up=1, down=1, padding=0, flip_filter=False, gain=1): """Slow reference implementation of `upfirdn2d()` using standard PyTorch ops. """ # Validate arguments. assert isinstance(x, torch.Tensor) and x.ndim == 4 if f is None: f = torch.ones([1, 1], dtype=torch.float32, device=x.device) assert isinstance(f, torch.Tensor) and f.ndim in [1, 2] assert f.dtype == torch.float32 and not f.requires_grad batch_size, num_channels, in_height, in_width = x.shape upx, upy = _parse_scaling(up) downx, downy = _parse_scaling(down) padx0, padx1, pady0, pady1 = _parse_padding(padding) # Check that upsampled buffer is not smaller than the filter. upW = in_width * upx + padx0 + padx1 upH = in_height * upy + pady0 + pady1 assert upW >= f.shape[-1] and upH >= f.shape[0] # Upsample by inserting zeros. x = x.reshape([batch_size, num_channels, in_height, 1, in_width, 1]) x = torch.nn.functional.pad(x, [0, upx - 1, 0, 0, 0, upy - 1]) x = x.reshape([batch_size, num_channels, in_height * upy, in_width * upx]) # Pad or crop. x = torch.nn.functional.pad(x, [max(padx0, 0), max(padx1, 0), max(pady0, 0), max(pady1, 0)]) x = x[:, :, max(-pady0, 0) : x.shape[2] - max(-pady1, 0), max(-padx0, 0) : x.shape[3] - max(-padx1, 0)] # Setup filter. f = f * (gain ** (f.ndim / 2)) f = f.to(x.dtype) if not flip_filter: f = f.flip(list(range(f.ndim))) # Convolve with the filter. f = f[np.newaxis, np.newaxis].repeat([num_channels, 1] + [1] * f.ndim) if f.ndim == 4: x = conv2d_gradfix.conv2d(input=x, weight=f, groups=num_channels) else: x = conv2d_gradfix.conv2d(input=x, weight=f.unsqueeze(2), groups=num_channels) x = conv2d_gradfix.conv2d(input=x, weight=f.unsqueeze(3), groups=num_channels) # Downsample by throwing away pixels. x = x[:, :, ::downy, ::downx] return x #---------------------------------------------------------------------------- _upfirdn2d_cuda_cache = dict() def _upfirdn2d_cuda(up=1, down=1, padding=0, flip_filter=False, gain=1): """Fast CUDA implementation of `upfirdn2d()` using custom ops. """ # Parse arguments. upx, upy = _parse_scaling(up) downx, downy = _parse_scaling(down) padx0, padx1, pady0, pady1 = _parse_padding(padding) # Lookup from cache. key = (upx, upy, downx, downy, padx0, padx1, pady0, pady1, flip_filter, gain) if key in _upfirdn2d_cuda_cache: return _upfirdn2d_cuda_cache[key] # Forward op. class Upfirdn2dCuda(torch.autograd.Function): @staticmethod def forward(ctx, x, f): # pylint: disable=arguments-differ assert isinstance(x, torch.Tensor) and x.ndim == 4 if f is None: f = torch.ones([1, 1], dtype=torch.float32, device=x.device) if f.ndim == 1 and f.shape[0] == 1: f = f.square().unsqueeze(0) # Convert separable-1 into full-1x1. assert isinstance(f, torch.Tensor) and f.ndim in [1, 2] y = x if f.ndim == 2: y = _plugin.upfirdn2d(y, f, upx, upy, downx, downy, padx0, padx1, pady0, pady1, flip_filter, gain) else: y = _plugin.upfirdn2d(y, f.unsqueeze(0), upx, 1, downx, 1, padx0, padx1, 0, 0, flip_filter, 1.0) y = _plugin.upfirdn2d(y, f.unsqueeze(1), 1, upy, 1, downy, 0, 0, pady0, pady1, flip_filter, gain) ctx.save_for_backward(f) ctx.x_shape = x.shape return y @staticmethod def backward(ctx, dy): # pylint: disable=arguments-differ f, = ctx.saved_tensors _, _, ih, iw = ctx.x_shape _, _, oh, ow = dy.shape fw, fh = _get_filter_size(f) p = [ fw - padx0 - 1, iw * upx - ow * downx + padx0 - upx + 1, fh - pady0 - 1, ih * upy - oh * downy + pady0 - upy + 1, ] dx = None df = None if ctx.needs_input_grad[0]: dx = _upfirdn2d_cuda(up=down, down=up, padding=p, flip_filter=(not flip_filter), gain=gain).apply(dy, f) assert not ctx.needs_input_grad[1] return dx, df # Add to cache. _upfirdn2d_cuda_cache[key] = Upfirdn2dCuda return Upfirdn2dCuda #---------------------------------------------------------------------------- def filter2d(x, f, padding=0, flip_filter=False, gain=1, impl='cuda'): r"""Filter a batch of 2D images using the given 2D FIR filter. By default, the result is padded so that its shape matches the input. User-specified padding is applied on top of that, with negative values indicating cropping. Pixels outside the image are assumed to be zero. Args: x: Float32/float64/float16 input tensor of the shape `[batch_size, num_channels, in_height, in_width]`. f: Float32 FIR filter of the shape `[filter_height, filter_width]` (non-separable), `[filter_taps]` (separable), or `None` (identity). padding: Padding with respect to the output. Can be a single number or a list/tuple `[x, y]` or `[x_before, x_after, y_before, y_after]` (default: 0). flip_filter: False = convolution, True = correlation (default: False). gain: Overall scaling factor for signal magnitude (default: 1). impl: Implementation to use. Can be `'ref'` or `'cuda'` (default: `'cuda'`). Returns: Tensor of the shape `[batch_size, num_channels, out_height, out_width]`. """ padx0, padx1, pady0, pady1 = _parse_padding(padding) fw, fh = _get_filter_size(f) p = [ padx0 + fw // 2, padx1 + (fw - 1) // 2, pady0 + fh // 2, pady1 + (fh - 1) // 2, ] return upfirdn2d(x, f, padding=p, flip_filter=flip_filter, gain=gain, impl=impl) #---------------------------------------------------------------------------- def upsample2d(x, f, up=2, padding=0, flip_filter=False, gain=1, impl='cuda'): r"""Upsample a batch of 2D images using the given 2D FIR filter. By default, the result is padded so that its shape is a multiple of the input. User-specified padding is applied on top of that, with negative values indicating cropping. Pixels outside the image are assumed to be zero. Args: x: Float32/float64/float16 input tensor of the shape `[batch_size, num_channels, in_height, in_width]`. f: Float32 FIR filter of the shape `[filter_height, filter_width]` (non-separable), `[filter_taps]` (separable), or `None` (identity). up: Integer upsampling factor. Can be a single int or a list/tuple `[x, y]` (default: 1). padding: Padding with respect to the output. Can be a single number or a list/tuple `[x, y]` or `[x_before, x_after, y_before, y_after]` (default: 0). flip_filter: False = convolution, True = correlation (default: False). gain: Overall scaling factor for signal magnitude (default: 1). impl: Implementation to use. Can be `'ref'` or `'cuda'` (default: `'cuda'`). Returns: Tensor of the shape `[batch_size, num_channels, out_height, out_width]`. """ upx, upy = _parse_scaling(up) padx0, padx1, pady0, pady1 = _parse_padding(padding) fw, fh = _get_filter_size(f) p = [ padx0 + (fw + upx - 1) // 2, padx1 + (fw - upx) // 2, pady0 + (fh + upy - 1) // 2, pady1 + (fh - upy) // 2, ] return upfirdn2d(x, f, up=up, padding=p, flip_filter=flip_filter, gain=gain*upx*upy, impl=impl) #---------------------------------------------------------------------------- def downsample2d(x, f, down=2, padding=0, flip_filter=False, gain=1, impl='cuda'): r"""Downsample a batch of 2D images using the given 2D FIR filter. By default, the result is padded so that its shape is a fraction of the input. User-specified padding is applied on top of that, with negative values indicating cropping. Pixels outside the image are assumed to be zero. Args: x: Float32/float64/float16 input tensor of the shape `[batch_size, num_channels, in_height, in_width]`. f: Float32 FIR filter of the shape `[filter_height, filter_width]` (non-separable), `[filter_taps]` (separable), or `None` (identity). down: Integer downsampling factor. Can be a single int or a list/tuple `[x, y]` (default: 1). padding: Padding with respect to the input. Can be a single number or a list/tuple `[x, y]` or `[x_before, x_after, y_before, y_after]` (default: 0). flip_filter: False = convolution, True = correlation (default: False). gain: Overall scaling factor for signal magnitude (default: 1). impl: Implementation to use. Can be `'ref'` or `'cuda'` (default: `'cuda'`). Returns: Tensor of the shape `[batch_size, num_channels, out_height, out_width]`. """ downx, downy = _parse_scaling(down) padx0, padx1, pady0, pady1 = _parse_padding(padding) fw, fh = _get_filter_size(f) p = [ padx0 + (fw - downx + 1) // 2, padx1 + (fw - downx) // 2, pady0 + (fh - downy + 1) // 2, pady1 + (fh - downy) // 2, ] return upfirdn2d(x, f, down=down, padding=p, flip_filter=flip_filter, gain=gain, impl=impl) #----------------------------------------------------------------------------
16,392
41.033333
120
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/ops/filtered_lrelu.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. import os import numpy as np import torch import warnings from .. import custom_ops from .. import misc from . import upfirdn2d from . import bias_act #---------------------------------------------------------------------------- _plugin = None def _init(): global _plugin if _plugin is None: _plugin = custom_ops.get_plugin( module_name='filtered_lrelu_plugin', sources=['filtered_lrelu.cpp', 'filtered_lrelu_wr.cu', 'filtered_lrelu_rd.cu', 'filtered_lrelu_ns.cu'], headers=['filtered_lrelu.h', 'filtered_lrelu.cu'], source_dir=os.path.dirname(__file__), extra_cuda_cflags=['--use_fast_math'], ) return True def _get_filter_size(f): if f is None: return 1, 1 assert isinstance(f, torch.Tensor) assert 1 <= f.ndim <= 2 return f.shape[-1], f.shape[0] # width, height def _parse_padding(padding): if isinstance(padding, int): padding = [padding, padding] assert isinstance(padding, (list, tuple)) assert all(isinstance(x, (int, np.integer)) for x in padding) padding = [int(x) for x in padding] if len(padding) == 2: px, py = padding padding = [px, px, py, py] px0, px1, py0, py1 = padding return px0, px1, py0, py1 #---------------------------------------------------------------------------- def filtered_lrelu(x, fu=None, fd=None, b=None, up=1, down=1, padding=0, gain=np.sqrt(2), slope=0.2, clamp=None, flip_filter=False, impl='cuda'): r"""Filtered leaky ReLU for a batch of 2D images. Performs the following sequence of operations for each channel: 1. Add channel-specific bias if provided (`b`). 2. Upsample the image by inserting N-1 zeros after each pixel (`up`). 3. Pad the image with the specified number of zeros on each side (`padding`). Negative padding corresponds to cropping the image. 4. Convolve the image with the specified upsampling FIR filter (`fu`), shrinking it so that the footprint of all output pixels lies within the input image. 5. Multiply each value by the provided gain factor (`gain`). 6. Apply leaky ReLU activation function to each value. 7. Clamp each value between -clamp and +clamp, if `clamp` parameter is provided. 8. Convolve the image with the specified downsampling FIR filter (`fd`), shrinking it so that the footprint of all output pixels lies within the input image. 9. Downsample the image by keeping every Nth pixel (`down`). The fused op is considerably more efficient than performing the same calculation using standard PyTorch ops. It supports gradients of arbitrary order. Args: x: Float32/float16/float64 input tensor of the shape `[batch_size, num_channels, in_height, in_width]`. fu: Float32 upsampling FIR filter of the shape `[filter_height, filter_width]` (non-separable), `[filter_taps]` (separable), or `None` (identity). fd: Float32 downsampling FIR filter of the shape `[filter_height, filter_width]` (non-separable), `[filter_taps]` (separable), or `None` (identity). b: Bias vector, or `None` to disable. Must be a 1D tensor of the same type as `x`. The length of vector must must match the channel dimension of `x`. up: Integer upsampling factor (default: 1). down: Integer downsampling factor. (default: 1). padding: Padding with respect to the upsampled image. Can be a single number or a list/tuple `[x, y]` or `[x_before, x_after, y_before, y_after]` (default: 0). gain: Overall scaling factor for signal magnitude (default: sqrt(2)). slope: Slope on the negative side of leaky ReLU (default: 0.2). clamp: Maximum magnitude for leaky ReLU output (default: None). flip_filter: False = convolution, True = correlation (default: False). impl: Implementation to use. Can be `'ref'` or `'cuda'` (default: `'cuda'`). Returns: Tensor of the shape `[batch_size, num_channels, out_height, out_width]`. """ assert isinstance(x, torch.Tensor) assert impl in ['ref', 'cuda'] if impl == 'cuda' and x.device.type == 'cuda' and _init(): return _filtered_lrelu_cuda(up=up, down=down, padding=padding, gain=gain, slope=slope, clamp=clamp, flip_filter=flip_filter).apply(x, fu, fd, b, None, 0, 0) return _filtered_lrelu_ref(x, fu=fu, fd=fd, b=b, up=up, down=down, padding=padding, gain=gain, slope=slope, clamp=clamp, flip_filter=flip_filter) #---------------------------------------------------------------------------- @misc.profiled_function def _filtered_lrelu_ref(x, fu=None, fd=None, b=None, up=1, down=1, padding=0, gain=np.sqrt(2), slope=0.2, clamp=None, flip_filter=False): """Slow and memory-inefficient reference implementation of `filtered_lrelu()` using existing `upfirdn2n()` and `bias_act()` ops. """ assert isinstance(x, torch.Tensor) and x.ndim == 4 fu_w, fu_h = _get_filter_size(fu) fd_w, fd_h = _get_filter_size(fd) if b is not None: assert isinstance(b, torch.Tensor) and b.dtype == x.dtype misc.assert_shape(b, [x.shape[1]]) assert isinstance(up, int) and up >= 1 assert isinstance(down, int) and down >= 1 px0, px1, py0, py1 = _parse_padding(padding) assert gain == float(gain) and gain > 0 assert slope == float(slope) and slope >= 0 assert clamp is None or (clamp == float(clamp) and clamp >= 0) # Calculate output size. batch_size, channels, in_h, in_w = x.shape in_dtype = x.dtype out_w = (in_w * up + (px0 + px1) - (fu_w - 1) - (fd_w - 1) + (down - 1)) // down out_h = (in_h * up + (py0 + py1) - (fu_h - 1) - (fd_h - 1) + (down - 1)) // down # Compute using existing ops. x = bias_act.bias_act(x=x, b=b) # Apply bias. x = upfirdn2d.upfirdn2d(x=x, f=fu, up=up, padding=[px0, px1, py0, py1], gain=up**2, flip_filter=flip_filter) # Upsample. x = bias_act.bias_act(x=x, act='lrelu', alpha=slope, gain=gain, clamp=clamp) # Bias, leaky ReLU, clamp. x = upfirdn2d.upfirdn2d(x=x, f=fd, down=down, flip_filter=flip_filter) # Downsample. # Check output shape & dtype. misc.assert_shape(x, [batch_size, channels, out_h, out_w]) assert x.dtype == in_dtype return x #---------------------------------------------------------------------------- _filtered_lrelu_cuda_cache = dict() def _filtered_lrelu_cuda(up=1, down=1, padding=0, gain=np.sqrt(2), slope=0.2, clamp=None, flip_filter=False): """Fast CUDA implementation of `filtered_lrelu()` using custom ops. """ assert isinstance(up, int) and up >= 1 assert isinstance(down, int) and down >= 1 px0, px1, py0, py1 = _parse_padding(padding) assert gain == float(gain) and gain > 0 gain = float(gain) assert slope == float(slope) and slope >= 0 slope = float(slope) assert clamp is None or (clamp == float(clamp) and clamp >= 0) clamp = float(clamp if clamp is not None else 'inf') # Lookup from cache. key = (up, down, px0, px1, py0, py1, gain, slope, clamp, flip_filter) if key in _filtered_lrelu_cuda_cache: return _filtered_lrelu_cuda_cache[key] # Forward op. class FilteredLReluCuda(torch.autograd.Function): @staticmethod def forward(ctx, x, fu, fd, b, si, sx, sy): # pylint: disable=arguments-differ assert isinstance(x, torch.Tensor) and x.ndim == 4 # Replace empty up/downsample kernels with full 1x1 kernels (faster than separable). if fu is None: fu = torch.ones([1, 1], dtype=torch.float32, device=x.device) if fd is None: fd = torch.ones([1, 1], dtype=torch.float32, device=x.device) assert 1 <= fu.ndim <= 2 assert 1 <= fd.ndim <= 2 # Replace separable 1x1 kernels with full 1x1 kernels when scale factor is 1. if up == 1 and fu.ndim == 1 and fu.shape[0] == 1: fu = fu.square()[None] if down == 1 and fd.ndim == 1 and fd.shape[0] == 1: fd = fd.square()[None] # Missing sign input tensor. if si is None: si = torch.empty([0]) # Missing bias tensor. if b is None: b = torch.zeros([x.shape[1]], dtype=x.dtype, device=x.device) # Construct internal sign tensor only if gradients are needed. write_signs = (si.numel() == 0) and (x.requires_grad or b.requires_grad) # Warn if input storage strides are not in decreasing order due to e.g. channels-last layout. strides = [x.stride(i) for i in range(x.ndim) if x.size(i) > 1] if any(a < b for a, b in zip(strides[:-1], strides[1:])): warnings.warn("low-performance memory layout detected in filtered_lrelu input", RuntimeWarning) # Call C++/Cuda plugin if datatype is supported. if x.dtype in [torch.float16, torch.float32]: if torch.cuda.current_stream(x.device) != torch.cuda.default_stream(x.device): warnings.warn("filtered_lrelu called with non-default cuda stream but concurrent execution is not supported", RuntimeWarning) y, so, return_code = _plugin.filtered_lrelu(x, fu, fd, b, si, up, down, px0, px1, py0, py1, sx, sy, gain, slope, clamp, flip_filter, write_signs) else: return_code = -1 # No Cuda kernel found? Fall back to generic implementation. Still more memory efficient than the reference implementation because # only the bit-packed sign tensor is retained for gradient computation. if return_code < 0: warnings.warn("filtered_lrelu called with parameters that have no optimized CUDA kernel, using generic fallback", RuntimeWarning) y = x.add(b.unsqueeze(-1).unsqueeze(-1)) # Add bias. y = upfirdn2d.upfirdn2d(x=y, f=fu, up=up, padding=[px0, px1, py0, py1], gain=up**2, flip_filter=flip_filter) # Upsample. so = _plugin.filtered_lrelu_act_(y, si, sx, sy, gain, slope, clamp, write_signs) # Activation function and sign handling. Modifies y in-place. y = upfirdn2d.upfirdn2d(x=y, f=fd, down=down, flip_filter=flip_filter) # Downsample. # Prepare for gradient computation. ctx.save_for_backward(fu, fd, (si if si.numel() else so)) ctx.x_shape = x.shape ctx.y_shape = y.shape ctx.s_ofs = sx, sy return y @staticmethod def backward(ctx, dy): # pylint: disable=arguments-differ fu, fd, si = ctx.saved_tensors _, _, xh, xw = ctx.x_shape _, _, yh, yw = ctx.y_shape sx, sy = ctx.s_ofs dx = None # 0 dfu = None; assert not ctx.needs_input_grad[1] dfd = None; assert not ctx.needs_input_grad[2] db = None # 3 dsi = None; assert not ctx.needs_input_grad[4] dsx = None; assert not ctx.needs_input_grad[5] dsy = None; assert not ctx.needs_input_grad[6] if ctx.needs_input_grad[0] or ctx.needs_input_grad[3]: pp = [ (fu.shape[-1] - 1) + (fd.shape[-1] - 1) - px0, xw * up - yw * down + px0 - (up - 1), (fu.shape[0] - 1) + (fd.shape[0] - 1) - py0, xh * up - yh * down + py0 - (up - 1), ] gg = gain * (up ** 2) / (down ** 2) ff = (not flip_filter) sx = sx - (fu.shape[-1] - 1) + px0 sy = sy - (fu.shape[0] - 1) + py0 dx = _filtered_lrelu_cuda(up=down, down=up, padding=pp, gain=gg, slope=slope, clamp=None, flip_filter=ff).apply(dy, fd, fu, None, si, sx, sy) if ctx.needs_input_grad[3]: db = dx.sum([0, 2, 3]) return dx, dfu, dfd, db, dsi, dsx, dsy # Add to cache. _filtered_lrelu_cuda_cache[key] = FilteredLReluCuda return FilteredLReluCuda #----------------------------------------------------------------------------
12,884
45.854545
164
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/ops/conv2d_resample.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """2D convolution with optional up/downsampling.""" import torch from .. import misc from . import conv2d_gradfix from . import upfirdn2d from .upfirdn2d import _parse_padding from .upfirdn2d import _get_filter_size #---------------------------------------------------------------------------- def _get_weight_shape(w): with misc.suppress_tracer_warnings(): # this value will be treated as a constant shape = [int(sz) for sz in w.shape] misc.assert_shape(w, shape) return shape #---------------------------------------------------------------------------- def _conv2d_wrapper(x, w, stride=1, padding=0, groups=1, transpose=False, flip_weight=True): """Wrapper for the underlying `conv2d()` and `conv_transpose2d()` implementations. """ _out_channels, _in_channels_per_group, kh, kw = _get_weight_shape(w) # Flip weight if requested. # Note: conv2d() actually performs correlation (flip_weight=True) not convolution (flip_weight=False). if not flip_weight and (kw > 1 or kh > 1): w = w.flip([2, 3]) # Execute using conv2d_gradfix. op = conv2d_gradfix.conv_transpose2d if transpose else conv2d_gradfix.conv2d return op(x, w, stride=stride, padding=padding, groups=groups) #---------------------------------------------------------------------------- @misc.profiled_function def conv2d_resample(x, w, f=None, up=1, down=1, padding=0, groups=1, flip_weight=True, flip_filter=False): r"""2D convolution with optional up/downsampling. Padding is performed only once at the beginning, not between the operations. Args: x: Input tensor of shape `[batch_size, in_channels, in_height, in_width]`. w: Weight tensor of shape `[out_channels, in_channels//groups, kernel_height, kernel_width]`. f: Low-pass filter for up/downsampling. Must be prepared beforehand by calling upfirdn2d.setup_filter(). None = identity (default). up: Integer upsampling factor (default: 1). down: Integer downsampling factor (default: 1). padding: Padding with respect to the upsampled image. Can be a single number or a list/tuple `[x, y]` or `[x_before, x_after, y_before, y_after]` (default: 0). groups: Split input channels into N groups (default: 1). flip_weight: False = convolution, True = correlation (default: True). flip_filter: False = convolution, True = correlation (default: False). Returns: Tensor of the shape `[batch_size, num_channels, out_height, out_width]`. """ # Validate arguments. assert isinstance(x, torch.Tensor) and (x.ndim == 4) assert isinstance(w, torch.Tensor) and (w.ndim == 4) and (w.dtype == x.dtype) assert f is None or (isinstance(f, torch.Tensor) and f.ndim in [1, 2] and f.dtype == torch.float32) assert isinstance(up, int) and (up >= 1) assert isinstance(down, int) and (down >= 1) assert isinstance(groups, int) and (groups >= 1) out_channels, in_channels_per_group, kh, kw = _get_weight_shape(w) fw, fh = _get_filter_size(f) px0, px1, py0, py1 = _parse_padding(padding) # Adjust padding to account for up/downsampling. if up > 1: px0 += (fw + up - 1) // 2 px1 += (fw - up) // 2 py0 += (fh + up - 1) // 2 py1 += (fh - up) // 2 if down > 1: px0 += (fw - down + 1) // 2 px1 += (fw - down) // 2 py0 += (fh - down + 1) // 2 py1 += (fh - down) // 2 # Fast path: 1x1 convolution with downsampling only => downsample first, then convolve. if kw == 1 and kh == 1 and (down > 1 and up == 1): x = upfirdn2d.upfirdn2d(x=x, f=f, down=down, padding=[px0,px1,py0,py1], flip_filter=flip_filter) x = _conv2d_wrapper(x=x, w=w, groups=groups, flip_weight=flip_weight) return x # Fast path: 1x1 convolution with upsampling only => convolve first, then upsample. if kw == 1 and kh == 1 and (up > 1 and down == 1): x = _conv2d_wrapper(x=x, w=w, groups=groups, flip_weight=flip_weight) x = upfirdn2d.upfirdn2d(x=x, f=f, up=up, padding=[px0,px1,py0,py1], gain=up**2, flip_filter=flip_filter) return x # Fast path: downsampling only => use strided convolution. if down > 1 and up == 1: x = upfirdn2d.upfirdn2d(x=x, f=f, padding=[px0,px1,py0,py1], flip_filter=flip_filter) x = _conv2d_wrapper(x=x, w=w, stride=down, groups=groups, flip_weight=flip_weight) return x # Fast path: upsampling with optional downsampling => use transpose strided convolution. if up > 1: if groups == 1: w = w.transpose(0, 1) else: w = w.reshape(groups, out_channels // groups, in_channels_per_group, kh, kw) w = w.transpose(1, 2) w = w.reshape(groups * in_channels_per_group, out_channels // groups, kh, kw) px0 -= kw - 1 px1 -= kw - up py0 -= kh - 1 py1 -= kh - up pxt = max(min(-px0, -px1), 0) pyt = max(min(-py0, -py1), 0) x = _conv2d_wrapper(x=x, w=w, stride=up, padding=[pyt,pxt], groups=groups, transpose=True, flip_weight=(not flip_weight)) x = upfirdn2d.upfirdn2d(x=x, f=f, padding=[px0+pxt,px1+pxt,py0+pyt,py1+pyt], gain=up**2, flip_filter=flip_filter) if down > 1: x = upfirdn2d.upfirdn2d(x=x, f=f, down=down, flip_filter=flip_filter) return x # Fast path: no up/downsampling, padding supported by the underlying implementation => use plain conv2d. if up == 1 and down == 1: if px0 == px1 and py0 == py1 and px0 >= 0 and py0 >= 0: return _conv2d_wrapper(x=x, w=w, padding=[py0,px0], groups=groups, flip_weight=flip_weight) # Fallback: Generic reference implementation. x = upfirdn2d.upfirdn2d(x=x, f=(f if up > 1 else None), up=up, padding=[px0,px1,py0,py1], gain=up**2, flip_filter=flip_filter) x = _conv2d_wrapper(x=x, w=w, groups=groups, flip_weight=flip_weight) if down > 1: x = upfirdn2d.upfirdn2d(x=x, f=f, down=down, flip_filter=flip_filter) return x #----------------------------------------------------------------------------
6,765
45.986111
130
py
DiffProxy
DiffProxy-main/stylegan/torch_utils/ops/fma.py
# Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. """Fused multiply-add, with slightly faster gradients than `torch.addcmul()`.""" import torch #---------------------------------------------------------------------------- def fma(a, b, c): # => a * b + c return _FusedMultiplyAdd.apply(a, b, c) #---------------------------------------------------------------------------- class _FusedMultiplyAdd(torch.autograd.Function): # a * b + c @staticmethod def forward(ctx, a, b, c): # pylint: disable=arguments-differ out = torch.addcmul(c, a, b) ctx.save_for_backward(a, b) ctx.c_shape = c.shape return out @staticmethod def backward(ctx, dout): # pylint: disable=arguments-differ a, b = ctx.saved_tensors c_shape = ctx.c_shape da = None db = None dc = None if ctx.needs_input_grad[0]: da = _unbroadcast(dout * b, a.shape) if ctx.needs_input_grad[1]: db = _unbroadcast(dout * a, b.shape) if ctx.needs_input_grad[2]: dc = _unbroadcast(dout, c_shape) return da, db, dc #---------------------------------------------------------------------------- def _unbroadcast(x, shape): extra_dims = x.ndim - len(shape) assert extra_dims >= 0 dim = [i for i in range(x.ndim) if x.shape[i] > 1 and (i < extra_dims or shape[i - extra_dims] == 1)] if len(dim): x = x.sum(dim=dim, keepdim=True) if extra_dims: x = x.reshape(-1, *x.shape[extra_dims+1:]) assert x.shape == shape return x #----------------------------------------------------------------------------
2,047
32.57377
105
py