code
stringlengths
17
6.64M
def lengths_to_mask(lengths: List[int], device: torch.device, max_len: int=None) -> Tensor: lengths = torch.tensor(lengths, device=device) max_len = (max_len if max_len else max(lengths)) mask = (torch.arange(max_len, device=device).expand(len(lengths), max_len) < lengths.unsqueeze(1)) return mask
def detach_to_numpy(tensor): return tensor.detach().cpu().numpy()
def remove_padding(tensors, lengths): return [tensor[:tensor_length] for (tensor, tensor_length) in zip(tensors, lengths)]
def nfeats_of(rottype): if (rottype in ['rotvec', 'axisangle']): return 3 elif (rottype in ['rotquat', 'quaternion']): return 4 elif (rottype in ['rot6d', '6drot', 'rotation6d']): return 6 elif (rottype in ['rotmat']): return 9 else: return TypeError("This rotation type doesn't have features.")
def axis_angle_to(newtype, rotations): if (newtype in ['matrix']): rotations = geometry.axis_angle_to_matrix(rotations) return rotations elif (newtype in ['rotmat']): rotations = geometry.axis_angle_to_matrix(rotations) rotations = matrix_to('rotmat', rotations) return rotations elif (newtype in ['rot6d', '6drot', 'rotation6d']): rotations = geometry.axis_angle_to_matrix(rotations) rotations = matrix_to('rot6d', rotations) return rotations elif (newtype in ['rotquat', 'quaternion']): rotations = geometry.axis_angle_to_quaternion(rotations) return rotations elif (newtype in ['rotvec', 'axisangle']): return rotations else: raise NotImplementedError
def matrix_to(newtype, rotations): if (newtype in ['matrix']): return rotations if (newtype in ['rotmat']): rotations = rotations.reshape((*rotations.shape[:(- 2)], 9)) return rotations elif (newtype in ['rot6d', '6drot', 'rotation6d']): rotations = geometry.matrix_to_rotation_6d(rotations) return rotations elif (newtype in ['rotquat', 'quaternion']): rotations = geometry.matrix_to_quaternion(rotations) return rotations elif (newtype in ['rotvec', 'axisangle']): rotations = geometry.matrix_to_axis_angle(rotations) return rotations else: raise NotImplementedError
def to_matrix(oldtype, rotations): if (oldtype in ['matrix']): return rotations if (oldtype in ['rotmat']): rotations = rotations.reshape((*rotations.shape[:(- 2)], 3, 3)) return rotations elif (oldtype in ['rot6d', '6drot', 'rotation6d']): rotations = geometry.rotation_6d_to_matrix(rotations) return rotations elif (oldtype in ['rotquat', 'quaternion']): rotations = geometry.quaternion_to_matrix(rotations) return rotations elif (oldtype in ['rotvec', 'axisangle']): rotations = geometry.axis_angle_to_matrix(rotations) return rotations else: raise NotImplementedError
def subsample(num_frames, last_framerate, new_framerate): step = int((last_framerate / new_framerate)) assert (step >= 1) frames = np.arange(0, num_frames, step) return frames
def upsample(motion, last_framerate, new_framerate): step = int((new_framerate / last_framerate)) assert (step >= 1) alpha = np.linspace(0, 1, (step + 1)) last = np.einsum('l,...->l...', (1 - alpha), motion[:(- 1)]) new = np.einsum('l,...->l...', alpha, motion[1:]) chuncks = (last + new)[:(- 1)] output = np.concatenate(chuncks.swapaxes(1, 0)) output = np.concatenate((output, motion[[(- 1)]])) return output
def lengths_to_mask(lengths): max_len = max(lengths) mask = (torch.arange(max_len, device=lengths.device).expand(len(lengths), max_len) < lengths.unsqueeze(1)) return mask
def collate_tensors(batch): dims = batch[0].dim() max_size = [max([b.size(i) for b in batch]) for i in range(dims)] size = ((len(batch),) + tuple(max_size)) canvas = batch[0].new_zeros(size=size) for (i, b) in enumerate(batch): sub_tensor = canvas[i] for d in range(dims): sub_tensor = sub_tensor.narrow(d, 0, b.size(d)) sub_tensor.add_(b) return canvas
def collate(batch): databatch = [b[0] for b in batch] labelbatch = [b[1] for b in batch] lenbatch = [len(b[0][0][0]) for b in batch] databatchTensor = collate_tensors(databatch) labelbatchTensor = torch.as_tensor(labelbatch) lenbatchTensor = torch.as_tensor(lenbatch) maskbatchTensor = lengths_to_mask(lenbatchTensor) batch = {'x': databatchTensor, 'y': labelbatchTensor, 'mask': maskbatchTensor, 'lengths': lenbatchTensor} return batch
def collate_data3d_slow(batch): batchTensor = {} for key in batch[0].keys(): databatch = [b[key] for b in batch] batchTensor[key] = collate_tensors(databatch) batch = batchTensor return batch
def collate_data3d(batch): batchTensor = {} for key in batch[0].keys(): databatch = [b[key] for b in batch] if (key == 'paths'): batchTensor[key] = databatch else: batchTensor[key] = torch.stack(databatch, axis=0) batch = batchTensor return batch
def main(): '\n get input text\n ToDo skip if user input text in command\n current tasks:\n 1 text 2 mtion\n 2 motion transfer\n 3 random sampling\n 4 reconstruction\n\n ToDo \n 1 use one funtion for all expoert\n 2 fitting smpl and export fbx in this file\n 3 \n\n ' cfg = parse_args(phase='demo') cfg.FOLDER = cfg.TEST.FOLDER cfg.Name = ('demo--' + cfg.NAME) logger = create_logger(cfg, phase='demo') if cfg.DEMO.EXAMPLE: from GraphMotion.utils.demo_utils import load_example_input (text, length) = load_example_input(cfg.DEMO.EXAMPLE) task = 'Example' elif cfg.DEMO.TASK: task = cfg.DEMO.TASK text = None else: task = 'Keyborad_input' text = input('Please enter texts, none for random latent sampling:') length = input('Please enter length, range 16~196, e.g. 50, none for random latent sampling:') if text: motion_path = input('Please enter npy_path for motion transfer, none for skip:') if (text and (not motion_path)): cfg.DEMO.MOTION_TRANSFER = False elif (text and motion_path): joints = np.load(motion_path) frames = subsample(len(joints), last_framerate=cfg.DEMO.FRAME_RATE, new_framerate=cfg.DATASET.KIT.FRAME_RATE) joints_sample = torch.from_numpy(joints[frames]).float() features = model.transforms.joints2jfeats(joints_sample[None]) motion = xx cfg.DEMO.MOTION_TRANSFER = True length = (200 if (not length) else length) length = [int(length)] text = [text] output_dir = Path(os.path.join(cfg.FOLDER, str(cfg.model.model_type), str(cfg.NAME), ('samples_' + cfg.TIME))) output_dir.mkdir(parents=True, exist_ok=True) if (cfg.ACCELERATOR == 'gpu'): os.environ['CUDA_VISIBLE_DEVICES'] = ','.join((str(x) for x in cfg.DEVICE)) device = torch.device('cuda') dataset = get_datasets(cfg, logger=logger, phase='test')[0] total_time = time.time() model = get_model(cfg, dataset) if (not text): logger.info(f'Begin specific task{task}') logger.info('Loading checkpoints from {}'.format(cfg.TEST.CHECKPOINTS)) state_dict = torch.load(cfg.TEST.CHECKPOINTS, map_location='cpu')['state_dict'] model.load_state_dict(state_dict, strict=True) logger.info('model {} loaded'.format(cfg.model.model_type)) model.sample_mean = cfg.TEST.MEAN model.fact = cfg.TEST.FACT model.to(device) model.eval() mld_time = time.time() with torch.no_grad(): rep_lst = [] rep_ref_lst = [] texts_lst = [] if text: batch = {'length': length, 'text': text} for rep in range(cfg.DEMO.REPLICATION): if cfg.DEMO.MOTION_TRANSFER: joints = model.forward_motion_style_transfer(batch) else: joints = model(batch) infer_time = (time.time() - mld_time) num_batch = 1 num_all_frame = sum(batch['length']) num_ave_frame = (sum(batch['length']) / len(batch['length'])) nsample = len(joints) id = 0 for i in range(nsample): npypath = str((output_dir / f'{task}_{length[i]}_batch{id}_{i}.npy')) with open(npypath.replace('.npy', '.txt'), 'w') as text_file: text_file.write(batch['text'][i]) np.save(npypath, joints[i].detach().cpu().numpy()) logger.info(f'''Motions are generated here: {npypath}''') if cfg.DEMO.OUTALL: rep_lst.append(joints) texts_lst.append(batch['text']) if cfg.DEMO.OUTALL: grouped_lst = [] for n in range(nsample): grouped_lst.append(torch.cat([r[n][None] for r in rep_lst], dim=0)[None]) combinedOut = torch.cat(grouped_lst, dim=0) try: npypath = str((output_dir / f'{task}_{length[i]}_all.npy')) np.save(npypath, combinedOut.detach().cpu().numpy()) with open(npypath.replace('npy', 'txt'), 'w') as text_file: for texts in texts_lst: for text in texts: text_file.write(text) text_file.write('\n') logger.info(f'''All reconstructed motions are generated here: {npypath}''') except: raise ValueError('Lengths of motions are different, so we cannot save all motions in one file.') if (not text): if (task == 'random_sampling'): text = 'random sampling' length = 196 (nsample, latent_dim) = (500, 256) batch = {'latent': torch.randn(1, nsample, latent_dim, device=model.device), 'length': ([int(length)] * nsample)} joints = model.gen_from_latent(batch) (num_batch, num_all_frame, num_ave_frame) = (100, (100 * 196), 196) infer_time = (time.time() - mld_time) for i in range(nsample): npypath = (output_dir / f"{text.split(' ')[0]}_{length}_{i}.npy") np.save(npypath, joints[i].detach().cpu().numpy()) logger.info(f'''Motions are generated here: {npypath}''') elif (task in ['reconstrucion', 'text_motion']): for rep in range(cfg.DEMO.REPLICATION): logger.info(f'Replication {rep}') joints_lst = [] ref_lst = [] for (id, batch) in enumerate(dataset.test_dataloader()): if (task == 'reconstrucion'): batch['motion'] = batch['motion'].to(device) length = batch['length'] (joints, joints_ref) = model.recon_from_motion(batch) elif (task == 'text_motion'): batch['motion'] = batch['motion'].to(device) (joints, joints_ref) = model(batch, return_ref=True) nsample = len(joints) length = batch['length'] for i in range(nsample): npypath = str((output_dir / f'{task}_{length[i]}_batch{id}_{i}_{rep}.npy')) np.save(npypath, joints[i].detach().cpu().numpy()) np.save(npypath.replace('.npy', '_ref.npy'), joints_ref[i].detach().cpu().numpy()) with open(npypath.replace('.npy', '.txt'), 'w') as text_file: text_file.write(batch['text'][i]) logger.info(f'''Reconstructed motions are generated here: {npypath}''') else: raise ValueError(f'Not support task {task}, only support random_sampling, reconstrucion, text_motion') total_time = (time.time() - total_time) print(f'MLD Infer time - This/Ave batch: {(infer_time / num_batch):.2f}') print(f'MLD Infer FPS - Total batch: {(num_all_frame / infer_time):.2f}') print(f'MLD Infer time - This/Ave batch: {(infer_time / num_batch):.2f}') print(f'MLD Infer FPS - Total batch: {(num_all_frame / infer_time):.2f}') print(f'MLD Infer FPS - Running Poses Per Second: {((num_ave_frame * infer_time) / num_batch):.2f}') print(f'MLD Infer FPS - {(num_all_frame / infer_time):.2f}s') print(f'MLD Infer FPS - Running Poses Per Second: {((num_ave_frame * infer_time) / num_batch):.2f}') print(f'MLD Infer FPS - time for 100 Poses: {((infer_time / (num_batch * num_ave_frame)) * 100):.2f}') print(f'Total time spent: {total_time:.2f} seconds (including model loading time and exporting time).') if cfg.DEMO.RENDER: from GraphMotion.utils.demo_utils import render_batch blenderpath = cfg.RENDER.BLENDER_PATH render_batch(os.path.dirname(npypath), execute_python=blenderpath, mode='sequence') logger.info(f'''Motions are rendered here: {os.path.dirname(npypath)}''')
class ProgressLogger(Callback): def __init__(self, metric_monitor: dict, precision: int=3): self.metric_monitor = metric_monitor self.precision = precision def on_train_start(self, trainer: Trainer, pl_module: LightningModule, **kwargs) -> None: logger.info('Training started') def on_train_end(self, trainer: Trainer, pl_module: LightningModule, **kwargs) -> None: logger.info('Training done') def on_validation_epoch_end(self, trainer: Trainer, pl_module: LightningModule, **kwargs) -> None: if trainer.sanity_checking: logger.info('Sanity checking ok.') def on_train_epoch_end(self, trainer: Trainer, pl_module: LightningModule, padding=False, **kwargs) -> None: metric_format = f'{{:.{self.precision}e}}' line = f'Epoch {trainer.current_epoch}' if padding: line = f"{line:>{len('Epoch xxxx')}}" metrics_str = [] losses_dict = trainer.callback_metrics for (metric_name, dico_name) in self.metric_monitor.items(): if (dico_name in losses_dict): metric = losses_dict[dico_name].item() metric = metric_format.format(metric) metric = f'{metric_name} {metric}' metrics_str.append(metric) if (len(metrics_str) == 0): return memory = f'Memory {psutil.virtual_memory().percent}%' line = ((((line + ': ') + ' '.join(metrics_str)) + ' ') + memory) logger.info(line)
def get_module_config(cfg_model, path='modules'): files = os.listdir(f'./configs/{path}/') for file in files: if file.endswith('.yaml'): with open((f'./configs/{path}/' + file), 'r') as f: cfg_model.merge_with(OmegaConf.load(f)) return cfg_model
def get_obj_from_str(string, reload=False): (module, cls) = string.rsplit('.', 1) if reload: module_imp = importlib.import_module(module) importlib.reload(module_imp) return getattr(importlib.import_module(module, package=None), cls)
def instantiate_from_config(config): if (not ('target' in config)): if (config == '__is_first_stage__'): return None elif (config == '__is_unconditional__'): return None raise KeyError('Expected key `target` to instantiate.') return get_obj_from_str(config['target'])(**config.get('params', dict()))
def parse_args(phase='train'): parser = ArgumentParser() group = parser.add_argument_group('Training options') if (phase in ['train', 'test', 'demo']): group.add_argument('--cfg', type=str, required=False, default='./configs/config.yaml', help='config file') group.add_argument('--cfg_assets', type=str, required=False, default='./configs/assets.yaml', help='config file for asset paths') group.add_argument('--batch_size', type=int, required=False, help='training batch size') group.add_argument('--device', type=int, nargs='+', required=False, help='training device') group.add_argument('--nodebug', action='store_true', required=False, help='debug or not') group.add_argument('--dir', type=str, required=False, help='evaluate existing npys') if (phase == 'demo'): group.add_argument('--render', action='store_true', help='Render visulizaed figures') group.add_argument('--render_mode', type=str, help='video or sequence') group.add_argument('--frame_rate', type=float, default=12.5, help='the frame rate for the input/output motion') group.add_argument('--replication', type=int, default=1, help='the frame rate for the input/output motion') group.add_argument('--example', type=str, required=False, help='input text and lengths with txt format') group.add_argument('--task', type=str, required=False, help='random_sampling, reconstrucion or text_motion') group.add_argument('--out_dir', type=str, required=False, help='output dir') group.add_argument('--allinone', action='store_true', required=False, help='output seperate or combined npy file') if (phase == 'render'): group.add_argument('--cfg', type=str, required=False, default='./configs/render.yaml', help='config file') group.add_argument('--cfg_assets', type=str, required=False, default='./configs/assets.yaml', help='config file for asset paths') group.add_argument('--npy', type=str, required=False, default=None, help='npy motion files') group.add_argument('--dir', type=str, required=False, default=None, help='npy motion folder') group.add_argument('--mode', type=str, required=False, default='sequence', help='render target: video, sequence, frame') group.add_argument('--joint_type', type=str, required=False, default=None, help='mmm or vertices for skeleton') params = parser.parse_args() cfg_base = OmegaConf.load('./configs/base.yaml') cfg_exp = OmegaConf.merge(cfg_base, OmegaConf.load(params.cfg)) cfg_model = get_module_config(cfg_exp.model, cfg_exp.model.target) cfg_assets = OmegaConf.load(params.cfg_assets) cfg = OmegaConf.merge(cfg_exp, cfg_model, cfg_assets) if (phase in ['train', 'test']): cfg.TRAIN.BATCH_SIZE = (params.batch_size if params.batch_size else cfg.TRAIN.BATCH_SIZE) cfg.DEVICE = (params.device if params.device else cfg.DEVICE) cfg.DEBUG = ((not params.nodebug) if (params.nodebug is not None) else cfg.DEBUG) cfg.DEBUG = (False if (phase == 'test') else cfg.DEBUG) if (phase == 'test'): cfg.DEBUG = False cfg.DEVICE = [0] print('Force no debugging and one gpu when testing') cfg.TEST.TEST_DIR = (params.dir if params.dir else cfg.TEST.TEST_DIR) if (phase == 'demo'): cfg.DEMO.RENDER = params.render cfg.DEMO.FRAME_RATE = params.frame_rate cfg.DEMO.EXAMPLE = params.example cfg.DEMO.TASK = params.task cfg.TEST.FOLDER = (params.out_dir if params.dir else cfg.TEST.FOLDER) cfg.DEMO.REPLICATION = params.replication cfg.DEMO.OUTALL = params.allinone if (phase == 'render'): if params.npy: cfg.RENDER.NPY = params.npy cfg.RENDER.INPUT_MODE = 'npy' if params.dir: cfg.RENDER.DIR = params.dir cfg.RENDER.INPUT_MODE = 'dir' cfg.RENDER.JOINT_TYPE = params.joint_type cfg.RENDER.MODE = params.mode if cfg.DEBUG: cfg.NAME = ('debug--' + cfg.NAME) cfg.LOGGER.WANDB.OFFLINE = True cfg.LOGGER.VAL_EVERY_STEPS = 1 return cfg
class HumanML3DDataModule(BASEDataModule): def __init__(self, cfg, batch_size, num_workers, collate_fn=None, phase='train', **kwargs): super().__init__(batch_size=batch_size, num_workers=num_workers, collate_fn=collate_fn) self.save_hyperparameters(logger=False) self.name = 'humanml3d' self.njoints = 22 if (phase == 'text_only'): self.Dataset = TextOnlyDataset else: self.Dataset = Text2MotionDatasetV2 self.cfg = cfg sample_overrides = {'split': 'val', 'tiny': True, 'progress_bar': False} self._sample_set = self.get_sample_set(overrides=sample_overrides) self.nfeats = self._sample_set.nfeats def feats2joints(self, features): mean = torch.tensor(self.hparams.mean).to(features) std = torch.tensor(self.hparams.std).to(features) features = ((features * std) + mean) return recover_from_ric(features, self.njoints) def joints2feats(self, features): features = process_file(features, self.njoints)[0] return features def renorm4t2m(self, features): ori_mean = torch.tensor(self.hparams.mean).to(features) ori_std = torch.tensor(self.hparams.std).to(features) eval_mean = torch.tensor(self.hparams.mean_eval).to(features) eval_std = torch.tensor(self.hparams.std_eval).to(features) features = ((features * ori_std) + ori_mean) features = ((features - eval_mean) / eval_std) return features def mm_mode(self, mm_on=True): if mm_on: self.is_mm = True self.name_list = self.test_dataset.name_list self.mm_list = np.random.choice(self.name_list, self.cfg.TEST.MM_NUM_SAMPLES, replace=False) self.test_dataset.name_list = self.mm_list else: self.is_mm = False self.test_dataset.name_list = self.name_list
class Humanact12DataModule(BASEDataModule): def __init__(self, cfg, batch_size, num_workers, collate_fn=None, phase='train', **kwargs): super().__init__(batch_size=batch_size, num_workers=num_workers, collate_fn=collate_fn) self.save_hyperparameters(logger=False) self.name = 'HumanAct12' self.Dataset = HumanAct12Poses self.cfg = cfg sample_overrides = {'num_seq_max': 2, 'split': 'test', 'tiny': True, 'progress_bar': False} self.nfeats = 150 self.njoints = 25 self.nclasses = 12
class KitDataModule(BASEDataModule): def __init__(self, cfg, phase='train', collate_fn=all_collate, batch_size: int=32, num_workers: int=16, **kwargs): super().__init__(batch_size=batch_size, num_workers=num_workers, collate_fn=collate_fn) self.save_hyperparameters(logger=False) self.name = 'kit' self.njoints = 21 if (phase == 'text_only'): self.Dataset = TextOnlyDataset else: self.Dataset = Text2MotionDatasetV2 self.cfg = cfg sample_overrides = {'split': 'val', 'tiny': True, 'progress_bar': False} self._sample_set = self.get_sample_set(overrides=sample_overrides) self.nfeats = self._sample_set.nfeats def feats2joints(self, features): mean = torch.tensor(self.hparams.mean).to(features) std = torch.tensor(self.hparams.std).to(features) features = ((features * std) + mean) return recover_from_ric(features, self.njoints) def renorm4t2m(self, features): ori_mean = torch.tensor(self.hparams.mean).to(features) ori_std = torch.tensor(self.hparams.std).to(features) eval_mean = torch.tensor(self.hparams.mean_eval).to(features) eval_std = torch.tensor(self.hparams.std_eval).to(features) features = ((features * ori_std) + ori_mean) features = ((features - eval_mean) / eval_std) return features def mm_mode(self, mm_on=True): if mm_on: self.is_mm = True self.name_list = self.test_dataset.name_list self.mm_list = np.random.choice(self.name_list, self.cfg.TEST.MM_NUM_SAMPLES, replace=False) self.test_dataset.name_list = self.mm_list else: self.is_mm = False self.test_dataset.name_list = self.name_list
class UestcDataModule(BASEDataModule): def __init__(self, cfg, batch_size, num_workers, collate_fn=None, method_name='vibe', phase='train', **kwargs): super().__init__(batch_size=batch_size, num_workers=num_workers, collate_fn=collate_fn) self.save_hyperparameters(logger=False) self.name = 'Uestc' self.Dataset = UESTC self.cfg = cfg self.nfeats = 150 self.njoints = 25 self.nclasses = 40
class HumanAct12Poses(Dataset): dataname = 'humanact12' def __init__(self, datapath='data/HumanAct12Poses', **kargs): self.datapath = datapath super().__init__(**kargs) pkldatafilepath = os.path.join(datapath, 'humanact12poses.pkl') with rich.progress.open(pkldatafilepath, 'rb', description='loading humanact12 pkl') as f: data = pkl.load(f) self._pose = [x for x in data['poses']] self._num_frames_in_video = [p.shape[0] for p in self._pose] self._joints = [x for x in data['joints3D']] self._actions = [x for x in data['y']] total_num_actions = 12 self.num_classes = total_num_actions self._train = list(range(len(self._pose))) keep_actions = np.arange(0, total_num_actions) self._action_to_label = {x: i for (i, x) in enumerate(keep_actions)} self._label_to_action = {i: x for (i, x) in enumerate(keep_actions)} self._action_classes = humanact12_coarse_action_enumerator def _load_joints3D(self, ind, frame_ix): return self._joints[ind][frame_ix] def _load_rotvec(self, ind, frame_ix): pose = self._pose[ind][frame_ix].reshape((- 1), 24, 3) return pose
def parse_info_name(path): name = os.path.splitext(os.path.split(path)[(- 1)])[0] info = {} current_letter = None for letter in name: if (letter in string.ascii_letters): info[letter] = [] current_letter = letter else: info[current_letter].append(letter) for key in info.keys(): info[key] = ''.join(info[key]) return info
def to_numpy(tensor): if torch.is_tensor(tensor): return tensor.cpu().numpy() elif (type(tensor).__module__ != 'numpy'): raise ValueError('Cannot convert {} to numpy array'.format(type(tensor))) return tensor
def to_torch(ndarray): if (type(ndarray).__module__ == 'numpy'): return torch.from_numpy(ndarray) elif (not torch.is_tensor(ndarray)): raise ValueError('Cannot convert {} to torch tensor'.format(type(ndarray))) return ndarray
def cleanexit(): import sys import os try: sys.exit(0) except SystemExit: os._exit(0)
def lengths_to_mask(lengths): max_len = max(lengths) mask = (torch.arange(max_len, device=lengths.device).expand(len(lengths), max_len) < lengths.unsqueeze(1)) return mask
def collate_tensors(batch): dims = batch[0].dim() max_size = [max([b.size(i) for b in batch]) for i in range(dims)] size = ((len(batch),) + tuple(max_size)) canvas = batch[0].new_zeros(size=size) for (i, b) in enumerate(batch): sub_tensor = canvas[i] for d in range(dims): sub_tensor = sub_tensor.narrow(d, 0, b.size(d)) sub_tensor.add_(b) return canvas
def collate(batch): databatch = [b[0] for b in batch] labelbatch = [b[1] for b in batch] lenbatch = [len(b[0][0][0]) for b in batch] databatchTensor = collate_tensors(databatch) labelbatchTensor = torch.as_tensor(labelbatch) lenbatchTensor = torch.as_tensor(lenbatch) maskbatchTensor = lengths_to_mask(lenbatchTensor) batch = {'x': databatchTensor, 'y': labelbatchTensor, 'mask': maskbatchTensor, 'lengths': lenbatchTensor} return batch
class BASEDataModule(pl.LightningDataModule): def __init__(self, collate_fn, batch_size: int, num_workers: int): super().__init__() self.dataloader_options = {'batch_size': batch_size, 'num_workers': num_workers, 'collate_fn': collate_fn} self.persistent_workers = True self.is_mm = False def get_sample_set(self, overrides={}): sample_params = self.hparams.copy() sample_params.update(overrides) split_file = pjoin(eval(f'self.cfg.DATASET.{self.name.upper()}.SPLIT_ROOT'), (self.cfg.EVAL.SPLIT + '.txt')) return self.Dataset(split_file=split_file, **sample_params) def __getattr__(self, item): if (item.endswith('_dataset') and (not item.startswith('_'))): subset = item[:(- len('_dataset'))] item_c = ('_' + item) if (item_c not in self.__dict__): subset = (subset.upper() if (subset != 'val') else 'EVAL') split = eval(f'self.cfg.{subset}.SPLIT') split_file = pjoin(eval(f'self.cfg.DATASET.{self.name.upper()}.SPLIT_ROOT'), (eval(f'self.cfg.{subset}.SPLIT') + '.txt')) self.__dict__[item_c] = self.Dataset(split_file=split_file, split=split, **self.hparams) return getattr(self, item_c) classname = self.__class__.__name__ raise AttributeError(f"'{classname}' object has no attribute '{item}'") def setup(self, stage=None): self.stage = stage if (stage in (None, 'fit')): _ = self.train_dataset _ = self.val_dataset if (stage in (None, 'test')): _ = self.test_dataset def train_dataloader(self): return DataLoader(self.train_dataset, shuffle=True, persistent_workers=True, **self.dataloader_options) def predict_dataloader(self): dataloader_options = self.dataloader_options.copy() dataloader_options['batch_size'] = (1 if self.is_mm else self.cfg.TEST.BATCH_SIZE) dataloader_options['num_workers'] = self.cfg.TEST.NUM_WORKERS dataloader_options['shuffle'] = False return DataLoader(self.test_dataset, persistent_workers=True, **dataloader_options) def val_dataloader(self): dataloader_options = self.dataloader_options.copy() dataloader_options['batch_size'] = self.cfg.EVAL.BATCH_SIZE dataloader_options['num_workers'] = self.cfg.EVAL.NUM_WORKERS dataloader_options['shuffle'] = False return DataLoader(self.val_dataset, persistent_workers=True, **dataloader_options) def test_dataloader(self): dataloader_options = self.dataloader_options.copy() dataloader_options['batch_size'] = (1 if self.is_mm else self.cfg.TEST.BATCH_SIZE) dataloader_options['num_workers'] = self.cfg.TEST.NUM_WORKERS dataloader_options['shuffle'] = False return DataLoader(self.test_dataset, persistent_workers=True, **dataloader_options)
def get_mean_std(phase, cfg, dataset_name): name = ('t2m' if (dataset_name == 'humanml3d') else dataset_name) assert (name in ['t2m', 'kit']) if (phase in ['val']): if (name == 't2m'): data_root = pjoin(cfg.model.t2m_path, name, 'Comp_v6_KLD01', 'meta') elif (name == 'kit'): data_root = pjoin(cfg.model.t2m_path, name, 'Comp_v6_KLD005', 'meta') else: raise ValueError('Only support t2m and kit') mean = np.load(pjoin(data_root, 'mean.npy')) std = np.load(pjoin(data_root, 'std.npy')) else: data_root = eval(f'cfg.DATASET.{dataset_name.upper()}.ROOT') mean = np.load(pjoin(data_root, 'Mean.npy')) std = np.load(pjoin(data_root, 'Std.npy')) return (mean, std)
def get_WordVectorizer(cfg, phase, dataset_name): if (phase not in ['text_only']): if (dataset_name.lower() in ['humanml3d', 'kit']): return WordVectorizer(cfg.DATASET.WORD_VERTILIZER_PATH, 'our_vab') else: raise ValueError('Only support WordVectorizer for HumanML3D') else: return None
def get_collate_fn(name, phase='train'): if (name.lower() in ['humanml3d', 'kit']): return mld_collate elif (name.lower() in ['humanact12', 'uestc']): return a2m_collate
def get_datasets(cfg, logger=None, phase='train'): dataset_names = eval(f'cfg.{phase.upper()}.DATASETS') datasets = [] for dataset_name in dataset_names: if (dataset_name.lower() in ['humanml3d', 'kit']): data_root = eval(f'cfg.DATASET.{dataset_name.upper()}.ROOT') (mean, std) = get_mean_std(phase, cfg, dataset_name) (mean_eval, std_eval) = get_mean_std('val', cfg, dataset_name) wordVectorizer = get_WordVectorizer(cfg, phase, dataset_name) collate_fn = get_collate_fn(dataset_name, phase) if (dataset_name.lower() in ['kit']): dataset = dataset_module_map[dataset_name.lower()](cfg=cfg, batch_size=cfg.TRAIN.BATCH_SIZE, num_workers=cfg.TRAIN.NUM_WORKERS, debug=cfg.DEBUG, collate_fn=collate_fn, mean=mean, std=std, mean_eval=mean_eval, std_eval=std_eval, w_vectorizer=wordVectorizer, text_dir=pjoin(data_root, 'texts'), motion_dir=pjoin(data_root, motion_subdir[dataset_name]), max_motion_length=cfg.DATASET.SAMPLER.MAX_LEN, min_motion_length=24, max_text_len=cfg.DATASET.SAMPLER.MAX_TEXT_LEN, unit_length=eval(f'cfg.DATASET.{dataset_name.upper()}.UNIT_LEN')) datasets.append(dataset) else: dataset = dataset_module_map[dataset_name.lower()](cfg=cfg, batch_size=cfg.TRAIN.BATCH_SIZE, num_workers=cfg.TRAIN.NUM_WORKERS, debug=cfg.DEBUG, collate_fn=collate_fn, mean=mean, std=std, mean_eval=mean_eval, std_eval=std_eval, w_vectorizer=wordVectorizer, text_dir=pjoin(data_root, 'texts'), motion_dir=pjoin(data_root, motion_subdir[dataset_name]), max_motion_length=cfg.DATASET.SAMPLER.MAX_LEN, min_motion_length=cfg.DATASET.SAMPLER.MIN_LEN, max_text_len=cfg.DATASET.SAMPLER.MAX_TEXT_LEN, unit_length=eval(f'cfg.DATASET.{dataset_name.upper()}.UNIT_LEN')) datasets.append(dataset) elif (dataset_name.lower() in ['humanact12', 'uestc']): collate_fn = get_collate_fn(dataset_name, phase) dataset = dataset_module_map[dataset_name.lower()](datapath=eval(f'cfg.DATASET.{dataset_name.upper()}.ROOT'), cfg=cfg, batch_size=cfg.TRAIN.BATCH_SIZE, num_workers=cfg.TRAIN.NUM_WORKERS, debug=cfg.DEBUG, collate_fn=collate_fn, num_frames=cfg.DATASET.HUMANACT12.NUM_FRAMES, sampling=cfg.DATASET.SAMPLER.SAMPLING, sampling_step=cfg.DATASET.SAMPLER.SAMPLING_STEP, pose_rep=cfg.DATASET.HUMANACT12.POSE_REP, max_len=cfg.DATASET.SAMPLER.MAX_LEN, min_len=cfg.DATASET.SAMPLER.MIN_LEN, num_seq_max=(cfg.DATASET.SAMPLER.MAX_SQE if (not cfg.DEBUG) else 100), glob=cfg.DATASET.HUMANACT12.GLOB, translation=cfg.DATASET.HUMANACT12.TRANSLATION) cfg.DATASET.NCLASSES = dataset.nclasses datasets.append(dataset) elif (dataset_name.lower() in ['amass']): raise NotImplementedError else: raise NotImplementedError cfg.DATASET.NFEATS = datasets[0].nfeats cfg.DATASET.NJOINTS = datasets[0].njoints return datasets
def is_float(numStr): flag = False numStr = str(numStr).strip().lstrip('-').lstrip('+') try: reg = re.compile('^[-+]?[0-9]+\\.[0-9]+$') res = reg.match(str(numStr)) if res: flag = True except Exception as ex: print(('is_float() - error: ' + str(ex))) return flag
def is_number(numStr): flag = False numStr = str(numStr).strip().lstrip('-').lstrip('+') if str(numStr).isdigit(): flag = True return flag
def get_opt(opt_path, device): opt = Namespace() opt_dict = vars(opt) skip = ('-------------- End ----------------', '------------ Options -------------', '\n') print('Reading', opt_path) with open(opt_path) as f: for line in f: if (line.strip() not in skip): (key, value) = line.strip().split(': ') if (value in ('True', 'False')): opt_dict[key] = bool(value) elif is_float(value): opt_dict[key] = float(value) elif is_number(value): opt_dict[key] = int(value) else: opt_dict[key] = str(value) opt_dict['which_epoch'] = 'latest' opt.save_root = pjoin(opt.checkpoints_dir, opt.dataset_name, opt.name) opt.model_dir = pjoin(opt.save_root, 'model') opt.meta_dir = pjoin(opt.save_root, 'meta') if (opt.dataset_name == 't2m'): opt.data_root = './dataset/HumanML3D' opt.motion_dir = pjoin(opt.data_root, 'new_joint_vecs') opt.text_dir = pjoin(opt.data_root, 'texts') opt.joints_num = 22 opt.dim_pose = 263 opt.max_motion_length = 196 elif (opt.dataset_name == 'kit'): opt.data_root = './dataset/KIT-ML' opt.motion_dir = pjoin(opt.data_root, 'new_joint_vecs') opt.text_dir = pjoin(opt.data_root, 'texts') opt.joints_num = 21 opt.dim_pose = 251 opt.max_motion_length = 196 else: raise KeyError('Dataset not recognized') opt.dim_word = 300 opt.num_classes = (200 // opt.unit_length) opt.dim_pos_ohot = len(POS_enumerator) opt.is_train = False opt.is_continue = False opt.device = device return opt
def save_json(save_path, data): with open(save_path, 'w') as file: json.dump(data, file)
def load_json(file_path): with open(file_path, 'r') as file: data = json.load(file) return data
def process(graph): (entities, relations) = ({}, []) for i in graph['verbs']: description = i['description'] pos = 0 flag = 0 (_words, _spans) = ([], []) for i in description.split(): (tags, verb) = ({}, 0) if ('[' in i): _role = i[1:(- 1)] flag = 1 _spans = [pos] _words = [] elif (']' in i): _words.append(i[:(- 1)]) entities[len(entities)] = {'role': _role, 'spans': _spans, 'words': _words} pos += 1 flag = 0 if (_role != 'V'): tags[len(entities)] = _role else: verb = len(entities) else: pos += 1 if flag: _words.append(i) _spans.append(pos) for i in tags: relations.append((verb, i, tags[i])) output = {'entities': entities, 'relations': relations} return output
class WordVectorizer(object): def __init__(self, meta_root, prefix): vectors = np.load(pjoin(meta_root, ('%s_data.npy' % prefix))) words = pickle.load(open(pjoin(meta_root, ('%s_words.pkl' % prefix)), 'rb')) word2idx = pickle.load(open(pjoin(meta_root, ('%s_idx.pkl' % prefix)), 'rb')) self.word2vec = {w: vectors[word2idx[w]] for w in words} def _get_pos_ohot(self, pos): pos_vec = np.zeros(len(POS_enumerator)) if (pos in POS_enumerator): pos_vec[POS_enumerator[pos]] = 1 else: pos_vec[POS_enumerator['OTHER']] = 1 return pos_vec def __len__(self): return len(self.word2vec) def __getitem__(self, item): (word, pos) = item.split('/') if (word in self.word2vec): word_vec = self.word2vec[word] vip_pos = None for (key, values) in VIP_dict.items(): if (word in values): vip_pos = key break if (vip_pos is not None): pos_vec = self._get_pos_ohot(vip_pos) else: pos_vec = self._get_pos_ohot(pos) else: word_vec = self.word2vec['unk'] pos_vec = self._get_pos_ohot('OTHER') return (word_vec, pos_vec)
class FrameSampler(): def __init__(self, sampling='conseq', sampling_step=1, request_frames=None, threshold_reject=0.75, max_len=1000, min_len=10): self.sampling = sampling self.sampling_step = sampling_step self.request_frames = request_frames self.threshold_reject = threshold_reject self.max_len = max_len self.min_len = min_len def __call__(self, num_frames): return get_frameix_from_data_index(num_frames, self.request_frames, self.sampling, self.sampling_step) def accept(self, duration): if (self.request_frames is None): if (duration > self.max_len): return False elif (duration < self.min_len): return False else: min_number = (self.threshold_reject * self.request_frames) if (duration < min_number): return False return True def get(self, key, default=None): return getattr(self, key, default) def __getitem__(self, key): return getattr(self, key)
def subsample(num_frames, last_framerate, new_framerate): step = int((last_framerate / new_framerate)) assert (step >= 1) frames = np.arange(0, num_frames, step) return frames
def upsample(motion, last_framerate, new_framerate): step = int((new_framerate / last_framerate)) assert (step >= 1) alpha = np.linspace(0, 1, (step + 1)) last = np.einsum('l,...->l...', (1 - alpha), motion[:(- 1)]) new = np.einsum('l,...->l...', alpha, motion[1:]) chuncks = (last + new)[:(- 1)] output = np.concatenate(chuncks.swapaxes(1, 0)) output = np.concatenate((output, motion[[(- 1)]])) return output
def get_frameix_from_data_index(num_frames: int, request_frames: Optional[int], sampling: str='conseq', sampling_step: int=1) -> Array: nframes = num_frames if (request_frames is None): frame_ix = np.arange(nframes) elif (request_frames > nframes): fair = False if fair: choices = np.random.choice(range(nframes), request_frames, replace=True) frame_ix = sorted(choices) else: ntoadd = max(0, (request_frames - nframes)) lastframe = (nframes - 1) padding = (lastframe * np.ones(ntoadd, dtype=int)) frame_ix = np.concatenate((np.arange(0, nframes), padding)) elif (sampling in ['conseq', 'random_conseq']): step_max = ((nframes - 1) // (request_frames - 1)) if (sampling == 'conseq'): if ((sampling_step == (- 1)) or ((sampling_step * (request_frames - 1)) >= nframes)): step = step_max else: step = sampling_step elif (sampling == 'random_conseq'): step = random.randint(1, step_max) lastone = (step * (request_frames - 1)) shift_max = ((nframes - lastone) - 1) shift = random.randint(0, max(0, (shift_max - 1))) frame_ix = (shift + np.arange(0, (lastone + 1), step)) elif (sampling == 'random'): choices = np.random.choice(range(nframes), request_frames, replace=False) frame_ix = sorted(choices) else: raise ValueError('Sampling not recognized.') return frame_ix
def lengths_to_mask(lengths): max_len = max(lengths) mask = (torch.arange(max_len, device=lengths.device).expand(len(lengths), max_len) < lengths.unsqueeze(1)) return mask
def collate_tensors(batch): dims = batch[0].dim() max_size = [max([b.size(i) for b in batch]) for i in range(dims)] size = ((len(batch),) + tuple(max_size)) canvas = batch[0].new_zeros(size=size) for (i, b) in enumerate(batch): sub_tensor = canvas[i] for d in range(dims): sub_tensor = sub_tensor.narrow(d, 0, b.size(d)) sub_tensor.add_(b) return canvas
def all_collate(batch): notnone_batches = [b for b in batch if (b is not None)] databatch = [b['motion'] for b in notnone_batches] if ('lengths' in notnone_batches[0]): lenbatch = [b['lengths'] for b in notnone_batches] else: lenbatch = [len(b['inp'][0][0]) for b in notnone_batches] databatchTensor = collate_tensors(databatch) lenbatchTensor = torch.as_tensor(lenbatch) maskbatchTensor = lengths_to_mask(lenbatchTensor, databatchTensor.shape[(- 1)]).unsqueeze(1).unsqueeze(1) motion = databatchTensor cond = {'y': {'mask': maskbatchTensor, 'lengths': lenbatchTensor}} if ('text' in notnone_batches[0]): textbatch = [b['text'] for b in notnone_batches] cond['y'].update({'text': textbatch}) if ('action_text' in notnone_batches[0]): action_text = [b['action_text'] for b in notnone_batches] cond['y'].update({'action_text': action_text}) return (motion, cond)
def mld_collate(batch): notnone_batches = [b for b in batch if (b is not None)] notnone_batches.sort(key=(lambda x: x[3]), reverse=True) adapted_batch = {'motion': collate_tensors([torch.tensor(b[4]).float() for b in notnone_batches]), 'text': [b[2] for b in notnone_batches], 'length': [b[5] for b in notnone_batches], 'word_embs': collate_tensors([torch.tensor(b[0]).float() for b in notnone_batches]), 'pos_ohot': collate_tensors([torch.tensor(b[1]).float() for b in notnone_batches]), 'text_len': collate_tensors([torch.tensor(b[3]) for b in notnone_batches]), 'tokens': [b[6] for b in notnone_batches], 'V': [b[7] for b in notnone_batches], 'entities': [b[8] for b in notnone_batches], 'relations': [b[9] for b in notnone_batches]} return adapted_batch
def a2m_collate(batch): databatch = [b[0] for b in batch] labelbatch = [b[1] for b in batch] lenbatch = [len(b[0][0][0]) for b in batch] labeltextbatch = [b[3] for b in batch] databatchTensor = collate_tensors(databatch) labelbatchTensor = torch.as_tensor(labelbatch).unsqueeze(1) lenbatchTensor = torch.as_tensor(lenbatch) maskbatchTensor = lengths_to_mask(lenbatchTensor) adapted_batch = {'motion': databatchTensor.permute(0, 3, 2, 1).flatten(start_dim=2), 'action': labelbatchTensor, 'action_text': labeltextbatch, 'mask': maskbatchTensor, 'length': lenbatchTensor} return adapted_batch
def parse_args(self, args=None, namespace=None): if (args is not None): return self.parse_args_bak(args=args, namespace=namespace) try: idx = sys.argv.index('--') args = sys.argv[(idx + 1):] except ValueError as e: args = [] return self.parse_args_bak(args=args, namespace=namespace)
def code_path(path=''): code_dir = hydra.utils.get_original_cwd() code_dir = Path(code_dir) return str((code_dir / path))
def working_path(path): return str((Path(os.getcwd()) / path))
def generate_id(): return ID
def get_last_checkpoint(path, ckpt_name='last.ckpt'): output_dir = Path(hydra.utils.to_absolute_path(path)) last_ckpt_path = ((output_dir / 'checkpoints') / ckpt_name) return str(last_ckpt_path)
def get_kitname(load_amass_data: bool, load_with_rot: bool): if (not load_amass_data): return 'kit-mmm-xyz' if (load_amass_data and (not load_with_rot)): return 'kit-amass-xyz' if (load_amass_data and load_with_rot): return 'kit-amass-rot'
def resolve_cfg_path(cfg: DictConfig): working_dir = os.getcwd() cfg.working_dir = working_dir
class ActorVae(nn.Module): def __init__(self, ablation, nfeats: int, latent_dim: list=[1, 256], ff_size: int=1024, num_layers: int=9, num_heads: int=4, dropout: float=0.1, is_vae: bool=True, activation: str='gelu', position_embedding: str='learned', **kwargs) -> None: super().__init__() self.latent_size = latent_dim[0] self.latent_dim = latent_dim[(- 1)] self.is_vae = is_vae input_feats = nfeats output_feats = nfeats self.encoder = ActorAgnosticEncoder(nfeats=input_feats, vae=True, latent_dim=self.latent_dim, ff_size=ff_size, num_layers=num_layers, num_heads=num_heads, dropout=dropout, activation=activation, **kwargs) self.decoder = ActorAgnosticDecoder(nfeats=output_feats, vae=True, latent_dim=self.latent_dim, ff_size=ff_size, num_layers=num_layers, num_heads=num_heads, dropout=dropout, activation=activation, **kwargs) def forward(self, features: Tensor, lengths: Optional[List[int]]=None): print('Should Not enter here') (z, dist) = self.encode(features, lengths) feats_rst = self.decode(z, lengths) return (feats_rst, z, dist) def encode(self, features: Tensor, lengths: Optional[List[int]]=None) -> Union[(Tensor, Distribution)]: dist = self.encoder(features, lengths) if self.is_vae: latent = sample_from_distribution(dist) else: latent = dist.unsqueeze(0) return (latent, dist) def decode(self, z: Tensor, lengths: List[int]): feats = self.decoder(z, lengths) return feats
class ActorAgnosticEncoder(nn.Module): def __init__(self, nfeats: int, vae: bool, latent_dim: int=256, ff_size: int=1024, num_layers: int=4, num_heads: int=4, dropout: float=0.1, activation: str='gelu', **kwargs) -> None: super().__init__() input_feats = nfeats self.vae = vae self.skel_embedding = nn.Linear(input_feats, latent_dim) if vae: self.mu_token = nn.Parameter(torch.randn(latent_dim)) self.logvar_token = nn.Parameter(torch.randn(latent_dim)) else: self.emb_token = nn.Parameter(torch.randn(latent_dim)) self.sequence_pos_encoding = PositionalEncoding(latent_dim, dropout) seq_trans_encoder_layer = nn.TransformerEncoderLayer(d_model=latent_dim, nhead=num_heads, dim_feedforward=ff_size, dropout=dropout, activation=activation) self.seqTransEncoder = nn.TransformerEncoder(seq_trans_encoder_layer, num_layers=num_layers) def forward(self, features: Tensor, lengths: Optional[List[int]]=None) -> Union[(Tensor, Distribution)]: if (lengths is None): lengths = [len(feature) for feature in features] device = features.device (bs, nframes, nfeats) = features.shape mask = lengths_to_mask(lengths, device) x = features x = self.skel_embedding(x) x = x.permute(1, 0, 2) if self.vae: mu_token = torch.tile(self.mu_token, (bs,)).reshape(bs, (- 1)) logvar_token = torch.tile(self.logvar_token, (bs,)).reshape(bs, (- 1)) xseq = torch.cat((mu_token[None], logvar_token[None], x), 0) token_mask = torch.ones((bs, 2), dtype=bool, device=x.device) aug_mask = torch.cat((token_mask, mask), 1) else: emb_token = torch.tile(self.emb_token, (bs,)).reshape(bs, (- 1)) xseq = torch.cat((emb_token[None], x), 0) token_mask = torch.ones((bs, 1), dtype=bool, device=x.device) aug_mask = torch.cat((token_mask, mask), 1) xseq = self.sequence_pos_encoding(xseq) final = self.seqTransEncoder(xseq, src_key_padding_mask=(~ aug_mask)) if self.vae: (mu, logvar) = (final[0], final[1]) std = logvar.exp().pow(0.5) dist = torch.distributions.Normal(mu, std) return dist else: return final[0]
class ActorAgnosticDecoder(nn.Module): def __init__(self, nfeats: int, latent_dim: int=256, ff_size: int=1024, num_layers: int=4, num_heads: int=4, dropout: float=0.1, activation: str='gelu', **kwargs) -> None: super().__init__() output_feats = nfeats self.latent_dim = latent_dim self.nfeats = nfeats self.sequence_pos_encoding = PositionalEncoding(latent_dim, dropout) seq_trans_decoder_layer = nn.TransformerDecoderLayer(d_model=latent_dim, nhead=num_heads, dim_feedforward=ff_size, dropout=dropout, activation=activation) self.seqTransDecoder = nn.TransformerDecoder(seq_trans_decoder_layer, num_layers=num_layers) self.final_layer = nn.Linear(latent_dim, output_feats) def forward(self, z: Tensor, lengths: List[int]): mask = lengths_to_mask(lengths, z.device) (bs, nframes) = mask.shape nfeats = self.nfeats time_queries = torch.zeros(nframes, bs, self.latent_dim, device=z.device) time_queries = self.sequence_pos_encoding(time_queries) output = self.seqTransDecoder(tgt=time_queries, memory=z, tgt_key_padding_mask=(~ mask)) output = self.final_layer(output) output[(~ mask.T)] = 0 feats = output.permute(1, 0, 2) return feats
def sample_from_distribution(dist, *, fact=1.0, sample_mean=False) -> Tensor: if sample_mean: return dist.loc.unsqueeze(0) if (fact is None): return dist.rsample().unsqueeze(0) eps = (dist.rsample() - dist.loc) z = (dist.loc + (fact * eps)) z = z.unsqueeze(0) return z
class Encoder_FC(nn.Module): def __init__(self, modeltype, njoints, nfeats, num_frames, num_classes, translation, pose_rep, glob, glob_rot, latent_dim=256, **kargs): super().__init__() self.modeltype = modeltype self.njoints = njoints self.nfeats = nfeats self.num_frames = num_frames self.num_classes = num_classes self.translation = translation self.pose_rep = pose_rep self.glob = glob self.glob_rot = glob_rot self.latent_dim = latent_dim self.activation = nn.GELU() self.input_dim = (((self.njoints * self.nfeats) * self.num_frames) + self.num_classes) self.fully_connected = nn.Sequential(nn.Linear(self.input_dim, 512), nn.GELU(), nn.Linear(512, 256), nn.GELU()) if (self.modeltype == 'cvae'): self.mu = nn.Linear(256, self.latent_dim) self.var = nn.Linear(256, self.latent_dim) else: self.final = nn.Linear(256, self.latent_dim) def forward(self, batch): (x, y) = (batch['x'], batch['y']) (bs, njoints, feats, nframes) = x.size() if (((njoints * feats) * nframes) != ((self.njoints * self.nfeats) * self.num_frames)): raise ValueError('This model is not adapted with this input') if (len(y.shape) == 1): y = F.one_hot(y, self.num_classes) y = y.to(dtype=x.dtype) x = x.reshape(bs, ((njoints * feats) * nframes)) x = torch.cat((x, y), 1) x = self.fully_connected(x) if (self.modeltype == 'cvae'): return {'mu': self.mu(x), 'logvar': self.var(x)} else: return {'z': self.final(x)}
class Decoder_FC(nn.Module): def __init__(self, modeltype, njoints, nfeats, num_frames, num_classes, translation, pose_rep, glob, glob_rot, latent_dim=256, **kargs): super().__init__() self.modeltype = modeltype self.njoints = njoints self.nfeats = nfeats self.num_frames = num_frames self.num_classes = num_classes self.translation = translation self.pose_rep = pose_rep self.glob = glob self.glob_rot = glob_rot self.latent_dim = latent_dim self.input_dim = (self.latent_dim + self.num_classes) self.output_dim = ((self.njoints * self.nfeats) * self.num_frames) self.fully_connected = nn.Sequential(nn.Linear(self.input_dim, 256), nn.GELU(), nn.Linear(256, 512), nn.GELU(), nn.Linear(512, self.output_dim), nn.GELU()) def forward(self, batch): (z, y) = (batch['z'], batch['y']) if (len(y.shape) == 1): y = F.one_hot(y, self.num_classes) y = y.to(dtype=z.dtype) z = torch.cat((z, y), dim=1) z = self.fully_connected(z) (bs, _) = z.size() z = z.reshape(bs, self.njoints, self.nfeats, self.num_frames) batch['output'] = z return batch
class GATLayer(nn.Module): def __init__(self, in_features=768, out_features=768, dropout=0.1, alpha=0.2, concat=True): super(GATLayer, self).__init__() self.dropout = dropout self.in_features = in_features self.out_features = out_features self.alpha = alpha self.concat = concat self.W = nn.Parameter(torch.empty(size=(in_features, out_features))) self.leakyrelu = nn.LeakyReLU(self.alpha) self.a = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARG0 = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARG1 = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARG2 = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARG3 = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARG4 = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARGM_LOC = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARGM_MNR = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARGM_TMP = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARGM_DIR = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.ARGM_ADV = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.MA = nn.Parameter(torch.empty(size=((2 * out_features), 1))) self.OTHERS = nn.Parameter(torch.empty(size=((2 * out_features), 1))) nn.init.xavier_uniform_(self.W.data, gain=1.414) nn.init.xavier_uniform_(self.a, gain=1.414) nn.init.xavier_uniform_(self.ARG0.data, gain=1.414) nn.init.xavier_uniform_(self.ARG1.data, gain=1.414) nn.init.xavier_uniform_(self.ARG2.data, gain=1.414) nn.init.xavier_uniform_(self.ARG3.data, gain=1.414) nn.init.xavier_uniform_(self.ARG4.data, gain=1.414) nn.init.xavier_uniform_(self.ARGM_LOC.data, gain=1.414) nn.init.xavier_uniform_(self.ARGM_MNR.data, gain=1.414) nn.init.xavier_uniform_(self.ARGM_TMP.data, gain=1.414) nn.init.xavier_uniform_(self.ARGM_DIR.data, gain=1.414) nn.init.xavier_uniform_(self.ARGM_ADV.data, gain=1.414) nn.init.xavier_uniform_(self.MA.data, gain=1.414) nn.init.xavier_uniform_(self.OTHERS.data, gain=1.414) def forward(self, h0, h1, multi_adj, adj): Wh0 = torch.einsum('bnd,de->bne', [h0, self.W]) Wh1 = torch.einsum('bnd,de->bne', [h1, self.W]) a_input = self._prepare_attentional_mechanism_input(Wh0, Wh1) e = self.leakyrelu(torch.matmul(a_input, self.a).squeeze(3)) e_ARG0 = self.leakyrelu(torch.matmul(a_input, self.ARG0).squeeze(3)) e_ARG1 = self.leakyrelu(torch.matmul(a_input, self.ARG1).squeeze(3)) e_ARG2 = self.leakyrelu(torch.matmul(a_input, self.ARG2).squeeze(3)) e_ARG3 = self.leakyrelu(torch.matmul(a_input, self.ARG3).squeeze(3)) e_ARG4 = self.leakyrelu(torch.matmul(a_input, self.ARG4).squeeze(3)) e_ARGM_LOC = self.leakyrelu(torch.matmul(a_input, self.ARGM_LOC).squeeze(3)) e_ARGM_MNR = self.leakyrelu(torch.matmul(a_input, self.ARGM_MNR).squeeze(3)) e_ARGM_TMP = self.leakyrelu(torch.matmul(a_input, self.ARGM_TMP).squeeze(3)) e_ARGM_DIR = self.leakyrelu(torch.matmul(a_input, self.ARGM_DIR).squeeze(3)) e_ARGM_ADV = self.leakyrelu(torch.matmul(a_input, self.ARGM_ADV).squeeze(3)) e_MA = self.leakyrelu(torch.matmul(a_input, self.MA).squeeze(3)) e_OTHERS = self.leakyrelu(torch.matmul(a_input, self.OTHERS).squeeze(3)) zero_vec = ((- 9000000000000000.0) * torch.ones_like(e)) attention = torch.where((adj > 0), e, zero_vec) zero_vec = torch.zeros_like(e_ARG0) attention_ARG0 = torch.where((multi_adj['ARG0'] > 0), e_ARG0, zero_vec) attention_ARG1 = torch.where((multi_adj['ARG1'] > 0), e_ARG1, zero_vec) attention_ARG2 = torch.where((multi_adj['ARG2'] > 0), e_ARG2, zero_vec) attention_ARG3 = torch.where((multi_adj['ARG3'] > 0), e_ARG3, zero_vec) attention_ARG4 = torch.where((multi_adj['ARG4'] > 0), e_ARG4, zero_vec) attention_ARGM_LOC = torch.where((multi_adj['ARGM-LOC'] > 0), e_ARGM_LOC, zero_vec) attention_ARGM_MNR = torch.where((multi_adj['ARGM-MNR'] > 0), e_ARGM_MNR, zero_vec) attention_ARGM_TMP = torch.where((multi_adj['ARGM-TMP'] > 0), e_ARGM_TMP, zero_vec) attention_ARGM_DIR = torch.where((multi_adj['ARGM-DIR'] > 0), e_ARGM_DIR, zero_vec) attention_ARGM_ADV = torch.where((multi_adj['ARGM-ADV'] > 0), e_ARGM_ADV, zero_vec) attention_OTHERS = torch.where((multi_adj['OTHERS'] > 0), e_OTHERS, zero_vec) attention_MA = torch.where((multi_adj['MA'] > 0), e_MA, zero_vec) attention = F.softmax((attention + (0.01 * (((((((((((attention_ARG0 + attention_ARG1) + attention_ARG2) + attention_ARG3) + attention_ARG4) + attention_ARGM_LOC) + attention_ARGM_MNR) + attention_ARGM_TMP) + attention_ARGM_DIR) + attention_ARGM_ADV) + attention_OTHERS) + attention_MA))), dim=1) attention = F.dropout(attention, self.dropout, training=self.training) h_prime = torch.matmul(attention, Wh1) if self.concat: return F.elu(h_prime) else: return h_prime def _prepare_attentional_mechanism_input(self, Wh0, Wh1): (N0, N1) = (Wh0.size()[1], Wh1.size()[1]) Wh0_repeated_in_chunks = Wh0.repeat_interleave(N1, dim=1) Wh1_repeated_alternating = Wh1.repeat(1, N0, 1) all_combinations_matrix = torch.cat([Wh0_repeated_in_chunks, Wh1_repeated_alternating], dim=(- 1)) return all_combinations_matrix.view((- 1), N0, N1, (2 * self.out_features))
class MotionDiscriminator(nn.Module): def __init__(self, input_size, hidden_size, hidden_layer, output_size=12, use_noise=None): super().__init__() self.input_size = input_size self.hidden_size = hidden_size self.hidden_layer = hidden_layer self.use_noise = use_noise self.recurrent = nn.GRU(input_size, hidden_size, hidden_layer) self.linear1 = nn.Linear(hidden_size, 30) self.linear2 = nn.Linear(30, output_size) def forward(self, motion_sequence, lengths=None, hidden_unit=None): (bs, njoints, nfeats, num_frames) = motion_sequence.shape motion_sequence = motion_sequence.reshape(bs, (njoints * nfeats), num_frames) motion_sequence = motion_sequence.permute(2, 0, 1) if (hidden_unit is None): hidden_unit = self.initHidden(motion_sequence.size(1), self.hidden_layer).to(motion_sequence.device) (gru_o, _) = self.recurrent(motion_sequence.float(), hidden_unit) out = gru_o[tuple(torch.stack(((lengths - 1), torch.arange(bs, device=motion_sequence.device))))] lin1 = self.linear1(out) lin1 = torch.tanh(lin1) lin2 = self.linear2(lin1) return lin2 def initHidden(self, num_samples, layer): return torch.randn(layer, num_samples, self.hidden_size, requires_grad=False)
class MotionDiscriminatorForFID(MotionDiscriminator): def forward(self, motion_sequence, lengths=None, hidden_unit=None): (bs, njoints, nfeats, num_frames) = motion_sequence.shape motion_sequence = motion_sequence.reshape(bs, (njoints * nfeats), num_frames) motion_sequence = motion_sequence.permute(2, 0, 1) if (hidden_unit is None): hidden_unit = self.initHidden(motion_sequence.size(1), self.hidden_layer).to(motion_sequence.device) (gru_o, _) = self.recurrent(motion_sequence.float(), hidden_unit) out = gru_o[tuple(torch.stack(((lengths - 1), torch.arange(bs, device=motion_sequence.device))))] lin1 = self.linear1(out) lin1 = torch.tanh(lin1) return lin1
class MLDTextEncoder(nn.Module): def __init__(self, cfg, modelpath: str, finetune: bool=False, vae: bool=True, latent_dim: int=256, ff_size: int=1024, num_layers: int=6, num_heads: int=4, dropout: float=0.1, activation: str='gelu', **kwargs) -> None: super().__init__() from transformers import AutoTokenizer, AutoModel from transformers import logging logging.set_verbosity_error() os.environ['TOKENIZERS_PARALLELISM'] = 'false' self.tokenizer = AutoTokenizer.from_pretrained(modelpath) self.text_model = AutoModel.from_pretrained(modelpath) if (not finetune): self.text_model.training = False for p in self.text_model.parameters(): p.requires_grad = False self.text_encoded_dim = self.text_model.config.hidden_size self.text_encoded_dim = latent_dim encoded_dim = self.text_model.config.hidden_size self.projection = nn.Sequential(nn.ReLU(), nn.Linear(encoded_dim, latent_dim)) vae = False if vae: self.mu_token = nn.Parameter(torch.randn(latent_dim)) self.logvar_token = nn.Parameter(torch.randn(latent_dim)) else: self.global_text_token = nn.Parameter(torch.randn(latent_dim)) self.sequence_pos_encoding = PositionalEncoding(latent_dim, dropout) seq_trans_encoder_layer = nn.TransformerEncoderLayer(d_model=latent_dim, nhead=num_heads, dim_feedforward=ff_size, dropout=dropout, activation=activation) self.seqTransEncoder = nn.TransformerEncoder(seq_trans_encoder_layer, num_layers=num_layers) if self.is_action_branch: action_trans_encoder_layer = nn.TransformerEncoderLayer(d_model=latent_dim, nhead=num_heads, dim_feedforward=ff_size, dropout=dropout, activation=activation) self.actionTransEncoder = nn.TransformerEncoder(action_trans_encoder_layer, num_layers=num_layers) self.mean_token = nn.Parameter(torch.randn(latent_dim)) self.std_token = nn.Parameter(torch.randn(latent_dim)) def global_branch(self, x, mask): bs = x.shape[0] x = x.permute(1, 0, 2) global_tokens = torch.tile(self.global_text_token, (bs,)).reshape(bs, (- 1)) if self.is_cross_token: mean_tokens = torch.tile(self.mean_token, (bs,)).reshape(bs, (- 1)) std_tokens = torch.tile(self.std_token, (bs,)).reshape(bs, (- 1)) xseq = torch.cat((mean_tokens[None], std_tokens[None], global_tokens[None], x), 0) token_mask = torch.ones((bs, 3), dtype=bool, device=x.device) aug_mask = torch.cat((token_mask, mask), 1) else: xseq = torch.cat((global_tokens[None], x), 0) token_mask = torch.ones((bs, 1), dtype=bool, device=x.device) aug_mask = torch.cat((token_mask, mask), 1) xseq = self.sequence_pos_encoding(xseq) text_tokens = self.seqTransEncoder(xseq, src_key_padding_mask=(~ aug_mask)) return text_tokens def action_branch(self, x, mask): bs = x.shape[0] mean_tokens = torch.tile(self.mean_token, (bs,)).reshape(bs, (- 1)) std_tokens = torch.tile(self.std_token, (bs,)).reshape(bs, (- 1)) actionSeq = torch.cat((mean_tokens[None], std_tokens[None], x), 0) token_mask = torch.ones((bs, 2), dtype=bool, device=x.device) aug_mask = torch.cat((token_mask, mask), 1) actionSeq = self.sequence_pos_encoding(actionSeq) action_tokens = self.actionTransEncoder(actionSeq, src_key_padding_mask=(~ aug_mask)) return action_tokens[0:2] def forward(self, texts: List[str]): (text_encoded, mask) = self.get_last_hidden_state(texts, return_mask=True) text_emb = self.projection(text_encoded) return text_emb def get_last_hidden_state(self, texts: List[str], return_mask: bool=False): encoded_inputs = self.tokenizer(texts, return_tensors='pt', padding=True) output = self.text_model(**encoded_inputs.to(self.text_model.device)) if (not return_mask): return output.last_hidden_state return (output.last_hidden_state, encoded_inputs.attention_mask.to(dtype=bool))
class MovementConvEncoder(nn.Module): def __init__(self, input_size, hidden_size, output_size): super(MovementConvEncoder, self).__init__() self.main = nn.Sequential(nn.Conv1d(input_size, hidden_size, 4, 2, 1), nn.Dropout(0.2, inplace=True), nn.LeakyReLU(0.2, inplace=True), nn.Conv1d(hidden_size, output_size, 4, 2, 1), nn.Dropout(0.2, inplace=True), nn.LeakyReLU(0.2, inplace=True)) self.out_net = nn.Linear(output_size, output_size) def forward(self, inputs): inputs = inputs.permute(0, 2, 1) outputs = self.main(inputs).permute(0, 2, 1) return self.out_net(outputs)
class MotionEncoderBiGRUCo(nn.Module): def __init__(self, input_size, hidden_size, output_size): super(MotionEncoderBiGRUCo, self).__init__() self.input_emb = nn.Linear(input_size, hidden_size) self.gru = nn.GRU(hidden_size, hidden_size, batch_first=True, bidirectional=True) self.output_net = nn.Sequential(nn.Linear((hidden_size * 2), hidden_size), nn.LayerNorm(hidden_size), nn.LeakyReLU(0.2, inplace=True), nn.Linear(hidden_size, output_size)) self.hidden_size = hidden_size self.hidden = nn.Parameter(torch.randn((2, 1, self.hidden_size), requires_grad=True)) def forward(self, inputs, m_lens): num_samples = inputs.shape[0] input_embs = self.input_emb(inputs) hidden = self.hidden.repeat(1, num_samples, 1) cap_lens = m_lens.data.tolist() emb = pack_padded_sequence(input_embs, cap_lens, batch_first=True) (gru_seq, gru_last) = self.gru(emb, hidden) gru_last = torch.cat([gru_last[0], gru_last[1]], dim=(- 1)) return self.output_net(gru_last)
class TextEncoderBiGRUCo(nn.Module): def __init__(self, word_size, pos_size, hidden_size, output_size): super(TextEncoderBiGRUCo, self).__init__() self.pos_emb = nn.Linear(pos_size, word_size) self.input_emb = nn.Linear(word_size, hidden_size) self.gru = nn.GRU(hidden_size, hidden_size, batch_first=True, bidirectional=True) self.output_net = nn.Sequential(nn.Linear((hidden_size * 2), hidden_size), nn.LayerNorm(hidden_size), nn.LeakyReLU(0.2, inplace=True), nn.Linear(hidden_size, output_size)) self.hidden_size = hidden_size self.hidden = nn.Parameter(torch.randn((2, 1, self.hidden_size), requires_grad=True)) def forward(self, word_embs, pos_onehot, cap_lens): num_samples = word_embs.shape[0] pos_embs = self.pos_emb(pos_onehot) inputs = (word_embs + pos_embs) input_embs = self.input_emb(inputs) hidden = self.hidden.repeat(1, num_samples, 1) cap_lens = cap_lens.data.tolist() emb = pack_padded_sequence(input_embs, cap_lens, batch_first=True) (gru_seq, gru_last) = self.gru(emb, hidden) gru_last = torch.cat([gru_last[0], gru_last[1]], dim=(- 1)) return self.output_net(gru_last)
class STGCN(nn.Module): 'Spatial temporal graph convolutional networks.\n Args:\n in_channels (int): Number of channels in the input data\n num_class (int): Number of classes for the classification task\n graph_args (dict): The arguments for building the graph\n edge_importance_weighting (bool): If ``True``, adds a learnable\n importance weighting to the edges of the graph\n **kwargs (optional): Other parameters for graph convolution units\n Shape:\n - Input: :math:`(N, in_channels, T_{in}, V_{in}, M_{in})`\n - Output: :math:`(N, num_class)` where\n :math:`N` is a batch size,\n :math:`T_{in}` is a length of input sequence,\n :math:`V_{in}` is the number of graph nodes,\n :math:`M_{in}` is the number of instance in a frame.\n ' def __init__(self, in_channels, num_class, kintree_path, graph_args, edge_importance_weighting, **kwargs): super().__init__() self.num_class = num_class self.losses = ['accuracy', 'cross_entropy', 'mixed'] self.criterion = torch.nn.CrossEntropyLoss(reduction='mean') self.graph = Graph(kintree_path=kintree_path, **graph_args) A = torch.tensor(self.graph.A, dtype=torch.float32, requires_grad=False) self.register_buffer('A', A) spatial_kernel_size = A.size(0) temporal_kernel_size = 9 kernel_size = (temporal_kernel_size, spatial_kernel_size) self.data_bn = nn.BatchNorm1d((in_channels * A.size(1))) kwargs0 = {k: v for (k, v) in kwargs.items() if (k != 'dropout')} self.st_gcn_networks = nn.ModuleList((st_gcn(in_channels, 64, kernel_size, 1, residual=False, **kwargs0), st_gcn(64, 64, kernel_size, 1, **kwargs), st_gcn(64, 64, kernel_size, 1, **kwargs), st_gcn(64, 64, kernel_size, 1, **kwargs), st_gcn(64, 128, kernel_size, 2, **kwargs), st_gcn(128, 128, kernel_size, 1, **kwargs), st_gcn(128, 128, kernel_size, 1, **kwargs), st_gcn(128, 256, kernel_size, 2, **kwargs), st_gcn(256, 256, kernel_size, 1, **kwargs), st_gcn(256, 256, kernel_size, 1, **kwargs))) if edge_importance_weighting: self.edge_importance = nn.ParameterList([nn.Parameter(torch.ones(self.A.size())) for i in self.st_gcn_networks]) else: self.edge_importance = ([1] * len(self.st_gcn_networks)) self.fcn = nn.Conv2d(256, num_class, kernel_size=1) def forward(self, motion): batch = {'output': motion} x = batch['output'].permute(0, 2, 3, 1).unsqueeze(4).contiguous() (N, C, T, V, M) = x.size() x = x.permute(0, 4, 3, 1, 2).contiguous() x = x.view((N * M), (V * C), T) x = self.data_bn(x) x = x.view(N, M, V, C, T) x = x.permute(0, 1, 3, 4, 2).contiguous() x = x.view((N * M), C, T, V) for (gcn, importance) in zip(self.st_gcn_networks, self.edge_importance): (x, _) = gcn(x, (self.A * importance)) x = F.avg_pool2d(x, x.size()[2:]) x = x.view(N, M, (- 1), 1, 1).mean(dim=1) batch['features'] = x.squeeze() x = self.fcn(x) x = x.view(x.size(0), (- 1)) batch['yhat'] = x return batch def compute_accuracy(self, batch): confusion = torch.zeros(self.num_class, self.num_class, dtype=int) yhat = batch['yhat'].max(dim=1).indices ygt = batch['y'] for (label, pred) in zip(ygt, yhat): confusion[label][pred] += 1 accuracy = (torch.trace(confusion) / torch.sum(confusion)) return accuracy def compute_loss(self, batch): cross_entropy = self.criterion(batch['yhat'], batch['y']) mixed_loss = cross_entropy acc = self.compute_accuracy(batch) losses = {'cross_entropy': cross_entropy.item(), 'mixed': mixed_loss.item(), 'accuracy': acc.item()} return (mixed_loss, losses)
class st_gcn(nn.Module): 'Applies a spatial temporal graph convolution over an input graph sequence.\n Args:\n in_channels (int): Number of channels in the input sequence data\n out_channels (int): Number of channels produced by the convolution\n kernel_size (tuple): Size of the temporal convolving kernel and graph convolving kernel\n stride (int, optional): Stride of the temporal convolution. Default: 1\n dropout (int, optional): Dropout rate of the final output. Default: 0\n residual (bool, optional): If ``True``, applies a residual mechanism. Default: ``True``\n Shape:\n - Input[0]: Input graph sequence in :math:`(N, in_channels, T_{in}, V)` format\n - Input[1]: Input graph adjacency matrix in :math:`(K, V, V)` format\n - Output[0]: Outpu graph sequence in :math:`(N, out_channels, T_{out}, V)` format\n - Output[1]: Graph adjacency matrix for output data in :math:`(K, V, V)` format\n where\n :math:`N` is a batch size,\n :math:`K` is the spatial kernel size, as :math:`K == kernel_size[1]`,\n :math:`T_{in}/T_{out}` is a length of input/output sequence,\n :math:`V` is the number of graph nodes.\n ' def __init__(self, in_channels, out_channels, kernel_size, stride=1, dropout=0, residual=True): super().__init__() assert (len(kernel_size) == 2) assert ((kernel_size[0] % 2) == 1) padding = (((kernel_size[0] - 1) // 2), 0) self.gcn = ConvTemporalGraphical(in_channels, out_channels, kernel_size[1]) self.tcn = nn.Sequential(nn.BatchNorm2d(out_channels), nn.ReLU(inplace=True), nn.Conv2d(out_channels, out_channels, (kernel_size[0], 1), (stride, 1), padding), nn.BatchNorm2d(out_channels), nn.Dropout(dropout, inplace=True)) if (not residual): self.residual = (lambda x: 0) elif ((in_channels == out_channels) and (stride == 1)): self.residual = (lambda x: x) else: self.residual = nn.Sequential(nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=(stride, 1)), nn.BatchNorm2d(out_channels)) self.relu = nn.ReLU(inplace=True) def forward(self, x, A): res = self.residual(x) (x, A) = self.gcn(x, A) x = (self.tcn(x) + res) return (self.relu(x), A)
class Graph(): " The Graph to model the skeletons extracted by the openpose\n Args:\n strategy (string): must be one of the follow candidates\n - uniform: Uniform Labeling\n - distance: Distance Partitioning\n - spatial: Spatial Configuration\n For more information, please refer to the section 'Partition Strategies'\n in our paper (https://arxiv.org/abs/1801.07455).\n layout (string): must be one of the follow candidates\n - openpose: Is consists of 18 joints. For more information, please\n refer to https://github.com/CMU-Perceptual-Computing-Lab/openpose#output\n - ntu-rgb+d: Is consists of 25 joints. For more information, please\n refer to https://github.com/shahroudy/NTURGB-D\n - smpl: Consists of 24/23 joints with without global rotation.\n max_hop (int): the maximal distance between two connected nodes\n dilation (int): controls the spacing between the kernel points\n " def __init__(self, kintree_path, layout='openpose', strategy='uniform', max_hop=1, dilation=1): self.max_hop = max_hop self.dilation = dilation self.kintree_path = kintree_path self.get_edge(layout) self.hop_dis = get_hop_distance(self.num_node, self.edge, max_hop=max_hop) self.get_adjacency(strategy) def __str__(self): return self.A def get_edge(self, layout): if (layout == 'openpose'): self.num_node = 18 self_link = [(i, i) for i in range(self.num_node)] neighbor_link = [(4, 3), (3, 2), (7, 6), (6, 5), (13, 12), (12, 11), (10, 9), (9, 8), (11, 5), (8, 2), (5, 1), (2, 1), (0, 1), (15, 0), (14, 0), (17, 15), (16, 14)] self.edge = (self_link + neighbor_link) self.center = 1 elif (layout == 'smpl'): self.num_node = 24 self_link = [(i, i) for i in range(self.num_node)] kt = pkl.load(open(self.kintree_path, 'rb')) neighbor_link = [(k, kt[1][(i + 1)]) for (i, k) in enumerate(kt[0][1:])] self.edge = (self_link + neighbor_link) self.center = 0 elif (layout == 'smpl_noglobal'): self.num_node = 23 self_link = [(i, i) for i in range(self.num_node)] kt = pkl.load(open(self.kintree_path, 'rb')) neighbor_link = [(k, kt[1][(i + 1)]) for (i, k) in enumerate(kt[0][1:])] neighbor_1base = [n for n in neighbor_link if ((n[0] != 0) and (n[1] != 0))] neighbor_link = [((i - 1), (j - 1)) for (i, j) in neighbor_1base] self.edge = (self_link + neighbor_link) self.center = 0 elif (layout == 'ntu-rgb+d'): self.num_node = 25 self_link = [(i, i) for i in range(self.num_node)] neighbor_1base = [(1, 2), (2, 21), (3, 21), (4, 3), (5, 21), (6, 5), (7, 6), (8, 7), (9, 21), (10, 9), (11, 10), (12, 11), (13, 1), (14, 13), (15, 14), (16, 15), (17, 1), (18, 17), (19, 18), (20, 19), (22, 23), (23, 8), (24, 25), (25, 12)] neighbor_link = [((i - 1), (j - 1)) for (i, j) in neighbor_1base] self.edge = (self_link + neighbor_link) self.center = (21 - 1) elif (layout == 'ntu_edge'): self.num_node = 24 self_link = [(i, i) for i in range(self.num_node)] neighbor_1base = [(1, 2), (3, 2), (4, 3), (5, 2), (6, 5), (7, 6), (8, 7), (9, 2), (10, 9), (11, 10), (12, 11), (13, 1), (14, 13), (15, 14), (16, 15), (17, 1), (18, 17), (19, 18), (20, 19), (21, 22), (22, 8), (23, 24), (24, 12)] neighbor_link = [((i - 1), (j - 1)) for (i, j) in neighbor_1base] self.edge = (self_link + neighbor_link) self.center = 2 else: raise NotImplementedError('This Layout is not supported') def get_adjacency(self, strategy): valid_hop = range(0, (self.max_hop + 1), self.dilation) adjacency = np.zeros((self.num_node, self.num_node)) for hop in valid_hop: adjacency[(self.hop_dis == hop)] = 1 normalize_adjacency = normalize_digraph(adjacency) if (strategy == 'uniform'): A = np.zeros((1, self.num_node, self.num_node)) A[0] = normalize_adjacency self.A = A elif (strategy == 'distance'): A = np.zeros((len(valid_hop), self.num_node, self.num_node)) for (i, hop) in enumerate(valid_hop): A[i][(self.hop_dis == hop)] = normalize_adjacency[(self.hop_dis == hop)] self.A = A elif (strategy == 'spatial'): A = [] for hop in valid_hop: a_root = np.zeros((self.num_node, self.num_node)) a_close = np.zeros((self.num_node, self.num_node)) a_further = np.zeros((self.num_node, self.num_node)) for i in range(self.num_node): for j in range(self.num_node): if (self.hop_dis[(j, i)] == hop): if (self.hop_dis[(j, self.center)] == self.hop_dis[(i, self.center)]): a_root[(j, i)] = normalize_adjacency[(j, i)] elif (self.hop_dis[(j, self.center)] > self.hop_dis[(i, self.center)]): a_close[(j, i)] = normalize_adjacency[(j, i)] else: a_further[(j, i)] = normalize_adjacency[(j, i)] if (hop == 0): A.append(a_root) else: A.append((a_root + a_close)) A.append(a_further) A = np.stack(A) self.A = A else: raise NotImplementedError('This Strategy is not supported')
class ConvTemporalGraphical(nn.Module): 'The basic module for applying a graph convolution.\n Args:\n in_channels (int): Number of channels in the input sequence data\n out_channels (int): Number of channels produced by the convolution\n kernel_size (int): Size of the graph convolving kernel\n t_kernel_size (int): Size of the temporal convolving kernel\n t_stride (int, optional): Stride of the temporal convolution. Default: 1\n t_padding (int, optional): Temporal zero-padding added to both sides of\n the input. Default: 0\n t_dilation (int, optional): Spacing between temporal kernel elements.\n Default: 1\n bias (bool, optional): If ``True``, adds a learnable bias to the output.\n Default: ``True``\n Shape:\n - Input[0]: Input graph sequence in :math:`(N, in_channels, T_{in}, V)` format\n - Input[1]: Input graph adjacency matrix in :math:`(K, V, V)` format\n - Output[0]: Outpu graph sequence in :math:`(N, out_channels, T_{out}, V)` format\n - Output[1]: Graph adjacency matrix for output data in :math:`(K, V, V)` format\n where\n :math:`N` is a batch size,\n :math:`K` is the spatial kernel size, as :math:`K == kernel_size[1]`,\n :math:`T_{in}/T_{out}` is a length of input/output sequence,\n :math:`V` is the number of graph nodes.\n ' def __init__(self, in_channels, out_channels, kernel_size, t_kernel_size=1, t_stride=1, t_padding=0, t_dilation=1, bias=True): super().__init__() self.kernel_size = kernel_size self.conv = nn.Conv2d(in_channels, (out_channels * kernel_size), kernel_size=(t_kernel_size, 1), padding=(t_padding, 0), stride=(t_stride, 1), dilation=(t_dilation, 1), bias=bias) def forward(self, x, A): assert (A.size(0) == self.kernel_size) x = self.conv(x) (n, kc, t, v) = x.size() x = x.view(n, self.kernel_size, (kc // self.kernel_size), t, v) x = torch.einsum('nkctv,kvw->nctw', (x, A)) return (x.contiguous(), A)
def get_hop_distance(num_node, edge, max_hop=1): A = np.zeros((num_node, num_node)) for (i, j) in edge: A[(j, i)] = 1 A[(i, j)] = 1 hop_dis = (np.zeros((num_node, num_node)) + np.inf) transfer_mat = [np.linalg.matrix_power(A, d) for d in range((max_hop + 1))] arrive_mat = (np.stack(transfer_mat) > 0) for d in range(max_hop, (- 1), (- 1)): hop_dis[arrive_mat[d]] = d return hop_dis
def normalize_digraph(A): Dl = np.sum(A, 0) num_node = A.shape[0] Dn = np.zeros((num_node, num_node)) for i in range(num_node): if (Dl[i] > 0): Dn[(i, i)] = (Dl[i] ** (- 1)) AD = np.dot(A, Dn) return AD
def normalize_undigraph(A): Dl = np.sum(A, 0) num_node = A.shape[0] Dn = np.zeros((num_node, num_node)) for i in range(num_node): if (Dl[i] > 0): Dn[(i, i)] = (Dl[i] ** (- 0.5)) DAD = np.dot(np.dot(Dn, A), Dn) return DAD
class VPosert(nn.Module): def __init__(self, cfg, **kwargs) -> None: super(VPosert, self).__init__() num_neurons = 512 self.latentD = 256 n_features = (196 * 263) self.encoder_net = nn.Sequential(BatchFlatten(), nn.BatchNorm1d(n_features), nn.Linear(n_features, num_neurons), nn.LeakyReLU(), nn.BatchNorm1d(num_neurons), nn.Dropout(0.1), nn.Linear(num_neurons, num_neurons), nn.Linear(num_neurons, num_neurons), NormalDistDecoder(num_neurons, self.latentD)) self.decoder_net = nn.Sequential(nn.Linear(self.latentD, num_neurons), nn.LeakyReLU(), nn.Dropout(0.1), nn.Linear(num_neurons, num_neurons), nn.LeakyReLU(), nn.Linear(num_neurons, n_features), ContinousRotReprDecoder()) def forward(self, features: Tensor, lengths: Optional[List[int]]=None): q_z = self.encode(features) feats_rst = self.decode(q_z) return (feats_rst, q_z) def encode(self, pose_body, lengths: Optional[List[int]]=None): "\n :param Pin: Nx(numjoints*3)\n :param rep_type: 'matrot'/'aa' for matrix rotations or axis-angle\n :return:\n " q_z = self.encoder_net(pose_body) q_z_sample = q_z.rsample() return (q_z_sample.unsqueeze(0), q_z) def decode(self, Zin, lengths: Optional[List[int]]=None): bs = Zin.shape[0] Zin = Zin[0] prec = self.decoder_net(Zin) return prec
class BatchFlatten(nn.Module): def __init__(self): super(BatchFlatten, self).__init__() self._name = 'batch_flatten' def forward(self, x): return x.view(x.shape[0], (- 1))
class ContinousRotReprDecoder(nn.Module): def __init__(self): super(ContinousRotReprDecoder, self).__init__() def forward(self, module_input): reshaped_input = module_input.view((- 1), 196, 263) return reshaped_input
class NormalDistDecoder(nn.Module): def __init__(self, num_feat_in, latentD): super(NormalDistDecoder, self).__init__() self.mu = nn.Linear(num_feat_in, latentD) self.logvar = nn.Linear(num_feat_in, latentD) def forward(self, Xout): return torch.distributions.normal.Normal(self.mu(Xout), F.softplus(self.logvar(Xout)))
def get_model(cfg, datamodule, phase='train'): modeltype = cfg.model.model_type if (modeltype == 'mld'): return get_module(cfg, datamodule) else: raise ValueError(f'Invalid model type {modeltype}.')
def get_module(cfg, datamodule): modeltype = cfg.model.model_type model_module = importlib.import_module(f'.modeltype.{cfg.model.model_type}', package='mld.models') Model = model_module.__getattribute__(f'{modeltype.upper()}') return Model(cfg=cfg, datamodule=datamodule)
class ACTORLosses(Metric): '\n Loss\n Modify loss\n \n ' def __init__(self, vae, mode, cfg): super().__init__(dist_sync_on_step=cfg.LOSS.DIST_SYNC_ON_STEP) self.vae = vae self.mode = mode losses = [] losses.append('recons_feature') losses.append('recons_verts') losses.append('recons_joints') losses.append('recons_limb') losses.append('latent_st2sm') losses.append('kl_motion') losses.append('total') for loss in losses: self.register_buffer(loss, torch.tensor(0.0)) self.register_buffer('count', torch.tensor(0)) self.losses = losses self._losses_func = {} self._params = {} for loss in losses: if (loss != 'total'): if (loss.split('_')[0] == 'kl'): self._losses_func[loss] = KLLoss() self._params[loss] = cfg.LOSS.LAMBDA_KL elif (loss.split('_')[0] == 'recons'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_REC elif (loss.split('_')[0] == 'cross'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_CROSS elif (loss.split('_')[0] == 'latent'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_LATENT elif (loss.split('_')[0] == 'cycle'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_CYCLE else: ValueError('This loss is not recognized.') def update(self, rs_set, dist_ref): total: float = 0.0 total += self._update_loss('recons_feature', rs_set['m_rst'], rs_set['m_ref']) total += self._update_loss('kl_motion', rs_set['dist_m'], dist_ref) self.total += total.detach() self.count += 1 return total def compute(self, split): count = getattr(self, 'count') return {loss: (getattr(self, loss) / count) for loss in self.losses} def _update_loss(self, loss: str, outputs, inputs): val = self._losses_func[loss](outputs, inputs) getattr(self, loss).__iadd__(val.detach()) weighted_loss = (self._params[loss] * val) return weighted_loss def loss2logname(self, loss: str, split: str): if (loss == 'total'): log_name = f'{loss}/{split}' else: (loss_type, name) = loss.split('_') log_name = f'{loss_type}/{name}/{split}' return log_name
class KLLoss(): def __init__(self): pass def __call__(self, q, p): div = torch.distributions.kl_divergence(q, p) return div.mean() def __repr__(self): return 'KLLoss()'
class KLLossMulti(): def __init__(self): self.klloss = KLLoss() def __call__(self, qlist, plist): return sum([self.klloss(q, p) for (q, p) in zip(qlist, plist)]) def __repr__(self): return 'KLLossMulti()'
class KLLoss(): def __init__(self): pass def __call__(self, q, p): div = torch.distributions.kl_divergence(q, p) return div.mean() def __repr__(self): return 'KLLoss()'
class KLLossMulti(): def __init__(self): self.klloss = KLLoss() def __call__(self, qlist, plist): return sum([self.klloss(q, p) for (q, p) in zip(qlist, plist)]) def __repr__(self): return 'KLLossMulti()'
class MLDLosses(Metric): '\n MLD Loss\n ' def __init__(self, vae, mode, cfg): super().__init__(dist_sync_on_step=cfg.LOSS.DIST_SYNC_ON_STEP) self.vae_type = cfg.TRAIN.ABLATION.VAE_TYPE self.mode = mode self.cfg = cfg self.predict_epsilon = cfg.TRAIN.ABLATION.PREDICT_EPSILON self.stage = cfg.TRAIN.STAGE losses = [] if (self.stage in ['diffusion', 'vae_diffusion']): losses.append('inst_loss') losses.append('x_loss') if (self.cfg.LOSS.LAMBDA_PRIOR != 0.0): losses.append('prior_loss') if (self.stage in ['vae', 'vae_diffusion']): losses.append('recons_feature') losses.append('recons_verts') losses.append('recons_joints') losses.append('recons_limb') losses.append('gen_feature') losses.append('gen_joints') losses.append('kl_motion') if (self.stage not in ['vae', 'diffusion', 'vae_diffusion']): raise ValueError(f'Stage {self.stage} not supported') losses.append('total') for loss in losses: self.add_state(loss, default=torch.tensor(0.0), dist_reduce_fx='sum') self.add_state('count', torch.tensor(0), dist_reduce_fx='sum') self.losses = losses self._losses_func = {} self._params = {} for loss in losses: if (loss.split('_')[0] == 'inst'): self._losses_func[loss] = nn.MSELoss(reduction='mean') self._params[loss] = 1 elif (loss.split('_')[0] == 'x'): self._losses_func[loss] = nn.MSELoss(reduction='mean') self._params[loss] = 1 elif (loss.split('_')[0] == 'prior'): self._losses_func[loss] = nn.MSELoss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_PRIOR if (loss.split('_')[0] == 'kl'): if (cfg.LOSS.LAMBDA_KL != 0.0): self._losses_func[loss] = KLLoss() self._params[loss] = cfg.LOSS.LAMBDA_KL elif (loss.split('_')[0] == 'recons'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_REC elif (loss.split('_')[0] == 'gen'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_GEN elif (loss.split('_')[0] == 'latent'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_LATENT else: ValueError('This loss is not recognized.') if (loss.split('_')[(- 1)] == 'joints'): self._params[loss] = cfg.LOSS.LAMBDA_JOINT def update(self, rs_set): total: float = 0.0 if (self.stage in ['vae', 'vae_diffusion']): total += self._update_loss('recons_feature', rs_set['m_rst'], rs_set['m_ref']) total += self._update_loss('recons_joints', rs_set['joints_rst'], rs_set['joints_ref']) total += self._update_loss('kl_motion', rs_set['dist_m'], rs_set['dist_ref']) if (self.stage in ['diffusion', 'vae_diffusion']): if self.predict_epsilon: total += self._update_loss('inst_loss', rs_set['noise_pred_1'], rs_set['noise_1']) total += self._update_loss('inst_loss', rs_set['noise_pred_2'], rs_set['noise_2']) total += self._update_loss('inst_loss', rs_set['noise_pred_3'], rs_set['noise_3']) else: total += self._update_loss('x_loss', rs_set['pred'], rs_set['latent']) if (self.cfg.LOSS.LAMBDA_PRIOR != 0.0): total += self._update_loss('prior_loss', rs_set['noise_prior'], rs_set['dist_m1']) if (self.stage in ['vae_diffusion']): total += self._update_loss('gen_feature', rs_set['gen_m_rst'], rs_set['m_ref']) total += self._update_loss('gen_joints', rs_set['gen_joints_rst'], rs_set['joints_ref']) self.total += total.detach() self.count += 1 return total def compute(self, split): count = getattr(self, 'count') return {loss: (getattr(self, loss) / count) for loss in self.losses} def _update_loss(self, loss: str, outputs, inputs): val = self._losses_func[loss](outputs, inputs) getattr(self, loss).__iadd__(val.detach()) weighted_loss = (self._params[loss] * val) return weighted_loss def loss2logname(self, loss: str, split: str): if (loss == 'total'): log_name = f'{loss}/{split}' else: (loss_type, name) = loss.split('_') log_name = f'{loss_type}/{name}/{split}' return log_name
class KLLoss(): def __init__(self): pass def __call__(self, q, p): div = torch.distributions.kl_divergence(q, p) return div.mean() def __repr__(self): return 'KLLoss()'
class KLLossMulti(): def __init__(self): self.klloss = KLLoss() def __call__(self, qlist, plist): return sum([self.klloss(q, p) for (q, p) in zip(qlist, plist)]) def __repr__(self): return 'KLLossMulti()'
class TemosLosses(Metric): "\n Loss\n Modify loss\n refer to temos loss\n add loss like deep-motion-editing\n 'gen_loss_total': l_total,\n 'gen_loss_adv': l_adv,\n 'gen_loss_recon_all': l_rec,\n 'gen_loss_recon_r': l_r_rec,\n 'gen_loss_recon_s': l_s_rec,\n 'gen_loss_feature_all': l_ft,\n 'gen_loss_feature_r': l_ft_r,\n 'gen_loss_feature_s': l_ft_s,\n 'gen_loss_feature_t': l_ft_t,\n 'gen_loss_quaternion': l_qt,\n 'gen_loss_twist': l_tw,\n 'gen_loss_triplet': l_triplet,\n 'gen_loss_joint': l_joint,\n \n " def __init__(self, vae, mode, cfg): super().__init__(dist_sync_on_step=cfg.LOSS.DIST_SYNC_ON_STEP) self.vae = vae self.mode = mode loss_on_both = False force_loss_on_jfeats = True ablation_no_kl_combine = False ablation_no_kl_gaussian = False ablation_no_motionencoder = False self.loss_on_both = loss_on_both self.ablation_no_kl_combine = ablation_no_kl_combine self.ablation_no_kl_gaussian = ablation_no_kl_gaussian self.ablation_no_motionencoder = ablation_no_motionencoder losses = [] if ((mode == 'xyz') or force_loss_on_jfeats): if (not ablation_no_motionencoder): losses.append('recons_jfeats2jfeats') losses.append('recons_text2jfeats') if (mode == 'smpl'): if (not ablation_no_motionencoder): losses.append('recons_rfeats2rfeats') losses.append('recons_text2rfeats') else: ValueError('This mode is not recognized.') if (vae or loss_on_both): kl_losses = [] if ((not ablation_no_kl_combine) and (not ablation_no_motionencoder)): kl_losses.extend(['kl_text2motion', 'kl_motion2text']) if (not ablation_no_kl_gaussian): if ablation_no_motionencoder: kl_losses.extend(['kl_text']) else: kl_losses.extend(['kl_text', 'kl_motion']) losses.extend(kl_losses) if ((not self.vae) or loss_on_both): if (not ablation_no_motionencoder): losses.append('latent_manifold') losses.append('total') for loss in losses: self.register_buffer(loss, torch.tensor(0.0)) self.register_buffer('count', torch.tensor(0)) self.losses = losses self._losses_func = {} self._params = {} for loss in losses: if (loss != 'total'): if (loss.split('_')[0] == 'kl'): self._losses_func[loss] = KLLoss() self._params[loss] = cfg.LOSS.LAMBDA_KL elif (loss.split('_')[0] == 'recons'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_REC elif (loss.split('_')[0] == 'latent'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_LATENT elif (loss.split('_')[0] == 'cycle'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_CYCLE else: ValueError('This loss is not recognized.') def update(self, f_text=None, f_motion=None, f_ref=None, lat_text=None, lat_motion=None, dis_text=None, dis_motion=None, dis_ref=None): total: float = 0.0 if ((self.mode == 'xyz') or self.force_loss_on_jfeats): if (not self.ablation_no_motionencoder): total += self._update_loss('recons_jfeats2jfeats', f_motion, f_ref) total += self._update_loss('recons_text2jfeats', f_text, f_ref) if (self.mode == 'smpl'): if (not self.ablation_no_motionencoder): total += self._update_loss('recons_rfeats2rfeats', f_motion.rfeats, f_ref.rfeats) total += self._update_loss('recons_text2rfeats', f_text.rfeats, f_ref.rfeats) if (self.vae or self.loss_on_both): if ((not self.ablation_no_kl_combine) and (not self.ablation_no_motionencoder)): total += self._update_loss('kl_text2motion', dis_text, dis_motion) total += self._update_loss('kl_motion2text', dis_motion, dis_text) if (not self.ablation_no_kl_gaussian): total += self._update_loss('kl_text', dis_text, dis_ref) if (not self.ablation_no_motionencoder): total += self._update_loss('kl_motion', dis_motion, dis_ref) if ((not self.vae) or self.loss_on_both): if (not self.ablation_no_motionencoder): total += self._update_loss('latent_manifold', lat_text, lat_motion) self.total += total.detach() self.count += 1 return total def compute(self, split): count = getattr(self, 'count') return {loss: (getattr(self, loss) / count) for loss in self.losses} def _update_loss(self, loss: str, outputs, inputs): val = self._losses_func[loss](outputs, inputs) getattr(self, loss).__iadd__(val.detach()) weighted_loss = (self._params[loss] * val) return weighted_loss def loss2logname(self, loss: str, split: str): if (loss == 'total'): log_name = f'{loss}/{split}' else: (loss_type, name) = loss.split('_') log_name = f'{loss_type}/{name}/{split}' return log_name
class KLLoss(): def __init__(self): pass def __call__(self, q, p): div = torch.distributions.kl_divergence(q, p) return div.mean() def __repr__(self): return 'KLLoss()'
class KLLossMulti(): def __init__(self): self.klloss = KLLoss() def __call__(self, qlist, plist): return sum([self.klloss(q, p) for (q, p) in zip(qlist, plist)]) def __repr__(self): return 'KLLossMulti()'
class TmostLosses(Metric): "\n Loss\n Modify loss\n refer to temos loss\n add loss like deep-motion-editing\n 'gen_loss_total': l_total,\n 'gen_loss_adv': l_adv,\n 'gen_loss_recon_all': l_rec,\n 'gen_loss_recon_r': l_r_rec,\n 'gen_loss_recon_s': l_s_rec,\n 'gen_loss_feature_all': l_ft,\n 'gen_loss_feature_r': l_ft_r,\n 'gen_loss_feature_s': l_ft_s,\n 'gen_loss_feature_t': l_ft_t,\n 'gen_loss_quaternion': l_qt,\n 'gen_loss_twist': l_tw,\n 'gen_loss_triplet': l_triplet,\n 'gen_loss_joint': l_joint,\n \n " def __init__(self, vae, mode, cfg): super().__init__(dist_sync_on_step=cfg.LOSS.DIST_SYNC_ON_STEP) self.vae = vae self.mode = mode losses = [] losses.append('recons_mm2m') losses.append('recons_t2m') losses.append('cross_mt2m') losses.append('cross_tm2m') losses.append('cycle_cmsm2mContent') losses.append('cycle_cmsm2mStyle') losses.append('latent_ct2cm') losses.append('latent_st2sm') losses.append('kl_motion') losses.append('kl_text') losses.append('kl_ct2cm') losses.append('kl_cm2ct') losses.append('total') for loss in losses: self.register_buffer(loss, torch.tensor(0.0)) self.register_buffer('count', torch.tensor(0)) self.losses = losses self.ablation_cycle = cfg.TRAIN.ABLATION.CYCLE self._losses_func = {} self._params = {} for loss in losses: if (loss != 'total'): if (loss.split('_')[0] == 'kl'): self._losses_func[loss] = KLLoss() self._params[loss] = cfg.LOSS.LAMBDA_KL elif (loss.split('_')[0] == 'recons'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_REC elif (loss.split('_')[0] == 'cross'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_CROSS elif (loss.split('_')[0] == 'latent'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_LATENT elif (loss.split('_')[0] == 'cycle'): self._losses_func[loss] = torch.nn.SmoothL1Loss(reduction='mean') self._params[loss] = cfg.LOSS.LAMBDA_CYCLE else: ValueError('This loss is not recognized.') def update(self, rs_set, dist_ref): total: float = 0.0 '\n loss list\n - triplet loss\n - anchor style1\n - pos style2\n - neg diff_style\n anchor = s_xa\n pos = s_xpos\n neg = self.gen.enc_style(co_data[diff_style], diff_style[-2:])\n l_triplet = self.triplet_loss(anchor, pos, neg)\n - \n ' total += self._update_loss('recons_mm2m', rs_set['rs_cm1sm1'], rs_set['m1']) total += self._update_loss('recons_t2m', rs_set['rs_ct1st1'], rs_set['m1']) total += self._update_loss('cross_mt2m', rs_set['rs_cm1st1'], rs_set['m1']) total += self._update_loss('cross_tm2m', rs_set['rs_ct1sm1'], rs_set['m1']) if self.ablation_cycle: total += self._update_loss('cycle_cmsm2mContent', rs_set['cyc_rs_cm1sm1'], rs_set['m1']) total += self._update_loss('cycle_cmsm2mStyle', rs_set['cyc_rs_cm2sm2'], rs_set['m2']) total += self._update_loss('latent_ct2cm', rs_set['lat_ct1'], rs_set['lat_cm1']) total += self._update_loss('latent_st2sm', rs_set['lat_st1'], rs_set['lat_sm1']) total += self._update_loss('kl_motion', rs_set['dist_cm1'], dist_ref) total += self._update_loss('kl_text', rs_set['dist_ct1'], dist_ref) total += self._update_loss('kl_ct2cm', rs_set['dist_ct1'], rs_set['dist_cm1']) total += self._update_loss('kl_cm2ct', rs_set['dist_cm1'], rs_set['dist_ct1']) self.total += total.detach() self.count += 1 return total def compute(self, split): count = getattr(self, 'count') return {loss: (getattr(self, loss) / count) for loss in self.losses} def _update_loss(self, loss: str, outputs, inputs): val = self._losses_func[loss](outputs, inputs) getattr(self, loss).__iadd__(val.detach()) weighted_loss = (self._params[loss] * val) return weighted_loss def loss2logname(self, loss: str, split: str): if (loss == 'total'): log_name = f'{loss}/{split}' else: (loss_type, name) = loss.split('_') log_name = f'{loss_type}/{name}/{split}' return log_name
class KLLoss(): def __init__(self): pass def __call__(self, q, p): div = torch.distributions.kl_divergence(q, p) return div.mean() def __repr__(self): return 'KLLoss()'
class KLLossMulti(): def __init__(self): self.klloss = KLLoss() def __call__(self, qlist, plist): return sum([self.klloss(q, p) for (q, p) in zip(qlist, plist)]) def __repr__(self): return 'KLLossMulti()'