import torch import numpy as np from torch.utils.data import Dataset import h5py import itertools from scipy import ndimage from torch.utils.data.sampler import Sampler import numpy as np def random_rot_flip(image_1, image_2, label): k = np.random.randint(0, 4) image_1 = np.rot90(image_1, k) image_2 = np.rot90(image_2, k) label = np.rot90(label, k) axis = np.random.randint(0, 2) image_1 = np.flip(image_1, axis=axis).copy() image_2 = np.flip(image_2, axis=axis).copy() label = np.flip(label, axis=axis).copy() return image_1, image_2, label def random_rotate(image_1, image_2, label): angle = np.random.randint(-20, 20) image_1 = ndimage.rotate(image_1, angle, order=0, reshape=False) image_2 = ndimage.rotate(image_2, angle, order=0, reshape=False) label = ndimage.rotate(label, angle, order=0, reshape=False) return image_1, image_2, label class MS(Dataset): """ MS Dataset """ def __init__(self, base_dir=None, split='train', num=None, transform=None): self._base_dir = base_dir self.transform = transform self.sample_list = [] train_path = self._base_dir+'/train.list' test_path = self._base_dir+'/test.list' if split=='train': with open(train_path, 'r') as f: self.image_list = f.readlines() elif split == 'test': with open(test_path, 'r') as f: self.image_list = f.readlines() self.image_list = [item.replace('\n','') for item in self.image_list] print("total {} samples".format(len(self.image_list))) def __len__(self): return len(self.image_list) def __getitem__(self, idx): image_name = self.image_list[idx] if "ms23" in image_name: h5f = h5py.File(image_name, 'r') image_1 = h5f['image'][:] image_2 = h5f['image'][:] label = h5f['label'][:] else: h5f = h5py.File(image_name, 'r') image_1 = h5f['image_1'][:] image_2 = h5f['image_2'][:] label = h5f['label'][:] sample = {'image_1': image_1, 'image_2': image_2, 'label': label} if self.transform: sample = self.transform(sample) return sample class WeightCrop(object): """ Crop randomly the image in a sample Args: output_size (int): Desired output size """ def __init__(self, output_size): self.output_size = output_size def __call__(self, sample): image_1, image_2, label = sample['image_1'], sample['image_2'], sample['label'] # pad the sample if necessary if label.shape[0] <= self.output_size[0] or label.shape[1] <= self.output_size[1] or label.shape[2] <= self.output_size[2]: pw = max((self.output_size[0] - label.shape[0]) // 2 + 3, 0) ph = max((self.output_size[1] - label.shape[1]) // 2 + 3, 0) pd = max((self.output_size[2] - label.shape[2]) // 2 + 3, 0) image_1 = np.pad(image_1, [(pw, pw), (ph, ph), (pd, pd)], mode='constant', constant_values=0) image_2 = np.pad(image_2, [(pw, pw), (ph, ph), (pd, pd)], mode='constant', constant_values=0) label = np.pad(label, [(pw, pw), (ph, ph), (pd, pd)], mode='constant', constant_values=0) (w, h, d) = image_1.shape if label.sum() > 0: mask = np.nonzero(label) num_label_pixel = mask[0].shape[0] center_index =np.random.randint(0, num_label_pixel-1) center_x, center_y, center_z = mask[0][center_index], mask[1][center_index], mask[2][center_index] w1 = np.random.randint(-10, 10)+self.output_size[0]//2 h1 = np.random.randint(-10, 10)+self.output_size[1]//2 d1 = np.random.randint(-10, 10)+self.output_size[2]//2 lefttop_x, lefttop_y, lefttop_z = center_x-w1, center_y-h1, center_z-d1 minx = max(lefttop_x, 0) miny = max(lefttop_y, 0) minz = max(lefttop_z, 0) maxx = minx + self.output_size[0] maxy = miny + self.output_size[1] maxz = minz + self.output_size[2] if maxx>= w or maxy >= h or maxz >=d: maxx = min(maxx, w-1) maxy = min(maxy, h-1) maxz = min(maxz, d-1) minx = maxx - self.output_size[0] miny = maxy - self.output_size[1] minz = maxz - self.output_size[2] label = label[minx:maxx, miny:maxy, minz:maxz] image_1 = image_1[minx:maxx, miny:maxy, minz:maxz] image_2 = image_2[minx:maxx, miny:maxy, minz:maxz] assert(label.shape == self.output_size) else: w1 = np.random.randint(0, w - self.output_size[0]) h1 = np.random.randint(0, h - self.output_size[1]) d1 = np.random.randint(0, d - self.output_size[2]) label = label[w1:w1 + self.output_size[0], h1:h1 + self.output_size[1], d1:d1 + self.output_size[2]] image_1 = image_1[w1:w1 + self.output_size[0], h1:h1 + self.output_size[1], d1:d1 + self.output_size[2]] image_2 = image_2[w1:w1 + self.output_size[0], h1:h1 + self.output_size[1], d1:d1 + self.output_size[2]] assert(label.shape == self.output_size) return {'image_1': image_1, 'image_2': image_2, 'label': label} class RandomRotFlip(object): """ Crop randomly flip the dataset in a sample Args: output_size (int): Desired output size """ def __call__(self, sample): image_1, image_2, label = sample['image_1'], sample['image_2'], sample['label'] image_1, image_2, label = random_rot_flip(image_1, image_2, label) return {'image_1': image_1, 'image_2': image_2, 'label': label} class RandomRot(object): """ Crop randomly flip the dataset in a sample Args: output_size (int): Desired output size """ def __call__(self, sample): image_1, image_2, label = sample['image_1'], sample['image_2'], sample['label'] image_1, image_2, label = random_rotate(image_1, image_2, label) return {'image_1': image_1, 'image_2': image_2, 'label': label} class ToTensor(object): """Convert ndarrays in sample to Tensors.""" def __call__(self, sample): image_1, image_2, label = sample['image_1'], sample['image_2'], sample['label'] image_1 = image_1.reshape(1, image_1.shape[0], image_1.shape[1], image_1.shape[2]).astype(np.float32) image_2 = image_2.reshape(1, image_2.shape[0], image_2.shape[1], image_2.shape[2]).astype(np.float32) return {'image_1': torch.from_numpy(image_1), 'image_2': torch.from_numpy(image_2), 'label': torch.from_numpy(label).long()} class TwoStreamBatchSampler(Sampler): """Iterate two sets of indices An 'epoch' is one iteration through the primary indices. During the epoch, the secondary indices are iterated through as many times as needed. """ def __init__(self, primary_indices, secondary_indices, primary_batch_size, secondary_batch_size): self.primary_indices = primary_indices self.secondary_indices = secondary_indices self.secondary_batch_size = secondary_batch_size self.primary_batch_size = primary_batch_size assert len(self.primary_indices) >= self.primary_batch_size > 0 assert len(self.secondary_indices) >= self.secondary_batch_size > 0 def __iter__(self): primary_iter = iterate_once(self.primary_indices) secondary_iter = iterate_eternally(self.secondary_indices) return ( primary_batch + secondary_batch for (primary_batch, secondary_batch) in zip(grouper(primary_iter, self.primary_batch_size), grouper(secondary_iter, self.secondary_batch_size)) ) def __len__(self): return len(self.primary_indices) // self.primary_batch_size def iterate_once(iterable): return np.random.permutation(iterable) def iterate_eternally(indices): def infinite_shuffles(): while True: yield np.random.permutation(indices) return itertools.chain.from_iterable(infinite_shuffles()) def grouper(iterable, n): "Collect data into fixed-length chunks or blocks" # grouper('ABCDEFG', 3) --> ABC DEF" args = [iter(iterable)] * n return zip(*args)