def train_transforms(image, masks):
seq = augmentation.get_train_augmenters_seq2()
hooks_masks = augmentation.get_train_masks_augmenters_deactivator()
# Convert the stochastic sequence of augmenters to a deterministic one.
# The deterministic sequence will always apply the exactly same effects to the images.
seq_det = seq.to_deterministic(
) # call this for each batch again, NOT only once at the start
image_aug = seq_det.augment_images([image])[0]
masks_aug = seq_det.augment_images([masks], hooks=hooks_masks)[0]
image_aug_tensor = transforms.ToTensor()(image_aug.copy())
masks_aug = (masks_aug >= MASK_THRESHOLD).astype(np.uint8)
return image_aug_tensor, masks_aug
def train_transforms(image, masks, labels=None):
seq = augmentation.get_train_augmenters_seq2(mode='constant')
hooks_masks = augmentation.get_train_masks_augmenters_deactivator()
# Convert the stochastic sequence of augmenters to a deterministic one.
# The deterministic sequence will always apply the exactly same effects to the images.
seq_det = seq.to_deterministic() # call this for each batch again, NOT only once at the start
image = seq_det.augment_images([image])[0]
image = transforms.ToTensor()(image.copy())
image = transforms.Normalize(IMAGES_MEAN, IMAGES_STD)(image)
masks = seq_det.augment_images([masks], hooks=hooks_masks)[0]
masks = (masks >= MASK_THRESHOLD).astype(np.uint8)
if labels is not None:
labels = seq_det.augment_images([labels], hooks=hooks_masks)[0]
labels = (labels > 0).astype(np.uint8)
return image, masks, labels
else:
return image, masks
def train_transforms(image, mask, labels):
seq = augmentation.get_train_augmenters_seq2(mode='constant')
hooks_masks = augmentation.get_train_masks_augmenters_deactivator()
# Convert the stochastic sequence of augmenters to a deterministic one.
# The deterministic sequence will always apply the exactly same effects to the images.
seq_det = seq.to_deterministic(
) # call this for each batch again, NOT only once at the start
image_aug = seq_det.augment_images([image])[0]
image_aug_tensor = transforms.ToTensor()(image_aug.copy())
image_aug_tensor = transforms.Normalize(IMAGES_MEAN,
IMAGES_STD)(image_aug_tensor)
mask_aug = seq_det.augment_images([mask], hooks=hooks_masks)[0]
mask_aug = (mask_aug >= MASK_THRESHOLD).astype(np.uint8)
labels_aug = seq_det.augment_images([labels], hooks=hooks_masks)[0]
for index in range(labels_aug.shape[-1]):
labels_aug[..., index] = (labels_aug[..., index] > 0).astype(np.uint8)
return image_aug_tensor, mask_aug, labels_aug
