Thank you for this very valuable tool! I'm working on an image-to-image translation task, where I would like to apply the same transformations (augmentations) to input and ground truth images. Is there any way to do this?
I thought that retrieving one augmentation instance per (input, ground truth) tuple might solve the issue. To test this, I entered an (input, input) tuple into the following code and tested whether the augmented images in data_aug are equal. However, this program returns False (although the images visually look the same). This was programmed using 3D scans and volumentations-3D.
def get_augmentation(patch_size):
return Compose([
Rotate((-15, 15), (0, 0), (0, 0), p=0.5),
ElasticTransform((0, 0.25), interpolation=2, p=0.1),
Flip(0, p=0.5),
], p=1.0)
aug = get_augmentation((79, 95, 79)) # 79, 95, 79 is the whole image, not a patch
data = {'image': np.array([input_img, input_img])}
data_aug = aug(**data)
img = data_aug['image']
print((img[0] == img[1]).all()) # this returns False
Moreover, sometimes one of the augmented images is blank. Could there be a mistake in the provided code, which causes this issue?
Thank you for this very valuable tool! I'm working on an image-to-image translation task, where I would like to apply the same transformations (augmentations) to input and ground truth images. Is there any way to do this?
I thought that retrieving one augmentation instance per (input, ground truth) tuple might solve the issue. To test this, I entered an (input, input) tuple into the following code and tested whether the augmented images in data_aug are equal. However, this program returns False (although the images visually look the same). This was programmed using 3D scans and volumentations-3D.
Moreover, sometimes one of the augmented images is blank. Could there be a mistake in the provided code, which causes this issue?
Thank you in advance and best wishes Elena