CropOrPad only crops Mask

TorchIO version 0.15.0

What I did Create the following sequence of transforms:

input_H = 160
input_W = 160
input_D = 80

transforms = [
    torchio.transforms.ToCanonical(),
    torchio.transforms.Resample(target_spacing=(1,1,1)),
    torchio.transforms.RescaleIntensity(out_min_max=(0,1),percentiles=(0.5,99.5)),
    torchio.transforms.ZNormalization(),
    torchio.transforms.CropOrPad(target_shape=(input_D,input_H,input_W),padding_mode='constant',padding_fill=0,mask_name='mask'),
]
transform = Compose(transforms)

dataset = ImagesDataset(subjects, transform=transform)
sample = dataset[0]

I then checked the size of one of the images in the sample e.g. sample['t2']['data'].shape which returns torch.Size([1, 247, 345, 346]) whereas sample['mask']['data'].shape returns torch.Size([1, 80, 160, 160])

What I expected to happen Both the image and mask have the same size

What actually happened The sizes are different.

Possible reason In the crop and pad function (here) these lines should probably be changed

        padding_params = tuple(padding.tolist()) if padding.any() else None
        cropping_params = tuple(cropping.tolist()) if padding.any() else None

should be

        padding_params = tuple(padding.tolist()) if padding.any() else None
        cropping_params = tuple(cropping.tolist()) if cropping.any() else None

[GReguig]

Ok, I will give it a look

[nwschurink]

Ah ok I’ve found out why the check is not working. The shape check is only performed when doing a center crop (e.g _compute_center_crop_or_pad) (here). However, the check is not performed when performing a crop on the mask’s center (e.g. _compute_mask_center_crop_or_pad)