Unexpected output from Resize

[kayhan-batmanghelich]

Hi,

Thank you for the nice library!

I am trying to make some course material and started playing with the library. I am still not used to it and this question might me being not familiar enough with monai. I tried to make a simple Resize and I think I am not using the API correctly because output size does not match my expectation. Here is a simple example:

sidelength = 64
loader = LoadNifti(image_only=True)
scaler = ScaleIntensity(minv=-1.0,maxv=1.0)
resizer = Resize(spatial_size=(sidelength, sidelength, sidelength), mode=('trilinear'))
tensorer = ToTensor()

inputImg = loader('392272.long.003_S_5150_base/norm.nii.gz')
print(inputImg.shape)

inputImg_scale = scaler(inputImg)
print(inputImg_scale.shape)

inputImg_scale_resize = resizer(inputImg_scale)
print(inputImg_scale_resize.shape)

inputImg_scale_resize_tensor = tensorer(inputImg_scale_resize)
print(inputImg_scale_resize_tensor.shape)

This is the output:

(256, 256, 256)
(256, 256, 256)
(256, 64, 64, 64)
torch.Size([256, 64, 64, 64])

Output supposed to be (64,64,64) why is it (256, 64, 64, 64)?

[rijobro]

Channel normally comes first, so here it assumes you have 256 channels and each image is 256x256 spatially. If you put in an AddChannel transform before the resizer, I think that'll solve your problem.

We also have a Compose transform, so once you've sorted out your teething problems, it might be simpler to do this (as you can execute all transforms in one go).

[kayhan-batmanghelich]

Thank you @rijobro That resolves the issue. Perhaps a note in the documentation helps although you have it in the transfomer

def get_medicalImg_tensor(sidelength):
    xform = Compose([
        LoadNifti(image_only=True),
        AddChannel(),
        ScaleIntensity(minv=-1.0,maxv=1.0),
        Resize((sidelength, sidelength, sidelength), mode=('trilinear')),
        ToTensor(),
    ])
    img = xform('392272.long.003_S_5150_base/norm.nii.gz')
    return img


img = get_medicalImg_tensor(64)
print(img.shape)