IndexError: too many indices for array about brain-segmentation-pytorch

[mateuszbuda]

Thanks for your interest in this repo.
Image slices are expected to be 3D: 2 spacial dimensions and 1 for 3 channels.
Masks slices are 2D.
One image volume is expected to be 4D: an array of 3D image slices.
Corresponding mask volume is expected to be 3D: an array of 2D mask slices.
At the end of the dataset initialization, mask slices are expanded to 3D volumes with one channel.

# add channel dimension to masks
self.volumes = [(v, m[..., np.newaxis]) for (v, m) in self.volumes]

Variable volumes is expected to be a list of tuples that contain 4D image volume and 4D mask volume.
The order of dimensions is: [slices, height, width, channels].
For example, an image volume with 10 slices and slices of size 200x300 has shape (10, 200, 300, 3) and corresponding mask volume has shape (10, 200, 300, 1).
I hope it helps.

[gireeshkbogu]

Thank you for the quick response and detailed explanation. I converted .nii.gz files to tiff by thinking that all the input to your code is 2D.
Do you know how can I convert .nii.gz files into .tff (3D images and 2D masks)?

[mateuszbuda]

The model is 2D with 3-channel input but on a higher level it segments volumes.
If you have files for only one modality, you can try to copy it to have 3-channel slices or use gray2rgb function from skimage: https://scikit-image.org/docs/dev/api/skimage.color.html#skimage.color.gray2rgb.
If you have files for three modalities, you have to register them first and then read in the same order and concatenate along the lest dimension/axis.
For masks, you can read them with imread function from skimage with as_gray=True: https://scikit-image.org/docs/dev/api/skimage.io.html#skimage.io.imread.