When running test.py the saved prediction results are missing appropriate orientation and origin since the output files are initiated with eye(4). This will cause issues when viewing the image data and the segmentation masks in most viewing software. Suggestion would be to remember the orientation and origin from the image volumes and to use this when initiating the output files, e.g. use img.get_affine() when reading the image data.
When running test.py the saved prediction results are missing appropriate orientation and origin since the output files are initiated with eye(4). This will cause issues when viewing the image data and the segmentation masks in most viewing software. Suggestion would be to remember the orientation and origin from the image volumes and to use this when initiating the output files, e.g. use img.get_affine() when reading the image data.