wrong dimension segmentations (146x174x146x72) about tractseg

[wasserth]

Hi,
the original problem was the following: If you resample the HCP image from 1.25mm to 2.5mm resolution you get [73,87,73]. When you upsample this back to 1.25mm you get [146,174,146]. So unfortunately this is different from what you had in the beginning. Now you would have to remember if you downsampled from a uneven shape and then cut one value if you upsample again. That's what is done in TractSeg. For ExpRunner I did not bother doing this. If I need the output to be [145,174,145] I just do data = data[:-1, :, :-1, :], or use TractSeg for inference (using the branch speed_up you can do TractSeg ... --exp_name <your experiment_name> and use TractSeg for inference with your custom experiments instead of doing ExpRunner --seg/--probs.
The code for training your own models (ExpRunner) is not as polished as the code for using the pretrained models (TractSeg) so unfortunately it is a bit more inconvenient to use.
I hope that answered your question.