mateuszbuda

Hi @fepegar, thanks for trying the model. The notebook on PyTorch Hub runs inference on one slice. This can be done, but the model was trained on images normalized per...

The problem is that it would require loading the entire 4D MRI volume in the code sample which is run as a test every time there is a PR to...

It simply means that `validation_cases` is larger than `len(self.patients)`. `patients` is a dictionary mapping patient IDs to volumes. Did you run it on your own dataset? You can add this...

Hi, this is correct, the model was trained on 3-channel input images. Trained weights can only be loaded for a model initialized with `in_channels=3`.

Image channels correspond to 3 MRI sequences: pre-contrast, FLAIR, and post-contrast. More detail in the paper: [https://arxiv.org/abs/1906.03720](https://arxiv.org/abs/1906.03720)

All preprocessing steps are implemented in python in this repo.

This removes the first and the last slice from each volume. This is done because they are often corrupted and never contain brain tissue.

Your volumes (input to `resize_sample`) are expected to be 4D arrays of shape [n_slices x slice_height x slice_width x n_channels]. Masks should be 3D arrays of shape [n_slices x slice_height...

Here are some ideas: - Dice loss of 1 indicates no overlap between GT and predictions. Did you observe it on the training or validation set? - Dice loss is...

Hi @manvirvirk If you want to run training on your dataset, you should be able to do this by implementing pytorch Dataset interface and replace [BrainSegmentationDataset](https://github.com/mateuszbuda/brain-segmentation-pytorch/blob/8ef2e2d423b67b53ec8113fc71a9b968bb0f66e7/dataset.py#L12) with it.