akhanss/BraTS-2020 — reverse-engineered prompt

I want a ready to run Python pipeline for the BraTS 2020 brain tumor segmentation challenge using a 3D U net. It should let me point to the BraTS dataset in the usual folder layout, with the training and validation cases and the flair, t1, t1ce, t2, and seg files, then handle preprocessing, create the train and validation split files, train the model, save the best checkpoint, and run inference on validation or test cases to write prediction files into an output folder.

Please make it feel like I can just set up the data folders and run one command to train and one command to do inference. Keep the environment compatible with the older TensorFlow setup this project expects, and include any install notes that matter, especially anything tricky around tables or medical imaging packages. If something is unclear, look up the current docs online and make the repo usable end to end without me having to figure out the wiring myself.