TODO: `omnitools` includes a PyTorch dataloader that should work out-of-the-box with annotations from the Omnidata annotator

Documentation

Note:: You could even put the implementation itself right here in the notebook. How to do that?

First: add #default_exp torch.dataloader to the top of the notebook. Then, any code cells marked at the top with #export will be written to omnidata-tools/torch/dataloader.py. The notebook will also be converted to documentation (same as it currently is!). So it doesn't require writing separate docs.