This is an unofficial implementation of https://github.com/google-research/google-research/tree/master/uflow
Currently in very early stage, only basic loss functions implemented.
For training with minecraft data, the ep1_pickle_doc.pkl must be placed into dataset/UFlow_data folder (see dataset.py)
Training:
python3 train.py
Torch checkpoint will be stored at save/checkpoint.pt
Testing:
python3 show_results.py
Use --help for flags or check out uflow_flags.py