Reproduce

[CTZhou-byte]

This is a very valuable work. I am very interested in UniSoccer. However, I am just an interested reader. Can you provide a more detailed Readme to start this project? I can't wait to see its effect.

During my reproduction process, I found that the corresponding JSON video file was missing. This makes it more difficult for someone who is not familiar with this direction to reproduce. In addition, there are some model paths. Can the author provide a complete demo, at least to quickly achieve inference?

Best,
Zhou

[jyrao]

Sorry for late reply since I was just back from holiday. The form of json file was in https://github.com/jyrao/UniSoccer/blob/main/train_data/json/MatchTime/classification_test.json. The dataloader would load the video path for only and predict the rest of the information like commentary into the csv form. This is the logic of we doing the inference in batch. Maybe the version you like is to end-to-end load the video and generate the output commentary, that is actually really close to the inference.py file. I might modify it later after finishing my current project. Or you can attempt first and we can exchange ideas here. Again thanks for following our work!