search

Walter0807 / MotionBERT

[ICCV 2023] PyTorch Implementation of "MotionBERT: A Unified Perspective on Learning Human Motion Representations"
Apache License 2.0
1.02k stars 123 forks source link

What the Xsub, Xview and 1-shot #74

Closed leviethung2103 closed 1 year ago

leviethung2103 commented 1 year ago

Hello,

I am starting to fine tuning the MB. However, I am confused with these terms: Xsub, Xview and 1-shot. Especially the term 1-shot.

In my scenario, I have one camera at the fixed location. How can fine-tune the MB ? Furthermore, my dataset only contains 2 actions (2 classes)

The option should I choose is Cross View, Cross Subject or 1-shot. Thank you

https://github.com/Walter0807/MotionBERT/blob/main/docs/action.md

Walter0807 commented 1 year ago

X-sub or X-view are ways to split the train/test set, which you can decide by yourself on your own dataset and no need to follow the practices on NTU-RGB+D.

The oneshot setting would be suitable if you have extremely few labeled data. Otherwise fine tuning on your dataset might be better.