Quo Vadis, Skeleton Action Recognition ?

This repository contains datasets and scripts for the paper Quo Vadis, Skeleton Action Recognition ?

This paper primarily contributes towards benchmarking of exisiting skeleton human action recognition models on larger datasets like NTU-120 and also provides insight on how these models performs for in the wild data as well.

We also introduce 3 new skeleton human action datasets:

• Skeletics-152
• Skeleton-mimetics
• Metaphorics

The details about each of these datasets can be found in skeletics-152 README, skeleton-mimetics README and metaphorics README.

Pre Trained models

This repository contains pre-trained model weights for all the models that were trained as mentioned in the paper. The weights of different models for different datasets can be found here

Scripts

The scripts needed to parse VIBE skeleton data into usable VA-CNN and GCN formats can be found here

Citation

@misc{gupta2020quo,
    title={Quo Vadis, Skeleton Action Recognition ?},
    author={Pranay Gupta and Anirudh Thatipelli and Aditya Aggarwal and Shubh Maheshwari and Neel Trivedi and Sourav Das and Ravi Kiran Sarvadevabhatla},
    year={2020},
    eprint={2007.02072},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}