---
license: cc-by-4.0
---
Video Test-Time Adaptation for Action Recognition (CVPR 2023)

- [Project Page](https://wlin-at.github.io/vitta)
- [GitHub Repo](https://github.com/wlin-at/ViTTA)
- [Arxiv Paper](https://arxiv.org/abs/2211.15393)

### Dataset Description
This dataset repo contains the following two datasets:  
- `Kinetics400_val_corruptions`: 12 corruption types for the 19877 validation videos on Kinetics400. 
- `SSv2_val_corruptions`: 12 corruption types for the 24777 validation videos on Something-Something v2.