| | --- |
| | 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. |
