Is Space-Time Attention All You Need for Video Understanding?
Paper • 2102.05095 • Published • 2
TimeSformer — UCF-101 Action Recognition
Implementation of Is Space-Time Attention All You Need for Video Understanding?
Trained on UCF-101 with divided space-time attention.
Model Config
| Param | Value |
|---|---|
| DIM | 768 |
| DEPTH | 12 |
| HEADS | 12 |
| FRAMES | 16 |
| IMG_SIZE | 224 |
| Classes | 101 |
Results
| Split | Top-1 | Top-5 |
|---|---|---|
| Val | 92.89% | 97.91% |
| Test | 91.64% | 98.08% |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support