Action Recognition on Kinetics zero-shot 600 (val)
83.1Top-1 AccTC-CLIP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TC-CLIPBackbone=ViT-L/14, Word Embeddings (WE)=true, Evaluation protocol=Zero-shot, Training dataset=Kinetics-4002024.04 | 83.1 | 97.3 | |
| ViFi-CLIP (reproduced)Backbone=ViT-L/14, Word Embeddings (WE)=true, Evaluation protocol=Zero-shot, Training dataset=Kinetics-4002024.04 | 81.1 | 96.7 | |
| Open-VCLIPBackbone=ViT-L/14, Word Embeddings (WE)=true, Evaluation protocol=Zero-shot, Training dataset=Kinetics-4002024.04 | 81.1 | 96.3 | |
| TC-CLIPBackbone=ViT-L/14, Word Embeddings (WE)=false, Evaluation protocol=Zero-shot, Training dataset=Kinetics-4002024.04 | 80.1 | 96.5 | |
| ViFi-CLIPBackbone=ViT-L/14, Word Embeddings (WE)=false, Evaluation protocol=Zero-shot, Training dataset=Kinetics-4002024.04 | 77.8 | 95.6 |