Video Classification on Kinetics 700 (test)
89Top-1 AccSotA (Fine-tuned)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SotA (Fine-tuned)shots/class=full2022.04 | 89 | — | — | |
| SotA (Contrastive)Method=Contrastive, shots/class=02022.04 | 69.6 | — | — | |
| Flamingo-80BMethod=RICES+ensembling, Prompt size=16, shots/class=52022.04 | 64.2 | — | — | |
| Flamingo-80BMethod=RICES, Prompt size=16, shots/class=52022.04 | 63.5 | — | — | |
| EVA-02-CLIPZero-shot=true, Backbone=ViT-E/14+, Training Scale=Enlarged2023.03 | 63.4 | — | — | |
| NFNet-F6Method=Our contrastive, shots/class=02022.04 | 62.9 | — | — | |
| Flamingo-80BMethod=RICES, Prompt size=16, shots/class=12022.04 | 62.7 | — | — | |
| EVA-02-CLIPZero-shot=true, Backbone=ViT-E/14, Training Scale=Enlarged2023.03 | 62.5 | — | — | |
| VideoCoCaZero-shot=true, Evaluation protocol=single view, Continued pretraining=true2022.12 | 62.5 | 83.8 | 73.2 | |
| VideoCoCa (w.o. videos)Zero-shot=true, Evaluation protocol=single view, Continued pretraining=false2022.12 | 62.2 | 83.7 | 73 | |
| Flamingo-9BMethod=RICES, Prompt size=16, shots/class=52022.04 | 60.9 | — | — | |
| EVA-01-CLIPZero-shot=true, Backbone=ViT-g/14+, Training Scale=Enlarged2023.03 | 60.9 | — | — | |
| Flamingo-80BMethod=RICES, Prompt size=8, shots/class=12022.04 | 60.4 | — | — | |
| EVA-02-CLIPZero-shot=true, Backbone=ViT-L/14+, Training Scale=Enlarged2023.03 | 60.2 | — | — | |
| CoCaZero-shot=true, Evaluation protocol=single view, Pooling=frame-level average2022.12 | 60.2 | 82.2 | 71.2 | |
| Flamingo-9BMethod=RICES, Prompt size=16, shots/class=12022.04 | 59.4 | — | — | |
| Open CLIPZero-shot=true, Backbone=ViT-G/14, Training Scale=Enlarged2023.03 | 59.2 | — | — | |
| EVA-02-CLIPZero-shot=true, Backbone=ViT-L/14, Training Scale=Standard2023.03 | 59.1 | — | — | |
| OpenAI CLIPZero-shot=true, Backbone=ViT-L/14+, Training Scale=Enlarged2023.03 | 58.5 | — | — | |
| EVA-01-CLIPZero-shot=true, Backbone=ViT-g/14, Training Scale=Enlarged2023.03 | 58.4 | — | — | |
| Flamingo-3BMethod=RICES, Prompt size=16, shots/class=52022.04 | 58.3 | — | — | |
| Flamingo-9BMethod=RICES, Prompt size=8, shots/class=12022.04 | 58 | — | — | |
| OpenAI CLIPZero-shot=true, Backbone=ViT-L/14, Training Scale=Standard2023.03 | 57.7 | — | — | |
| Flamingo-3BMethod=RICES, Prompt size=16, shots/class=12022.04 | 56.9 | — | — | |
| Open CLIPZero-shot=true, Backbone=ViT-g/14, Training Scale=Enlarged2023.03 | 56.9 | — | — | |
| Open CLIPZero-shot=true, Backbone=ViT-H/14, Training Scale=Enlarged2023.03 | 56.1 | — | — | |
| Flamingo-3BMethod=RICES, Prompt size=8, shots/class=12022.04 | 55.9 | — | — | |
| Flamingo-80BMethod=Random, Prompt size=16, shots/class=< 0.022022.04 | 51.2 | — | — | |
| Open CLIPZero-shot=true, Backbone=ViT-L/14, Training Scale=Standard2023.03 | 50.8 | — | — | |
| EVA-02-CLIPZero-shot=true, Backbone=ViT-B/16, Training Scale=Standard2023.03 | 50 | — | — | |
| OpenAI CLIPZero-shot=true, Backbone=ViT-B/16, Training Scale=Standard2023.03 | 49.3 | — | — | |
| Open CLIPZero-shot=true, Backbone=ViT-B/16, Training Scale=Standard2023.03 | 46.8 | — | — | |
| CLIP-LZero-shot=true, Evaluation protocol=single view2022.12 | — | — | 69.6 | |
| NFNet-F6Zero-shot=true, Evaluation protocol=single view2022.12 | — | — | 62.9 |