Video Classification on Kinetics-700
92.1Top-1 AccuracyAbsolute SOTA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Absolute SOTA2024.06 | 92.1 | — | |
| MiCoBackbone=ViT-g2024.06 | 91.6 | — | |
| EVA-CLIP-18BZero-shot=true, #Frames=162024.02 | 79.8 | 72.2 | |
| EVA-CLIP-18BZero-shot=true, #Frames=82024.02 | 79.7 | 72.1 | |
| EVA-CLIP-18BZero-shot=true, #Frames=12024.02 | 75 | 68.2 | |
| PyraTokVAE=true2026.01 | 74.08 | — | |
| InternVL-CZero-shot=true, #Frames=12024.02 | 73.7 | 66.4 | |
| EVA-CLIP-8BZero-shot=true, #Frames=12024.02 | 73.6 | 66.1 | |
| EVA-02-CLIP-E/14+Zero-shot=true, #Frames=12024.02 | 72.1 | 64.4 | |
| En-VidTr-LInput=32 x 22021.04 | 70.8 | — | |
| SF101Input=64 x 22021.04 | 70.2 | — | |
| VidTr-LInput=32 x 22021.04 | 70.2 | — | |
| CSN152Input=32 x 22021.04 | 70.1 | — | |
| VidTr-MInput=16 x 42021.04 | 69.5 | — | |
| PEcoreGVAE=false2026.01 | 69.1 | — | |
| EVA-01-CLIP-g/14+Zero-shot=true, #Frames=12024.02 | 68.8 | 61.5 | |
| OpenCLIP-G/14Zero-shot=true, #Frames=12024.02 | 68.7 | 60.3 | |
| DFN5B-CLIP-H/14+Zero-shot=true, #Frames=12024.02 | 68.4 | 60.7 | |
| DFN5B-CLIP-H/14Zero-shot=true, #Frames=12024.02 | 67 | 59.2 | |
| LARPVAE=true2026.01 | 66.89 | — | |
| EVA-01-CLIP-g/14Zero-shot=true, #Frames=12024.02 | 66.2 | 58.8 | |
| InternVideo2VAE=false2026.01 | 64.9 | — | |
| SigLIP2-g-optVAE=false2026.01 | 61.8 | — | |
| SweetTokVAE=true2026.01 | 61.45 | — | |
| InternVL-CNumber of frames (#F)=8, Zero-shot=true2023.12 | 60.6 | 71.5 | |
| InternVLVAE=false2026.01 | 60.6 | — | |
| CLIP (reported)Pre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 60.3 | — | |
| CLIPPre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 59.5 | — | |
| OmniTokenizerVAE=true2026.01 | 58.71 | — | |
| I3DInput=32x22021.04 | 58.7 | — | |
| MLCDPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 57.7 | — | |
| InternVL-CNumber of frames (#F)=1, Zero-shot=true2023.12 | 56.8 | 67.5 | |
| xCLIPLinear probing=true, Backbone=ViT-B/16, Pre-train=IT35M2022.10 | 56 | — | |
| FLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 55.8 | — | |
| VideoVae+VAE=true2026.01 | 55.55 | — | |
| ViCLIPNumber of frames (#F)=8, Zero-shot=true2023.12 | 54.3 | 66.4 | |
| CLIPLinear probing=true, Backbone=ViT-B/16, Pre-train=IT35M2022.10 | 53.9 | — | |
| nCLIPLinear probing=true, Backbone=ViT-B/16, Pre-train=IT35M2022.10 | 51 | — | |
| OpenCLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 50.5 | — | |
| X3D-XLInput=16 x 52021.04 | 47.1 | — | |
| ImageBindBackbone=ViT-H2024.06 | 42.9 | — | |
| I3D101Input=32 x 42021.04 | 40.3 | — | |
| xCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 35.4 | — | |
| Meta-TransformerBackbone=ViT-L2024.06 | 33.2 | — | |
| CLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 32.4 | — | |
| nCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 28.4 | — | |
| EVA-01-CLIP-g+Number of frames (#F)=1, Zero-shot=true2023.12 | — | 60.9 | |
| EVA-02-CLIP-E+Number of frames (#F)=1, Zero-shot=true2023.12 | — | 63.4 | |
| OpenCLIP-gNumber of frames (#F)=1, Zero-shot=true2023.12 | — | 56.9 | |
| OpenCLIP-GNumber of frames (#F)=1, Zero-shot=true2023.12 | — | 59.2 |