Video Action Classification on COIN
95.3Top-1 AccPEcoreG
Evaluation Results
| Method | Links | |
|---|---|---|
| PEcoreGParameters=1.9B, Evaluation Protocol=Attentive probe2025.06 | 95.3 | |
| SigLIP2Parameters=1.2B, Evaluation Protocol=Attentive probe2025.06 | 95.1 | |
| InternVideo2_s2-1BParameters=1B, Evaluation Protocol=Attentive probe2025.06 | 93.8 | |
| HERMES2024.08 | 93.5 | |
| MA-LMM2024.08 | 93.2 | |
| MA-LMM2024.04 | 93.2 | |
| V-JEPA 2 ViT-g384Parameters=1B, Input format=32x8x3 frames, Input Resolution=384x3842025.06 | 91.1 | |
| S52024.08 | 90.8 | |
| S52024.04 | 90.8 | |
| DINOv2Parameters=1.1B, Evaluation Protocol=Attentive probe2025.06 | 90.7 | |
| V-JEPA 2 ViT-gParameters=1B, Input format=32x8x3 frames2025.06 | 90.7 | |
| D-Sprv.2024.04 | 90 | |
| TranS4mer2024.08 | 89.2 | |
| VIS4mer2024.08 | 88.4 | |
| ViS4mer2024.04 | 88.4 | |
| V-JEPA 2 ViT-HParameters=600M, Input format=32x8x3 frames2025.06 | 87.9 | |
| V-JEPA ViT-HParameters=600M, Evaluation Protocol=Attentive probe2025.06 | 87.1 | |
| V-JEPA 2 ViT-LParameters=300M, Input format=32x8x3 frames2025.06 | 86.8 | |
| TSN2024.04 | 73.4 | |
| ProcedureVRLPretraining Supervision=Unsupervised: ASR, Pretraining Dataset=HT100M, Protocol=Fine-tuning2023.03 | 56.9 | |
| DistantSupPretraining Supervision=Unsupervised: ASR + wikiHow, Pretraining Dataset=HT100M, Protocol=Fine-tuning2023.03 | 54.1 | |
| TimeSformerPretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics, Protocol=Fine-tuning2023.03 | 48.3 | |
| DistantSup†Pretraining Supervision=Unsupervised: ASR + wikiHow, Pretraining Dataset=HT100M, Protocol=Fine-tuning2023.03 | 46.6 | |
| TimeSformerPretraining Supervision=Unsupervised: ASR w. MIL-NCE [39], Pretraining Dataset=HT100M, Protocol=Fine-tuning2023.03 | 46.5 | |
| CLIPPretraining Supervision=Unsupervised: captions, Pretraining Dataset=CLIP400M, Protocol=Fine-tuning2023.03 | 45.9 | |
| VideoCLIPPretraining Supervision=Unsupervised: ASR, Pretraining Dataset=HT100M, Protocol=Fine-tuning2023.03 | 39.4 | |
| S3DPretraining Supervision=Unsupervised: ASR w. MIL-NCE [39], Pretraining Dataset=HT100M, Protocol=Fine-tuning2023.03 | 37.5 | |
| TSN (RGB+Flow)Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics, Protocol=Fine-tuning2023.03 | 36.5 | |
| SlowFastPretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics, Protocol=Fine-tuning2023.03 | 32.9 | |
| ClipBERTPretraining Supervision=Supervised: captions, Pretraining Dataset=COCO+VG, Protocol=Fine-tuning2023.03 | 30.8 | |
| ProcedureVRLPretraining Supervision=Unsupervised: ASR, Pretraining Dataset=HT100M, Protocol=Zero-shot2023.03 | 16.6 | |
| CLIPPretraining Supervision=Unsupervised: captions, Pretraining Dataset=CLIP400M, Protocol=Zero-shot2023.03 | 14.8 | |
| DistantSup†Pretraining Supervision=Unsupervised: ASR + wikiHow, Pretraining Dataset=HT100M, Protocol=Zero-shot2023.03 | 10.2 |