Video Classification on UCF101 (Top-1/Top-5 Accuracy)
98.24Top-1 AccuracyX-Ray Visual
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| X-Ray VisualEncoder=EViT-2b, Protocol=Linear probing2026.02 | 98.24 | 99.86 | |
| PE-GEncoder=PE-G, Protocol=Linear probing2026.02 | 96.99 | 99.89 | |
| DiNO-v3Encoder=ViT-H, Protocol=Linear probing2026.02 | 94.3 | 99.52 | |
| VideoCoCa (w.o. videos)protocol=zero-shot, continued pretraining=false, video adaptor=plugin attentional poolers2022.12 | 86.6 | 98.5 | |
| VideoCoCaprotocol=zero-shot, continued pretraining=true, video adaptor=plugin attentional poolers2022.12 | 86.6 | 98.4 | |
| CoCaprotocol=zero-shot, pooling=frame-level average pooling2022.12 | 85.1 | 97.5 | |
| MOV-Lprotocol=zero-shot2022.12 | 80.9 | — | |
| X-Florenceprotocol=zero-shot2022.12 | 73.2 | — | |
| X-CLIPprotocol=zero-shot2022.12 | 72 | — | |
| RESTprotocol=zero-shot2022.12 | 69.1 | — | |
| ActionCLIPprotocol=zero-shot2022.12 | 58.3 | — | |
| SiGLIP 2Encoder=ViT-g/16, Protocol=Linear probing2026.02 | 55.31 | 68.93 | |
| ER-ZSARprotocol=zero-shot2022.12 | 51.8 | — |