Video-Text Retrieval on InstrAct-Semantic
28.1R@1InstrAction
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| InstrActionHard Negative Strategy=Order-Swapped HNs2026.04 | 28.1 | 78 | 3 | |
| CLIP-ViPDomain=Out-of-domain (OOD), Evaluation Protocol=Fine-tuned, Training Subset=Filtered Cooking Subset (175690 Clips)2026.04 | 18.4 | 63.2 | 4 | |
| VideoPrismDomain=In-domain (ID), Evaluation Protocol=Zero-shot2026.04 | 18.2 | 74.4 | 4 | |
| PerceptionLMDomain=In-domain (ID), Evaluation Protocol=Zero-shot2026.04 | 16.5 | 63.2 | 4 | |
| CLIP4ClipDomain=Out-of-domain (OOD), Evaluation Protocol=Fine-tuned, Training Subset=Filtered Cooking Subset (175690 Clips)2026.04 | 14.8 | 58 | 5 | |
| ViCLIPDomain=In-domain (ID), Evaluation Protocol=Zero-shot2026.04 | 14.7 | 61.2 | 4 | |
| UNiVLDomain=In-domain (ID), Evaluation Protocol=Zero-shot2026.04 | 13.2 | 60.3 | 5 | |
| MIL-NCEDomain=In-domain (ID), Evaluation Protocol=Zero-shot2026.04 | 7.2 | 49.4 | 6 |