Composed Video Retrieval on CoVR-R (test)
89.73Recall@1Reason-Then-Retrieve
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Reason-Then-RetrieveStage=Test2026.06 | 89.73 | 95.79 | 96.63 | 97.98 | — | |
| Our Approach+RUsing Video Captions=✘, Modification Text Fusion=SA, Backbone=Qwen-VL, Frames=FrameR=1, Zero-shot=true2026.03 | 49.88 | 66.99 | 72.97 | 85.14 | 8.31 | |
| Our ApproachUsing Video Captions=✘, Modification Text Fusion=SA, Backbone=Qwen-VL, Frames=FrameR=1, Zero-shot=true2026.03 | 44.32 | 61.91 | 67.33 | 79.9 | 7.46 | |
| BSE-CoVRUsing Video Captions=✔, Modification Text Fusion=CA, Backbone=BLIP, Frames=15, Zero-shot=true2026.03 | 37.9 | 57.67 | 64.48 | 79.47 | 6.42 | |
| MVFT-JI†Using Video Captions=✔, Modification Text Fusion=CA, Backbone=BLIP, Frames=15, Zero-shot=true2026.03 | 34.4 | 54.15 | 62.3 | 77.4 | 6.28 | |
| BSE-CoVRUsing Video Captions=✔, Modification Text Fusion=Avg, Backbone=BLIP, Frames=15, Zero-shot=true2026.03 | 33.43 | 53.88 | 61.31 | 77.32 | 6.05 | |
| Thawakar et al.Using Video Captions=✔, Modification Text Fusion=Avg, Backbone=BLIP, Frames=15, Zero-shot=true2026.03 | 31.71 | 52.18 | 59.53 | 74.67 | 5.62 | |
| Thawakar et al.Using Video Captions=✔, Modification Text Fusion=CA, Backbone=BLIP, Frames=15, Zero-shot=true2026.03 | 30.9 | 50.63 | 60.11 | 75.33 | 5.88 | |
| CoVR-BLIPUsing Video Captions=✘, Modification Text Fusion=Avg, Backbone=BLIP, Frames=15, Zero-shot=true2026.03 | 30.3 | 51.07 | 57.05 | 73.82 | 4.85 | |
| MVFT-JI†Using Video Captions=✘, Modification Text Fusion=CA, Backbone=BLIP, Frames=15, Zero-shot=true2026.03 | 29.48 | 49.22 | 58.02 | 74.55 | 6 | |
| CoVR-BLIPUsing Video Captions=✘, Modification Text Fusion=CA, Backbone=BLIP, Frames=15, Zero-shot=true2026.03 | 28.06 | 47.8 | 55.93 | 73.76 | 5.1 |