Image Retrieval on COCO (test)
60Recall@1UNCHA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| UNCHABackbone=ViT-B/16, Evaluation protocol=Zero-shot2026.03 | 60 | 71 | — | |
| HyCoCLIPBackbone=ViT-B/16, Evaluation protocol=Zero-shot2026.03 | 58.4 | 69.3 | — | |
| CLIPBackbone=ViT-B/16, Evaluation protocol=Zero-shot2026.03 | 57.4 | 68.5 | — | |
| MERUBackbone=ViT-B/16, Evaluation protocol=Zero-shot2026.03 | 57.4 | 68.6 | — | |
| UNCHABackbone=ViT-S/16, Evaluation protocol=Zero-shot2026.03 | 56.2 | 67.6 | — | |
| HyCoCLIPBackbone=ViT-S/16, Evaluation protocol=Zero-shot2026.03 | 55.2 | 66.6 | — | |
| CLIPBackbone=ViT-S/16, Evaluation protocol=Zero-shot2026.03 | 53.7 | 65.2 | — | |
| MERUBackbone=ViT-S/16, Evaluation protocol=Zero-shot2026.03 | 53.6 | 65.3 | — | |
| ATMG†Backbone=ViT-B/16, Evaluation protocol=Zero-shot2026.03 | 51.2 | 62.6 | — | |
| ATMG†Backbone=ViT-S/16, Evaluation protocol=Zero-shot2026.03 | 50.3 | 62.1 | — | |
| PTM-STPairs=500, Ratio=4.4‰2026.03 | 6.6 | 20.5 | 30.7 | |
| PTM-STPairs=200, Ratio=1.7‰2026.03 | 3.9 | 13.7 | 22.2 | |
| LoRSPairs=500, Ratio=4.4‰2026.03 | 3.3 | 11.8 | 19.2 | |
| MTT-VLPairs=500, Ratio=4.4‰2026.03 | 2.5 | 8.9 | 15.8 | |
| PTM-STPairs=100, Ratio=0.8‰2026.03 | 2.3 | 9 | 15.6 | |
| LoRSPairs=200, Ratio=1.7‰2026.03 | 2.2 | 8.6 | 14.7 | |
| DATMPairs=500, Ratio=4.4‰2026.03 | 2.2 | 8.1 | 14.3 | |
| LoRSPairs=100, Ratio=0.8‰2026.03 | 1.7 | 6.9 | 11.9 | |
| MTT-VLPairs=200, Ratio=1.7‰2026.03 | 1.7 | 6.5 | 12.3 | |
| DATMPairs=200, Ratio=1.7‰2026.03 | 1.6 | 6.3 | 11.8 | |
| MTT-VLPairs=100, Ratio=0.8‰2026.03 | 1.3 | 5.4 | 9.5 | |
| DATMPairs=100, Ratio=0.8‰2026.03 | 1.2 | 5.1 | 8.9 |