Composed Image Retrieval on CIRR, FashionIQ, GeneCIS
38.3Average PerformanceBGE-large
Evaluation results
| Method | Links | |
|---|---|---|
| BGE-largeModel Category=CLIP-based Models, Evaluation Protocol=Pretrained2026.03 | 38.3 | |
| RzenEmbed-7BModel Category=VLM-based Models, Evaluation Protocol=Pretrained2026.03 | 37.6 | |
| BGE-baseModel Category=CLIP-based Models, Evaluation Protocol=Pretrained2026.03 | 36.5 | |
| GME-7BModel Category=VLM-based Models, Evaluation Protocol=Pretrained2026.03 | 36.1 | |
| GME-2BModel Category=VLM-based Models, Evaluation Protocol=Pretrained2026.03 | 33.8 | |
| MM-Embed-7BModel Category=VLM-based Models, Evaluation Protocol=Pretrained2026.03 | 27.2 | |
| CLIP4CIR-RN50x4Model Category=CLIP-based Models, Evaluation Protocol=Pretrained2026.03 | 26 | |
| CLIP4CIR-RN50Model Category=CLIP-based Models, Evaluation Protocol=Pretrained2026.03 | 24.9 | |
| SEARLE-largeModel Category=CLIP-based Models, Evaluation Protocol=Pretrained2026.03 | 24.7 | |
| SEARLE-baseModel Category=CLIP-based Models, Evaluation Protocol=Pretrained2026.03 | 23.6 |