Composed Retrieval on Bookset Multi-Text to Image
52.7Recall@1HOMIE (Qwen3-VL-8B)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HOMIE (Qwen3-VL-8B)Model Category=MLLM-Based2025.02 | 52.7 | 79.5 | 88.5 | |
| HOMIE (Qwen3-VL-2B)Model Category=MLLM-Based2025.02 | 49.8 | 76 | 85.9 | |
| MUSKFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 47.9 | 74.2 | 84.3 | |
| CONCHFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 43.7 | 70.8 | 81.1 | |
| Patho-CLIP-LFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 40.7 | 69.7 | 78.8 | |
| Pathgen-CLIP-LFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 39.8 | 70.8 | 80.7 | |
| Pathgen-CLIPFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 39.2 | 66.5 | 78.9 | |
| Patho-CLIPFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 35.6 | 61.8 | 73.6 | |
| CONCHFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 34.4 | 63.9 | 75.6 | |
| MUSKFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 34 | 68.9 | 82.2 | |
| Patho-CLIP-LFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 33.6 | 62.3 | 77 | |
| Patho-CLIPFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 30.1 | 58.2 | 71.1 | |
| Pathgen-CLIPFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 28.3 | 59.8 | 71.5 | |
| Pathgen-CLIP-LFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 27.7 | 58.2 | 74.6 | |
| PLIPFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 26.8 | 49.6 | 59.8 | |
| QuiltNetFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 26.6 | 45.5 | 52.6 | |
| PathCLIPFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 22.4 | 44.6 | 57.9 | |
| PathCLIPFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 19.3 | 39.8 | 56.4 | |
| PathoR1-7BModel Category=MLLM-Based2025.02 | 18.2 | 43.2 | 52.3 | |
| BiomedCLIPFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 17.8 | 39.9 | 52.6 | |
| BiomedCLIPFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 15.8 | 37.5 | 52.5 | |
| QuiltNetFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 15.4 | 34.6 | 46.3 | |
| PLIPFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 12.1 | 35.2 | 53.9 | |
| Lingshu-7BModel Category=MLLM-Based2025.02 | 10.2 | 24.8 | 35 | |
| PubmedCLIPFusion Strategy=Add, Model Category=Dual-Encoder2025.02 | 7.7 | 17.3 | 23.9 | |
| LamRA-7BModel Category=MLLM-Based2025.02 | 7.4 | 20.5 | 29.3 | |
| HuatuoGPT-V-7BModel Category=MLLM-Based2025.02 | 6.1 | 18.9 | 25.8 | |
| Qwen2.5-VL-7BModel Category=MLLM-Based2025.02 | 3.9 | 12.7 | 23.2 | |
| GME-7BModel Category=MLLM-Based2025.02 | 3.7 | 14.5 | 22.1 | |
| Qwen3-VL-8BModel Category=MLLM-Based2025.02 | 3.3 | 10.9 | 15.6 | |
| PubmedCLIPFusion Strategy=RRF, Model Category=Dual-Encoder2025.02 | 2.5 | 9.8 | 16.4 | |
| E5-V-7BModel Category=MLLM-Based2025.02 | 2 | 6.4 | 10.9 | |
| Qwen3-VL-2BModel Category=MLLM-Based2025.02 | 1.8 | 5.5 | 8.4 |