Multi-modal Retrieval on M-BEIR (test)
69.7Average RecallV-Retrver-7B
Evaluation results
| Method | Links | |||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| V-Retrver-7BParameters=7B2026.02 | 69.7 | 51.8 | 87.5 | — | 90.2 | 82.9 | — | 52.2 | 94.8 | 37.8 | — | — | — | — | — | — | — | 40.3 | 96.9 | 39.8 | 69.8 | 73.2 | 51.2 | 73.5 | 87.8 | 85 | — | — | — | — | — | — | — | — | — | — | — | |
| U-MARVEL-7BParameters=7B2026.02 | 64.8 | 49.4 | 85.6 | — | 89.4 | 81.4 | — | 50.5 | 88.4 | 37.7 | — | — | — | — | — | — | — | 34.2 | 98.5 | 34.7 | 63.7 | 62.9 | 38.2 | 63.2 | 80.8 | 78.9 | — | — | — | — | — | — | — | — | — | — | — | |
| LamRAEvaluation Protocol=Supervised, Model Paradigm=LMMs, Reranking strategy=With Reranking2024.12 | 63.7 | 48 | 85.2 | 32.9 | 96.7 | 75.8 | 87.7 | 48.6 | 92.3 | 36.1 | 33.5 | 59.2 | 64.1 | 37.8 | 63.3 | 79.2 | 78.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LamRA-7BParameters=7B2026.02 | 63.7 | 48 | 85.2 | — | 87.7 | 75.8 | — | 48.6 | 92.3 | 36.1 | — | — | — | — | — | — | — | 32.9 | 96.7 | 33.5 | 59.2 | 64.1 | 37.8 | 63.3 | 79.2 | 78.3 | — | — | — | — | — | — | — | — | — | — | — | |
| TRACE2026.03 | 58.8 | 42.1 | 82.3 | 30.5 | — | — | — | — | — | — | — | — | — | 36.4 | 57.3 | 57.4 | 55.8 | — | 87.8 | — | — | — | — | — | — | 67.1 | — | — | — | — | 33.6 | — | 64.1 | 41.2 | 91.3 | 33.2 | 78.5 | |
| ELVA-7B (Ours)Supervision Type=Supervised, Model Architecture=MLLM2026.06 | 58.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LamRA-RetEvaluation Protocol=Supervised, Model Paradigm=LMMs, Reranking strategy=Retriever only2024.12 | 56.6 | 41.6 | 81.5 | 28.7 | 86 | 62.6 | 81.2 | 39.6 | 90.6 | 30.4 | 32.1 | 54.1 | 52.1 | 33.2 | 53.1 | 76.2 | 63.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MuCo-7B# Params=7B, Training=Fine-tuning2026.02 | 56.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 54.1 | 71.6 | 47.4 | 70.4 | — | — | — | — | — | — | — | |
| LamRA-Ret2026.03 | 56.6 | 41.6 | 81.5 | 28.7 | — | — | — | — | — | — | — | — | — | 33.2 | 53.1 | 54.1 | 52.1 | — | 86 | — | — | — | — | — | — | 63.3 | — | — | — | — | 32.1 | — | 62.6 | 39.6 | 90.6 | 30.4 | 76.2 | |
| LamRA-Ret-7BSupervision Type=Supervised, Model Architecture=MLLM2026.06 | 56.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LamRA-Ret# Params=7B, Training=Fine-tuning2026.02 | 54.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 52.9 | 71.8 | 45.2 | 65 | — | — | — | — | — | — | — | |
| PUMA-3BSupervision Type=Supervised, Model Architecture=MLLM2026.06 | 54.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| M3Task-UEM# Params=7B, Training=Fine-tuning2026.02 | 53.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 54 | 74.9 | 41.9 | 55.7 | — | — | — | — | — | — | — | |
| ELVA-2B (Ours)Supervision Type=Supervised, Model Architecture=MLLM2026.06 | 53.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MM-Embed-7BParameters=7B2026.02 | 52.7 | 41 | 71.3 | — | 85 | 68.8 | — | 41.3 | 90.1 | 18.4 | — | — | — | — | — | — | — | 17.1 | 95.9 | 32.4 | 42.1 | 42.3 | 25.7 | 50 | 64.1 | 57.7 | — | — | — | — | — | — | — | — | — | — | — | |
| MM-Embed# Params=7B, Training=Fine-tuning2026.02 | 52.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 50.9 | 76.9 | 40 | 60.9 | — | — | — | — | — | — | — | |
| MM-Embed-7BSupervision Type=Supervised, Model Architecture=MLLM2026.06 | 52.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| U-MARVEL+Evaluation Protocol=Zero-shot, Held-out Training Strategy=Trained on remaining 5 tasks, Reranker=true2025.07 | 52.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 49 | 55.3 | — | — | — | — | — | — | — | — | 61.2 | — | — | — | — | 32.1 | 64.9 | — | — | — | — | — | |
| MuCo-2B# Params=2B, Training=Fine-tuning2026.02 | 51.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 49.1 | 68.2 | 41.6 | 64.8 | — | — | — | — | — | — | — | |
| LamRA-Ret-2BSupervision Type=Supervised, Model Architecture=MLLM2026.06 | 51.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LamRAEvaluation Protocol=Zero-shot, Held-out Training Strategy=Trained on remaining 5 tasks, Reranker=true2025.07 | 50.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 46.9 | 54.2 | — | — | — | — | — | — | — | — | 59.1 | — | — | — | — | 29.2 | 65.1 | — | — | — | — | — | |
| UniIR-CLIP_SFEvaluation Protocol=Supervised, Model Paradigm=Dual Encoder2024.12 | 50.6 | 42.6 | 81.1 | 18 | 84.7 | 59.4 | 78.7 | 43.1 | 92.3 | 18.3 | 32 | 45.5 | 27.9 | 24.4 | 44.6 | 67.6 | 48.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniIR-CLIP_SFvariant=SF2026.02 | 50.6 | 42.6 | 81.1 | — | 78.7 | 59.4 | — | 43.1 | 92.3 | 18.3 | — | — | — | — | — | — | — | 18 | 84.7 | 32 | 45.5 | 27.9 | 24.4 | 44.6 | 67.6 | 48.9 | — | — | — | — | — | — | — | — | — | — | — | |
| UniIR-CLIPFusion Strategy=SF2026.03 | 50.6 | 42.6 | 81.1 | 18 | — | — | — | — | — | — | — | — | — | 24.4 | 44.6 | 45.5 | 27.9 | — | 84.7 | — | — | — | — | — | — | 48.9 | — | — | — | — | 32 | — | 59.4 | 43.1 | 92.3 | 18.3 | 67.6 | |
| UniIR-CLIP_SFSupervision Type=Supervised, Model Architecture=Dual Encoder2026.06 | 50.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Vision-R1-7BParameters=7B2026.02 | 50.2 | 41.9 | 75 | — | 69.1 | 51.3 | — | 35.4 | 85.1 | 22.4 | — | — | — | — | — | — | — | 22 | 70.6 | 25.9 | 48.8 | 44 | 29.2 | 57.7 | 66.2 | 59 | — | — | — | — | — | — | — | — | — | — | — | |
| Vision-R1-7BSupervision Type=Supervised, Model Architecture=MLLM2026.06 | 50.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VLM-R1-7BParameters=7B2026.02 | 50 | 40.5 | 77.2 | — | 67.9 | 50 | — | 36.2 | 86.3 | 20.9 | — | — | — | — | — | — | — | 22.5 | 72.3 | 26.4 | 48.8 | 37.5 | 29.9 | 57.4 | 64 | 62.3 | — | — | — | — | — | — | — | — | — | — | — | |
| LamRA-Ret# Params=2B, Training=Fine-tuning2026.02 | 50 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 47.6 | 66.2 | 41 | 61 | — | — | — | — | — | — | — | |
| VLM-R1-7BSupervision Type=Supervised, Model Architecture=MLLM2026.06 | 50 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniIR# Params=0.4B, Training=Fine-tuning2026.02 | 48.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 51 | 69.1 | 32.9 | 52.4 | — | — | — | — | — | — | — | |
| U-MARVELEvaluation Protocol=Zero-shot, Held-out Training Strategy=Trained on remaining 5 tasks, Reranker=false2025.07 | 48 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 44.3 | 49.2 | — | — | — | — | — | — | — | — | 52.6 | — | — | — | — | 31.5 | 62.6 | — | — | — | — | — | |
| UniIR-BLIP_FFEvaluation Protocol=Supervised, Model Paradigm=Dual Encoder2024.12 | 46.8 | 23.4 | 79.7 | 26.1 | 80 | 50.9 | 79.8 | 22.8 | 89.9 | 28.9 | 33 | 41 | 22.4 | 29.2 | 52.2 | 55.8 | 33 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniIR-BLIP_FFvariant=FF2026.02 | 46.8 | 23.4 | 79.7 | — | 79.8 | 50.9 | — | 22.8 | 89.9 | 28.9 | — | — | — | — | — | — | — | 26.1 | 80 | 33 | 41 | 22.4 | 29.2 | 52.2 | 55.8 | 33 | — | — | — | — | — | — | — | — | — | — | — | |
| UniIR-BLIPFusion Strategy=FF2026.03 | 46.8 | 23.4 | 79.7 | 26.1 | — | — | — | — | — | — | — | — | — | 29.2 | 52.2 | 41 | 22.4 | — | 80 | — | — | — | — | — | — | 33 | — | — | — | — | 33 | — | 50.9 | 22.8 | 89.9 | 28.9 | 55.8 | |
| UniIR-BLIP_FFSupervision Type=Supervised, Model Architecture=Dual Encoder2026.06 | 46.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7BParameters=7B2026.02 | 46.7 | 40.2 | 71.9 | — | 64.5 | 49.4 | — | 29.3 | 84.6 | 19.4 | — | — | — | — | — | — | — | 20.3 | 71.9 | 25.5 | 42.4 | 32.1 | 25 | 55.1 | 60.8 | 54.9 | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7BSupervision Type=Zero-shot2026.06 | 46.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LamRA-RetEvaluation Protocol=Zero-shot, Held-out Training Strategy=Trained on remaining 5 tasks, Reranker=false2025.07 | 45.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 44.7 | 44 | — | — | — | — | — | — | — | — | 49.5 | — | — | — | — | 27.2 | 62.8 | — | — | — | — | — | |
| UniIR-CLIPEvaluation Protocol=Supervised2025.07 | 44.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 45.5 | 27.9 | — | — | — | — | — | — | — | — | 48.9 | — | — | — | — | 32 | 67.6 | — | — | — | — | — | |
| Qwen2.5-VL-3BParameters=3B2026.02 | 42.4 | 36 | 67.8 | — | 61.7 | 45.2 | — | 23.3 | 82.3 | 12 | — | — | — | — | — | — | — | 16.1 | 69.5 | 20.9 | 36.7 | 22.3 | 24.3 | 53.5 | 56.4 | 49.8 | — | — | — | — | — | — | — | — | — | — | — | |
| SigLIPEvaluation Protocol=Zero-shot2024.12 | 37.2 | 30.1 | 75.7 | 36.5 | 39.8 | 27 | 43.5 | 30.8 | 88.2 | 34.2 | 28.9 | 29.7 | 25.1 | 14.4 | 22.7 | 41.7 | 27.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SigLIP2026.02 | 37.2 | 30.1 | 75.7 | — | 43.5 | 27 | — | 30.8 | 88.2 | 34.2 | — | — | — | — | — | — | — | 36.5 | 39.8 | 28.9 | 29.7 | 25.1 | 14.4 | 22.7 | 41.7 | 27.4 | — | — | — | — | — | — | — | — | — | — | — | |
| SigLIP2026.03 | 37.2 | 30.1 | 75.7 | 36.5 | — | — | — | — | — | — | — | — | — | 14.4 | 22.7 | 29.7 | 25.1 | — | 39.8 | — | — | — | — | — | — | 27.4 | — | — | — | — | 28.9 | — | 27 | 30.8 | 88.2 | 34.2 | 41.7 | |
| SigLIPSupervision Type=Zero-shot2026.06 | 37.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniIR-BLIPEvaluation Protocol=Supervised2025.07 | 37 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 41 | 22.4 | — | — | — | — | — | — | — | — | 33 | — | — | — | — | 33 | 55.8 | — | — | — | — | — | |
| CLIP-LEvaluation Protocol=Zero-shot2024.12 | 32.5 | 43.3 | 61.1 | 6.6 | 36.2 | 43.3 | 45.1 | 41.3 | 79 | 7.7 | 26.1 | 24.2 | 20.5 | 7 | 13.2 | 38.8 | 26.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIP-LBackbone=CLIP-L2026.02 | 32.5 | 43.3 | 61.1 | — | 45.1 | 43.3 | — | 41.3 | 79 | 7.7 | — | — | — | — | — | — | — | 6.6 | 36.2 | 26.1 | 24.2 | 20.5 | 7 | 13.2 | 38.8 | 26.4 | — | — | — | — | — | — | — | — | — | — | — | |
| CLIP-LBackbone=CLIP-L2026.03 | 32.5 | 43.3 | 61.1 | 6.6 | — | — | — | — | — | — | — | — | — | 7 | 13.2 | 24.2 | 20.5 | — | 36.2 | — | — | — | — | — | — | 26.4 | — | — | — | — | 26.1 | — | 43.3 | 41.3 | 79 | 7.7 | 38.8 | |
| CLIP-LSupervision Type=Zero-shot2026.06 | 32.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BLIPEvaluation Protocol=Zero-shot2024.12 | 26.8 | 16.4 | 74.4 | 15.9 | 44.9 | 26.8 | 20.3 | 17.2 | 83.2 | 19.9 | 27.4 | 16.1 | 10.2 | 2.3 | 10.6 | 27.4 | 16.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BLIP2026.02 | 26.8 | 16.4 | 74.4 | — | 20.3 | 26.8 | — | 17.2 | 83.2 | 19.9 | — | — | — | — | — | — | — | 15.9 | 44.9 | 27.4 | 16.1 | 10.2 | 2.3 | 10.6 | 27.4 | 16.6 | — | — | — | — | — | — | — | — | — | — | — | |
| BLIP2026.03 | 26.8 | 16.4 | 74.4 | 15.9 | — | — | — | — | — | — | — | — | — | 2.3 | 10.6 | 16.1 | 10.2 | — | 44.9 | — | — | — | — | — | — | 16.6 | — | — | — | — | 27.4 | — | 26.8 | 17.2 | 83.2 | 19.9 | 27.4 | |
| BLIPSupervision Type=Zero-shot2026.06 | 26.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BLIP2Evaluation Protocol=Zero-shot2024.12 | 24.8 | 16.7 | 63.8 | 14 | 38.6 | 26.9 | 24.5 | 15 | 80 | 14.2 | 25.4 | 12.2 | 5.5 | 4.4 | 11.8 | 27.3 | 15.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BLIP22026.02 | 24.8 | 16.7 | 63.8 | — | 24.5 | 26.9 | — | 15 | 80 | 14.2 | — | — | — | — | — | — | — | 14 | 38.6 | 25.4 | 12.2 | 5.5 | 4.4 | 11.8 | 27.3 | 15.8 | — | — | — | — | — | — | — | — | — | — | — | |
| BLIP22026.03 | 24.8 | 16.7 | 63.8 | 14 | — | — | — | — | — | — | — | — | — | 4.4 | 11.8 | 12.2 | 5.5 | — | 38.6 | — | — | — | — | — | — | 15.8 | — | — | — | — | 25.4 | — | 26.9 | 15 | 80 | 14.2 | 27.3 | |
| BLIP2Supervision Type=Zero-shot2026.06 | 24.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2-VL-7BEvaluation Protocol=Zero-shot2024.12 | 23 | 9.3 | 55.1 | 5 | 42 | 26.2 | 9.4 | 5.4 | 46.6 | 4 | 21.3 | 21.4 | 22.5 | 4.3 | 16.3 | 43.6 | 36.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7BParameters=7B2026.03 | 23 | 9.3 | 55.1 | 5 | — | — | — | — | — | — | — | — | — | 4.3 | 16.3 | 21.4 | 22.5 | — | 42 | — | — | — | — | — | — | 36.2 | — | — | — | — | 21.3 | — | 26.2 | 5.4 | 46.6 | 4 | 43.6 | |
| Qwen2-VL-7BSupervision Type=Zero-shot2026.06 | 23 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |