Visual Information Retrieval and Reasoning on ViDoSeek
74.87Overall AccuracyVISOR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VISORBackbone=Qwen2.5-VL-7B, Multi-agent architecture=false, Fine-tuned=true2026.04 | 74.87 | 73.49 | 76.66 | |
| EVisRAGBackbone=Qwen2.5-VL-7B, Multi-agent architecture=false, Fine-tuned=true2026.04 | 69.79 | 67.75 | 72.43 | |
| M3RAGBackbone=Qwen2.5-VL-7B, Multi-agent architecture=true, Fine-tuned=false, External Source=Published results2026.04 | 69.36 | — | — | |
| ViDoRAGBackbone=Qwen2.5-VL-7B, Multi-agent architecture=true, Fine-tuned=false2026.04 | 69 | 66.05 | 72.83 | |
| VISORBackbone=Qwen2.5-VL-3B, Multi-agent architecture=false, Fine-tuned=true2026.04 | 69 | 67.75 | 70.62 | |
| EVisRAGBackbone=Qwen2.5-VL-3B, Multi-agent architecture=false, Fine-tuned=true2026.04 | 68.82 | 66.05 | 72.43 | |
| VRAG-RLBackbone=Qwen2.5-VL-3B, Multi-agent architecture=false, Fine-tuned=true, External Source=Published results2026.04 | 67.76 | 63.1 | 73.8 | |
| VRAG-RLBackbone=Qwen2.5-VL-7B, Multi-agent architecture=false, Fine-tuned=true, External Source=Published results2026.04 | 66.78 | 60.6 | 74.8 | |
| R1-RouterBackbone=Qwen2.5-VL-7B, Multi-agent architecture=false, Fine-tuned=true2026.04 | 66.11 | 64.19 | 68.61 | |
| R1-RouterBackbone=Qwen2.5-VL-3B, Multi-agent architecture=false, Fine-tuned=true2026.04 | 65.59 | 62.64 | 69.42 | |
| MMSearch-R1Backbone=Qwen2.5-VL-7B, Multi-agent architecture=false, Fine-tuned=true2026.04 | 57.53 | 55.97 | 59.56 | |
| Search-R1-VLBackbone=Qwen2.5-VL-7B, Multi-agent architecture=false, Fine-tuned=true, External Source=Published results2026.04 | 44.77 | 40.5 | 50.3 | |
| ViDoRAGBackbone=Qwen2.5-VL-3B, Multi-agent architecture=true, Fine-tuned=false2026.04 | 34.33 | 31.32 | 38.23 | |
| ReActBackbone=Qwen2.5-VL-7B, Multi-agent architecture=false, Fine-tuned=false, External Source=Published results2026.04 | 33.85 | 27.5 | 42.1 | |
| Vanilla RAGBackbone=Qwen2.5-VL-7B, Multi-agent architecture=false, Fine-tuned=false, External Source=Published results2026.04 | 32.88 | 26.4 | 41.3 | |
| Search-R1-VLBackbone=Qwen2.5-VL-3B, Multi-agent architecture=false, Fine-tuned=true, External Source=Published results2026.04 | 24.32 | 20.1 | 29.8 | |
| Vanilla RAGBackbone=Qwen2.5-VL-3B, Multi-agent architecture=false, Fine-tuned=false, External Source=Published results2026.04 | 13.23 | 10.1 | 17.3 | |
| ReActBackbone=Qwen2.5-VL-3B, Multi-agent architecture=false, Fine-tuned=false, External Source=Published results2026.04 | 9.96 | 6.7 | 14.2 |