Visual Attribute Classification on MAAC-Bench
76.2Top-1 Accuracy (Macro)ReVisIT
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ReVisITBackbone=GPT-4.1, Retrieval strategy=cosine, Retrieval embedder=Qwen3-VL-Embedding-8B, Number of shots=52026.07 | 76.2 | 69.8 | 67.7 | 83.3 | 81.2 | 79.2 | |
| ReVisITBackbone=GPT-4.1, Retrieval strategy=random sampling, Number of shots=52026.07 | 72.8 | 58.3 | 64.6 | 83.3 | 83.5 | 74 | |
| GPT-4.1Setup=vanilla2026.07 | 72.1 | 57.3 | 64.6 | 84.4 | 77.6 | 76.6 | |
| ReVisITBackbone=Doubao-1.5, Retrieval strategy=cosine, Retrieval embedder=Qwen3-VL-Embedding-8B, Number of shots=52026.07 | 70.6 | 47.9 | 75 | 82.3 | 76.5 | 71.4 | |
| ReVisITBackbone=Qwen3-VL-30B, Access=open, Retrieval strategy=cosine, Retrieval embedder=Qwen3-VL-Embedding-8B, Number of shots=52026.07 | 68.9 | 50 | 63.5 | 84.4 | 76.5 | 70.3 | |
| Doubao-1.5-vision-pro-32kSetup=vanilla2026.07 | 65.2 | 40.6 | 60.4 | 79.2 | 71.8 | 74 | |
| Qwen3-VL-30B-A3BSetup=vanilla, Access=open2026.07 | 63.5 | 42.7 | 61.5 | 75 | 69.4 | 68.8 |