Visual Grounding on V* (Attribute, Spatial, Average Accuracy)
91.1Average AccuracyLazyMCoT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LazyMCoTBackbone=Qwen3-VL-8B-Instruct, Training Free=true2026.06 | 91.1 | 92.2 | 89.5 | |
| DeepScanBackbone=Qwen2.5-VL-7B-Instruct, Training Free=true2026.06 | 90.6 | 93 | 86.8 | |
| LazyMCoTBackbone=Qwen2.5-VL-7B-Instruct, Training Free=true2026.06 | 90.6 | 92.2 | 88.2 | |
| DeepEyesBackbone=Qwen2.5-VL-7B-Instruct, Training Free=false2026.06 | 90 | 92.1 | 86.8 | |
| HiDeBackbone=Qwen2.5-VL-7B-Instruct, Training Free=true2026.06 | 90 | 94.8 | 82.9 | |
| LazyMCoTBackbone=InternVL3-8B-Instruct, Training Free=true2026.06 | 87.3 | 89.7 | 83.9 | |
| HiDeBackbone=InternVL3-8B-Instruct, Training Free=true2026.06 | 86.4 | 89.6 | 81.6 | |
| TreeVGRBackbone=Qwen2.5-VL-7B-Instruct, Training Free=false2026.06 | 85.9 | 86.1 | 85.5 | |
| DyfoBackbone=Qwen2.5-VL-7B-Instruct, Training Free=true2026.06 | 84.3 | 82.6 | 86.8 | |
| ViCropBackbone=InternVL3-8B-Instruct, Training Free=true2026.06 | 83.3 | 88.7 | 75 | |
| Base (Qwen3-VL-8B-Instruct)Backbone=Qwen3-VL-8B-Instruct, Training Free=false2026.06 | 83.2 | 85.2 | 80.3 | |
| Thyme-VLBackbone=Qwen2.5-VL-7B-Instruct, Training Free=false2026.06 | 82.2 | 83.5 | 80.3 | |
| ViCropBackbone=Qwen2.5-VL-7B-Instruct, Training Free=true2026.06 | 82.2 | 89.6 | 71.1 | |
| ZoomRefineBackbone=Qwen2.5-VL-7B-Instruct, Training Free=true2026.06 | 82.2 | 85.3 | 77.6 | |
| Base (InternVL3-8B-Instruct)Backbone=InternVL3-8B-Instruct, Training Free=false2026.06 | 80.6 | 81.7 | 78.9 | |
| Base (Qwen2.5-VL-7B-Instruct)Backbone=Qwen2.5-VL-7B-Instruct, Training Free=false2026.06 | 79.1 | 80.9 | 76.3 | |
| GPT-4oTraining Free=true2026.06 | 66 | — | — |