Visual Grounding on RefCOCO UMD (val)
93AccuracyBase Model
Evaluation results
| Method | Links | |
|---|---|---|
| Base ModelBackbone=InternVL3-8B-Instruct2025.12 | 93 | |
| ECVGPOBackbone=InternVL3-8B-Instruct2025.12 | 92.15 | |
| GRPOBackbone=InternVL3-8B-Instruct2025.12 | 91.65 | |
| ECVGPOBackbone=Qwen2.5-VL-3B-Instruct2025.12 | 90.55 | |
| GRPOBackbone=Qwen2.5-VL-3B-Instruct2025.12 | 90.16 | |
| Base ModelBackbone=InternVL3-2B-Instruct2025.12 | 89.8 | |
| ECVGPOBackbone=InternVL3-2B-Instruct2025.12 | 89.49 | |
| GRPOBackbone=InternVL3-2B-Instruct2025.12 | 89.21 | |
| SFTBackbone=Qwen2.5-VL-3B-Instruct2025.12 | 88.75 | |
| SFTBackbone=InternVL3-8B-Instruct2025.12 | 88.3 | |
| Base ModelBackbone=Qwen2.5-VL-3B-Instruct2025.12 | 88.05 | |
| SFTBackbone=InternVL3-2B-Instruct2025.12 | 86.65 |