Image Grounding on Ref-L4
83.2Acc@0.5EchoGen
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| EchoGenCapability=Understanding and Generation2026.03 | 83.2 | 75.42 | 47.32 | 68.46 | — | |
| CogVLM-g.Capability=Understanding only2026.03 | 81.7 | 70.77 | 48.35 | 66.09 | — | |
| Qwen-VL-ChatCapability=Understanding only2026.03 | 73.8 | 58.05 | 37.16 | 55.94 | — | |
| KOSMOS-2Capability=Understanding only2026.03 | 48.53 | 38.34 | 17.54 | 34.72 | — | |
| OursTraining=RL-enhanced2025.07 | — | — | — | — | 75.26 | |
| Qwen2.5-VL-7BModel Size=7B2025.07 | — | — | — | — | 60.67 | |
| Qwen2.5-VL-7B + SFTTraining=SFT2025.07 | — | — | — | — | 69.03 |