Grounding on Renal Ultrasound cross-center (test)
69F1 Score @ IoU=0.25Echo-α-Grounding
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Echo-α-Grounding2026.04 | 69 | 56.73 | 30.76 | 69.8 | 59.6 | 37.72 | |
| Echo-α-GroundingOptimization=RL, tool-access=true2026.04 | 69 | 56.73 | 30.76 | 69.8 | 59.6 | 37.72 | |
| SFT + tool2026.04 | 65.45 | 53.26 | 28.88 | 65.91 | 57.85 | 35.03 | |
| SFT + tooltool-access=true2026.04 | 65.45 | 53.26 | 28.88 | 65.91 | 57.85 | 35.03 | |
| Specialized Detector2026.04 | 64.75 | 52.63 | 27.91 | 66.04 | 56.11 | 35.3 | |
| Specialized Detector2026.04 | 64.75 | 52.63 | 27.91 | 66.04 | 56.11 | 35.3 | |
| Direct MLLM + toolBackbone=Qwen3-VL2026.04 | 61.49 | 50.11 | 24.69 | 63.4 | 52.66 | 33.97 | |
| Direct MLLM + toolBackbone=Qwen3-VL, tool-access=true2026.04 | 61.49 | 50.11 | 24.69 | 63.4 | 52.66 | 33.97 | |
| SFT2026.04 | 47.61 | 36.3 | 10.73 | 43.49 | 33.15 | 9.8 | |
| SFTmode=shared SFT model without RL adaptation2026.04 | 47.61 | 36.3 | 10.73 | 43.49 | 33.15 | 9.8 | |
| Direct MLLMBackbone=Qwen3-VL2026.04 | 15.66 | 10.1 | 1.44 | 14.05 | 9.6 | 1.38 | |
| Direct MLLMBackbone=Qwen3-VL2026.04 | 15.66 | 10.1 | 1.44 | 14.05 | 9.6 | 1.38 |