Medical Image Classification on Aggregated Medical Datasets
24.87PPEQPOND
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| EQPONDBackbone=MedGemma-4B2025.10 | 24.87 | 9.09 | 6.35 | 4.84 | 9.19 | 4.18 | 8.36 | 78.82 | 34.84 | 75.67 | 33.23 | |
| GRPO+RSBackbone=MedGemma-4B2025.10 | 24.55 | 6.97 | 4.78 | 4.22 | 8.34 | 4.2 | 8.77 | 79.02 | 28.25 | 75.84 | 27.11 | |
| RLOOBackbone=Qwen-2.5-VL-7B2025.10 | 22.34 | 6.67 | 5.68 | 3.3 | 6.93 | 4.86 | 10 | 78.22 | 25.23 | 74.59 | 24.43 | |
| GRPOBackbone=MedGemma-4B2025.10 | 20.89 | 6.3 | 5.26 | 3.87 | 7.53 | 4.19 | 8.57 | 79.38 | 31.34 | 76.19 | 30.17 | |
| RLOOBackbone=MedGemma-4B2025.10 | 20.45 | 10.35 | 5.51 | 5.92 | 11.73 | 4.29 | 8.79 | 79.76 | 32.37 | 76.48 | 30.56 | |
| Re++Backbone=MedGemma-4B2025.10 | 20.3 | 7.78 | 5.69 | 4.69 | 8.98 | 4.44 | 8.99 | 78.76 | 31.05 | 75.41 | 29.66 | |
| GRPO+RSBackbone=Qwen-2.5-VL-7B2025.10 | 19.62 | 6.85 | 6.44 | 3.19 | 6.28 | 5.5 | 11.26 | 75.61 | 25.8 | 71.67 | 25 | |
| GRPO+DROBackbone=MedGemma-4B2025.10 | 18.2 | 7.52 | 5.68 | 4.56 | 8.95 | 4.55 | 9.39 | 80.17 | 31.46 | 76.69 | 30.09 | |
| GRPOBackbone=Qwen-2.5-VL-7B2025.10 | 17.9 | 7.93 | 4.85 | 3.87 | 8.21 | 4.85 | 9.92 | 78.4 | 26.01 | 76.21 | 24.25 | |
| Re++Backbone=Qwen-2.5-VL-7B2025.10 | 16.66 | 6.66 | 6.37 | 3.22 | 6.47 | 5.06 | 10.33 | 75.31 | 25.99 | 71.69 | 25.18 | |
| EQPOBackbone=Qwen-2.5-VL-7B2025.10 | 15.42 | 5.62 | 5 | 2.54 | 5.22 | 4.42 | 8.95 | 78.52 | 26.57 | 77.14 | 26.02 | |
| EQPO (EquiMedGemma)Backbone=MedGemma-4B2025.10 | 12.95 | 6.84 | 5.53 | 3.79 | 7.24 | 4.11 | 8.53 | 80.4 | 32.75 | 77.23 | 31.55 |