Radiology Report Generation on Tmild Mildly Suboptimal Benchmark (test)
37.6BLEU-1R2GenGPT
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| R2GenGPTdual-loop training strategy=true2026.04 | 37.6 | 10.3 | 16.1 | 27.1 | 41.1 | 29.8 | 32.1 | 53 | 26.9 | |
| R2GenGPT2026.04 | 33.8 | 9.2 | 16.6 | 26 | 35.3 | 28.8 | 29.1 | 52.4 | 25.5 | |
| Qwen3VL4Bsupervised fine-tuning=true2026.04 | 22.2 | 5.1 | 13.2 | 19.9 | 34.9 | 25.6 | 26.9 | 52.1 | 25.7 | |
| Qwen3VL8Bsupervised fine-tuning=true2026.04 | 21.9 | 4.8 | 12.8 | 19.3 | 36.6 | 22.4 | 23.9 | 50.9 | 24.3 | |
| GPT5.22026.04 | 12.8 | 0.7 | 10.4 | 10.8 | 22.5 | 23.8 | 19.7 | 47.3 | 16.7 | |
| Qwen3VL8B2026.04 | 7.4 | 0.5 | 10.2 | 9.5 | 18.1 | 16.9 | 15.3 | 49.7 | 14.3 | |
| Qwen3VL4B2026.04 | 7.1 | 0.7 | 10.2 | 9.9 | 21.1 | 27.3 | 20.5 | 49.9 | 13.7 |