Radiology Report Generation on Tsevere Severely Suboptimal Benchmark (test)
38.3BLEU-1R2GenGPT
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| R2GenGPTdual-loop training strategy=true2026.04 | 38.3 | 10.3 | 16 | 26.8 | 32.3 | 27.4 | 27.7 | 53.6 | 27.5 | |
| R2GenGPT2026.04 | 36.7 | 9.5 | 15.9 | 26.5 | 32 | 27.4 | 27.5 | 53.5 | 26.9 | |
| Qwen3VL4Bsupervised fine-tuning=true2026.04 | 24.6 | 5.2 | 13 | 19.9 | 35.4 | 26.3 | 27.4 | 52.7 | 25 | |
| Qwen3VL8Bsupervised fine-tuning=true2026.04 | 23.6 | 5.2 | 13 | 19.8 | 37.3 | 26.5 | 27.5 | 52.1 | 24.8 | |
| GPT5.22026.04 | 13 | 0.7 | 10.4 | 10.5 | 21 | 20.9 | 17.7 | 47.2 | 16.2 | |
| Qwen3VL8B2026.04 | 7.9 | 0.6 | 10.6 | 9.8 | 21.8 | 20.3 | 18.5 | 49.8 | 14.6 | |
| Qwen3VL4B2026.04 | 7.6 | 0.7 | 10.5 | 10.2 | 22.5 | 27.3 | 20.8 | 50.6 | 14.4 |