Multi-label classification on PadChest-GR
22.35F1 ScoreMedSigLIP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MedSigLIPThresholding strategy=Performance-based (Risk ×), Evaluation protocol=Zero-shot2026.01 | 22.35 | 21.49 | 43.23 | 19.5 | |
| CLIP Base + TAFThresholding strategy=Risk-constrained (Risk ✓), Evaluation protocol=Fine-tuning2026.01 | 21.17 | 15.01 | 33.69 | 1.6 | |
| MedSigLIPThresholding strategy=Risk-constrained (Risk ✓), Evaluation protocol=Zero-shot2026.01 | 17.93 | 8.24 | 31.65 | 0.9 | |
| BiomedCLIPThresholding strategy=Performance-based (Risk ×), Evaluation protocol=Zero-shot2026.01 | 15.36 | 8.71 | 40.3 | 11.6 | |
| BiomedCLIPThresholding strategy=Risk-constrained (Risk ✓), Evaluation protocol=Zero-shot2026.01 | 14.75 | 6.98 | 37.57 | 2.4 | |
| MedCLIPThresholding strategy=Performance-based (Risk ×), Evaluation protocol=Zero-shot2026.01 | 14.04 | 0.03 | 24.52 | 0.2 | |
| MedCLIPThresholding strategy=Risk-constrained (Risk ✓), Evaluation protocol=Zero-shot2026.01 | 13.99 | 0.03 | 24.45 | 0.7 | |
| CLIP BaseThresholding strategy=Performance-based (Risk ×), Evaluation protocol=Zero-shot2026.01 | 11.51 | 3.43 | 33.27 | 0.5 | |
| CLIP BaseThresholding strategy=Risk-constrained (Risk ✓), Evaluation protocol=Zero-shot2026.01 | 10.14 | 2.32 | 34.56 | 1.5 | |
| PubMedCLIPThresholding strategy=Risk-constrained (Risk ✓), Evaluation protocol=Zero-shot2026.01 | 10.09 | 3.61 | 38.55 | 2.2 | |
| PubMedCLIPThresholding strategy=Performance-based (Risk ×), Evaluation protocol=Zero-shot2026.01 | 9.68 | 2.48 | 37.64 | 1.5 |