Interval-change classification on MS-CXR-T
68.5Average Macro AccuracyBioViL-T
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BioViL-TModel Type=External SOTA2026.04 | 68.5 | — | — | — | — | — | |
| Gemini 3.0 FlashModel Type=Large Models2026.04 | 67.3 | — | — | — | — | — | |
| ProTransEvaluation Protocol=Text prototype2026.06 | 65.81 | 63.86 | 66.42 | 69.05 | 60.06 | 69.68 | |
| MedGemma 1.5 4BModel Type=Small Models2026.04 | 65.7 | — | — | — | — | — | |
| Coca-CXREvaluation Protocol=Text prototype, Next token prediction=true2026.06 | 65 | 69.6 | 68.1 | 56.4 | 59.3 | 71.8 | |
| TempA-VLPEvaluation Protocol=Text prototype2026.06 | 64.8 | 65.2 | 59.4 | 73.4 | 43.1 | 77.1 | |
| BioViL-T_TILASetup=Supervised, Protocol=Standard2026.04 | 64.1 | 63.3 | 66.7 | 67.1 | 52.3 | 70.9 | |
| ProTransEvaluation Protocol=SVM2026.06 | 63.54 | 61.54 | 61.81 | 69.35 | 55.46 | 69.53 | |
| Gemini 3.0 ProModel Type=Large Models2026.04 | 62.9 | — | — | — | — | — | |
| MedSTEvaluation Protocol=Text prototype2026.06 | 62.66 | 61.06 | 63.03 | 67.79 | 56.14 | 65.29 | |
| MedSTEvaluation Protocol=SVM2026.06 | 61.12 | 60.57 | 58.47 | 65 | 54.18 | 67.35 | |
| MedGemma 1 4BModel Type=Small Models2026.04 | 61.1 | — | — | — | — | — | |
| BioViL-TEvaluation Protocol=SVM2026.06 | 59.02 | 56.93 | 53.94 | 67.24 | 55.46 | 61.55 | |
| Gemma 3 4BModel Type=Small Models2026.04 | 59 | — | — | — | — | — | |
| BioViL-T_TILASetup=Supervised, Protocol=Consistency2026.04 | 57.4 | 60.2 | 58.3 | 57.8 | 43.1 | 67.4 | |
| BioViL-TEvaluation Protocol=Text prototype2026.06 | 56.3 | 51.99 | 50.24 | 66.39 | 55.5 | 57.39 | |
| BioViL-T_TILASetup=Retrieval, Protocol=Standard2026.04 | 54.1 | 55.3 | 58.7 | 58.1 | 40.3 | 58 | |
| Diff-RRGEvaluation Protocol=SVM2026.06 | 53.93 | 47.94 | 48.66 | 67.1 | 55.46 | 50.5 | |
| Qwen3 VL 4BModel Type=Small Models2026.04 | 53.5 | — | — | — | — | — | |
| Diff-RRGEvaluation Protocol=Text prototype2026.06 | 53.36 | 55.15 | 48.79 | 61.04 | 50.71 | 51.12 | |
| Gemma 3 27BModel Type=Small Models2026.04 | 52.7 | — | — | — | — | — | |
| BioViL-TSetup=Retrieval, Protocol=Combined2026.04 | 51.2 | 53.7 | 54.3 | 50.8 | 37.1 | 59.8 | |
| BioViL-T_SigLIPSetup=Retrieval, Protocol=Standard2026.04 | 50.2 | 48.9 | 58.3 | 55.6 | 34.4 | 53.6 | |
| MedGemma 1 27BModel Type=Small Models2026.04 | 50.1 | — | — | — | — | — | |
| CNN-TF_TILASetup=Supervised, Protocol=Standard2026.04 | 49.1 | 48.6 | 55.8 | 42.9 | 37.5 | 60.7 | |
| CNN-TFSetup=Supervised, Protocol=Standard2026.04 | 47.9 | 46.6 | 55.2 | 41.3 | 35.2 | 61.2 | |
| CheXRelFormerSetup=Supervised, Protocol=Standard2026.04 | 47.5 | 51.1 | 56 | 37.1 | 34.1 | 59 | |
| BioViL-T_TILASetup=Zero-shot, Protocol=Standard2026.04 | 47.2 | 46.5 | 46.3 | 51.9 | 38.8 | 52.3 | |
| BioViL-TSetup=Retrieval, Protocol=Reversed2026.04 | 44.8 | 42.8 | 51.7 | 45.1 | 36.2 | 48.2 | |
| ALTA_TILASetup=Retrieval, Protocol=Standard2026.04 | 44.5 | 49.2 | 45.9 | 45.5 | 35.4 | 46.4 | |
| BioViL-TSetup=Retrieval, Protocol=Standard2026.04 | 43.1 | 49.1 | 42.8 | 38.3 | 33.1 | 51.8 | |
| ALTASetup=Retrieval, Protocol=Standard2026.04 | 37.7 | 46.8 | 33.9 | 36 | 31.1 | 40.4 | |
| BioViL-TSetup=Zero-shot, Protocol=Standard2026.04 | 37.4 | 34.9 | 37.4 | 34.5 | 37 | 42.8 | |
| BioViL-TSetup=Retrieval, Protocol=Consistency2026.04 | 21.3 | 23.4 | 21.9 | 21.7 | 10.1 | 29.2 |