Human Artifact Detection on ArtifactLens Suite (cross-dataset 1.0)
81.7Overall F1ArtifactLens
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| ArtifactLensEvaluation Protocol=ArtifactLens, Backbone=Gemini-2.5-pro2026.02 | 81.7 | 75.5 | 90.6 | 75.9 | 84 | 84.1 | 84.9 | 79.8 | |
| ArtifactLensEvaluation Protocol=ArtifactLens, Backbone=GPT-4o2026.02 | 80.2 | 76.6 | 86.5 | 79.6 | 81.4 | 82.2 | 81.7 | 76 | |
| HADMEvaluation Protocol=Finetuned, In-distribution Training=HAD2026.02 | 75.8 | 77.3 | 75.9 | 61.1 | 86.5 | 72.9 | 81 | 77.3 | |
| MagicAssessorEvaluation Protocol=Finetuned, In-distribution Training=MagicBench2026.02 | 75.4 | 80.8 | 72 | 70.2 | 81.4 | 73.8 | 75.4 | 77.8 | |
| ArtifactLensEvaluation Protocol=ArtifactLens, Backbone=Gemini-2.5-flash2026.02 | 74.7 | 75.2 | 77.8 | 64 | 79.2 | 83.1 | 79.5 | 67.5 | |
| AHDEvaluation Protocol=Finetuned, In-distribution Training=AIGC-HA2026.02 | 62 | 75.7 | 55.4 | 79.1 | 53.7 | 79.1 | 58.1 | 64.5 | |
| ArtifactLensEvaluation Protocol=ArtifactLens, Backbone=GPT-4o-mini2026.02 | 57.1 | 80.5 | 45.1 | 46.2 | 65.4 | 69 | 61.1 | 43.8 | |
| Gemini-2.5-proEvaluation Protocol=Zero-shot2026.02 | 56 | 81.7 | 43.5 | 50.4 | 45.1 | 66.3 | 62.3 | 79.1 | |
| DiffdoctorEvaluation Protocol=Finetuned2026.02 | 54.4 | 78.3 | 42.5 | 37.2 | 59.3 | 50.3 | 63.6 | 61.8 | |
| ArtifactLensEvaluation Protocol=ArtifactLens, Backbone=QwenVL-2.5-7B2026.02 | 50.1 | 78 | 37.6 | 36.8 | 47.9 | 61.1 | 54.4 | 40.2 | |
| Gemini-2.5-flashEvaluation Protocol=Zero-shot2026.02 | 36.9 | 78.9 | 25.4 | 30.8 | 19.8 | 54.2 | 42.7 | 70.6 | |
| GPT-4oEvaluation Protocol=Zero-shot2026.02 | 21.7 | 80.7 | 13.5 | 34.7 | 3.8 | 37.7 | 10.6 | 43.7 | |
| GPT-4o-miniEvaluation Protocol=Zero-shot2026.02 | 8.7 | 81.6 | 4.7 | 10.8 | 0.2 | 14.4 | 9.3 | 10.3 | |
| QwenVL-2.5-7BEvaluation Protocol=Zero-shot2026.02 | 1.7 | 64.6 | 0.9 | 2.3 | 0.1 | 1.9 | 2.5 | 1.1 |