Degradation-induced covariate shift detection on COCO synthetically corrupted variants
88.64AUROC (Severity 1)DM backbone YOLOv10-m
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DM backbone YOLOv10-mApproach category=DM, Backbone / Detector architecture=YOLOv10-m2026.02 | 88.64 | 89.7 | 89.75 | 95.28 | 97.14 | |
| DM backbone YOLOv9-mApproach category=DM, Backbone / Detector architecture=YOLOv9-m2026.02 | 87.64 | 89.06 | 90.07 | 95.66 | 96.62 | |
| DM backbone YOLOv11-mApproach category=DM, Backbone / Detector architecture=YOLOv11-m2026.02 | 85.84 | 89.12 | 89.95 | 94.58 | 96.04 | |
| DM backbone RT DETR-lApproach category=DM, Backbone / Detector architecture=RT DETR-l2026.02 | 83.88 | 88.29 | 90.37 | 95.61 | 97.11 | |
| ARNIQUAEvaluation protocol=embedding-based; prototype anchored2026.02 | 73.6 | 80.07 | 80.58 | 83.3 | 85.74 | |
| MANIQAEvaluation protocol=embedding-based; prototype anchored2026.02 | 65.87 | 71.55 | 73.42 | 76.97 | 77.3 | |
| CLIPIQAEvaluation protocol=score-based2026.02 | 57.78 | 60.43 | 62.29 | 69.28 | 69.71 | |
| CLIPIQA+Evaluation protocol=score-based2026.02 | 56.42 | 58.11 | 61.03 | 63.45 | 67.81 | |
| Entropy (Classification)Backbone / Detector architecture=FasterRCNN, Evaluation protocol=ES2026.02 | 54.7 | 59.97 | 62.84 | 68.47 | 77.6 | |
| Entropy (Classification)Backbone / Detector architecture=FasterRCNN, Evaluation protocol=NLL2026.02 | 53.31 | 59.65 | 64.35 | 70.43 | 77.33 | |
| QualiCLIPEvaluation protocol=score-based2026.02 | 53.17 | 56.77 | 66.28 | 70.6 | 72.53 | |
| Confidence ScoreBackbone / Detector architecture=RetinaNet, Evaluation protocol=NLL2026.02 | 52.91 | 56.93 | 61.51 | 69.69 | 77.65 | |
| MANIQAEvaluation protocol=score-based2026.02 | 52.88 | 54.49 | 55.27 | 56.67 | 61.56 | |
| Entropy (Regression)Backbone / Detector architecture=RetinaNet, Evaluation protocol=ES2026.02 | 52.73 | 55.15 | 57.67 | 60.82 | 66.51 | |
| TraceBackbone / Detector architecture=RetinaNet, Evaluation protocol=ES2026.02 | 52.61 | 55.31 | 57.14 | 60.29 | 65.44 | |
| QualiCLIP+Evaluation protocol=score-based2026.02 | 52.56 | 57.9 | 65.49 | 69.18 | 70.87 | |
| Confidence ScoreBackbone / Detector architecture=RetinaNet, Evaluation protocol=ES2026.02 | 52.55 | 56.77 | 60.72 | 68.1 | 75.28 | |
| ARNIQUAEvaluation protocol=score-based2026.02 | 52.09 | 52.83 | 58.26 | 58.43 | 60.23 | |
| Entropy (Regression)Backbone / Detector architecture=FasterRCNN, Evaluation protocol=NLL2026.02 | 51.65 | 51.66 | 55.72 | 56.9 | 64.87 | |
| TraceBackbone / Detector architecture=DETR, Evaluation protocol=NLL2026.02 | 51.14 | 54.83 | 58.96 | 59.68 | 63.61 | |
| M-NF backboneFeature levels=all; C1-C52026.02 | 49.9 | 50.25 | 58.76 | 63.9 | 69.12 | |
| CLIPIQA+Evaluation protocol=embedding-based; prototype anchored2026.02 | 49.85 | 48.65 | 47.57 | 46.61 | 45.33 | |
| NF backboneFeature levels=all; C1-C52026.02 | 49.8 | 50.2 | 56.12 | 64.9 | 73.45 | |
| M-NF backboneFeature levels=C1, C2, C32026.02 | 49.8 | 50.2 | 58.61 | 63.8 | 68.94 | |
| NF backboneFeature levels=C1, C2, C32026.02 | 49.7 | 50.1 | 55.12 | 64.4 | 73.56 | |
| M-NF backboneFeature levels=C4, C52026.02 | 49.7 | 50.1 | 55.76 | 61.6 | 67.54 | |
| DeterminantBackbone / Detector architecture=RetinaNet, Evaluation protocol=ES2026.02 | 49.65 | 50.57 | 51.29 | 52.66 | 56.39 | |
| NF backboneFeature levels=C4, C52026.02 | 49.6 | 49.9 | 51.12 | 56.2 | 61.53 | |
| DeterminantBackbone / Detector architecture=RetinaNet, Evaluation protocol=NLL2026.02 | 49.48 | 51.09 | 50.94 | 54.13 | 58.23 | |
| CLIPIQAEvaluation protocol=embedding-based; prototype anchored2026.02 | 49.36 | 47.71 | 47.58 | 46.65 | 45.76 | |
| QualiCLIP+Evaluation protocol=embedding-based; prototype anchored2026.02 | 47.04 | 42.45 | 45.31 | 50.23 | 56.24 | |
| QualiCLIPEvaluation protocol=embedding-based; prototype anchored2026.02 | 45.36 | 41.53 | 46.77 | 49.79 | 56.16 |