Anomaly Detection on Fishyscapes Static
86.2APOutlier Head
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Outlier Headscore=random patches, re-training=true, OoD data=true, time [s]=0.122019.04 | 86.2 | 2.4 | — | |
| Outlier Headscore=combined probability, re-training=true, OoD data=true, time [s]=0.122019.04 | 84 | 10.3 | — | |
| Outlier Headscore=fixed patches, re-training=true, OoD data=true, time [s]=0.122019.04 | 82.9 | 5.1 | — | |
| SynBoostscore=dissimilarity score, re-training=false, OoD data=true, time [s]=2.322019.04 | 72.6 | 18.8 | — | |
| Learned Embedding Densityscore=minimum NLL, re-training=false, OoD data=false, time [s]=1.532019.04 | 62.1 | 17.4 | — | |
| Learned Embedding Densityscore=logistic regression, re-training=false, OoD data=true, time [s]=1.532019.04 | 57.2 | 13.4 | — | |
| Bayesian DeepLabscore=mutual information, re-training=true, OoD data=false, time [s]=3.622019.04 | 48.7 | 15.5 | — | |
| OoD trainingscore=void classifier, re-training=true, OoD data=true, time [s]=0.052019.04 | 45 | 19.4 | — | |
| kNN Embeddingscore=density, re-training=false, OoD data=false, time [s]=7.892019.04 | 44 | 20.2 | — | |
| Learned Embedding Densityscore=single-layer NLL, re-training=false, OoD data=false, time [s]=0.292019.04 | 40.9 | 21.3 | — | |
| Dirichlet DeepLabscore=prior entropy, re-training=true, OoD data=true, time [s]=0.052019.04 | 31.3 | 84.6 | — | |
| Image Resynthesisscore=resynthesis difference, re-training=false, OoD data=false, time [s]=2.392019.04 | 29.6 | 27.1 | — | |
| OoD trainingscore=max-entropy, re-training=true, OoD data=true, time [s]=0.292019.04 | 27.5 | 23.6 | — | |
| kNN Embeddingscore=relative class density, re-training=false, OoD data=false, time [s]=6.452019.04 | 15.8 | 100 | — | |
| Softmaxscore=entropy, re-training=false, OoD data=false, time [s]=0.292019.04 | 15.4 | 39.8 | — | |
| Softmaxscore=max-probability, re-training=false, OoD data=false, time [s]=0.052019.04 | 12.9 | 39.8 | — | |
| Randomscore=random uncertainty, re-training=false, OoD data=false, time [s]=-2019.04 | 2.5 | 95 | — | |
| Dirichlet priorAux data=true, Img rsyn.=false2022.07 | 0.846 | 30 | — | |
| OOD HeadAux data=true, Img rsyn.=false2022.07 | 0.84 | 10.3 | — | |
| SynBoostAux data=true, Img rsyn.=true2022.07 | 0.726 | 18.8 | — | |
| DenseHybridAux data=true, Img rsyn.=false2022.07 | 0.723 | 5.5 | — | |
| Embed. Dens.Aux data=false, Img rsyn.=false2022.07 | 0.621 | 17.4 | — | |
| Standardized MLAux data=false, Img rsyn.=false2022.07 | 0.531 | 19.6 | — | |
| Mutual informationAux data=true, Img rsyn.=false2022.07 | 0.487 | 15.5 | — | |
| Void ClassifierAux data=true, Img rsyn.=false2022.07 | 0.45 | 19.4 | — | |
| Proposed single-reference anomaly detection methodBackbone=DINOv3 [16], Training Protocol=training-free, Reference Setup=single-reference, Resolution=fixed spatial resolution2026.05 | 0.4115 | 81.7 | 74.62 | |
| Image Resyn.Aux data=true, Img rsyn.=false2022.07 | 0.296 | 27.1 | — | |
| Max softmaxAux data=false, Img rsyn.=false2022.07 | 0.129 | 39.8 | — |