Anomaly Detection on INet + SpaceImages Imagenette + AstronomyImages (test)
0.9625AUC-ROCRangeAD
Evaluation results
| Method | Links | |
|---|---|---|
| RangeADBackbone=Swin-v2-Vision Transformer [26], Feature extraction point=in (after GELU and Convolution-Layers), Quantile=0.012026.03 | 0.9625 | |
| RangeADBackbone=Swin-v2-Vision Transformer [26], Feature extraction point=in (after GELU and Convolution-Layers), Quantile=0.0012026.03 | 0.9592 | |
| RangeADBackbone=Swin-v2-Vision Transformer [26], Feature extraction point=after (after Transformer-Block), Quantile=0.012026.03 | 0.9094 | |
| RangeADBackbone=Swin-v2-Vision Transformer [26], Feature extraction point=after (after Transformer-Block), Quantile=0.0012026.03 | 0.909 | |
| DeepSVDDTraining data=10% of the INet testset2026.03 | 0.6757 | |
| Isolation Forest2026.03 | 0.4977 | |
| AutoencoderTraining data=10% of the INet testset2026.03 | 0.1864 |