Video Anomaly Detection on UBnormal (test)
79.6AUCQVAD
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| QVADSUPERVISION=TRAIN-FREE, EXPL.=✓2026.04 | 79.6 | — | — | — | — | |
| LRPOSupervision=Verbalized Learning, Explanation=✓, Training Data=100% / 100%, Backbone=InternVL3.5-8B2026.07 | 76.24 | — | — | — | — | |
| LRPOSupervision=Verbalized Learning, Explanation=✓, Training Data=2.5% / 6%, Backbone=InternVL3.5-8B2026.07 | 75.81 | — | — | — | — | |
| PANDASUPERVISION=TRAIN-FREE, EXPL.=✓2026.04 | 75.78 | — | — | — | — | |
| Our_woccSupervision Protocol=Unsupervised, Note=Trained as a whole in framework2024.01 | 74.76 | — | — | — | — | |
| Our_wocc (STG-NF)Supervision Protocol=One-Class Classification, Note=Degenerated to supervised OCC2024.01 | 72.81 | — | — | — | — | |
| AnomalyRulerImage Only=Yes, Training=No2024.07 | 71.9 | — | — | — | — | |
| FOLLOW THE RULESSUPERVISION=TRAIN-FREE, EXPL.=✓2026.04 | 71.9 | — | — | — | — | |
| AnomalyRulerVenue=ECCV’24, Supervision=Training-free, Explanation=✓, Training Data=-2026.07 | 71.9 | — | — | — | — | |
| STG-NFVenue=ICCV-23, Image Only=No, Training=Yes2024.07 | 71.8 | — | — | — | — | |
| STG-NFSupervision Protocol=One-Class Classification2024.01 | 71.8 | — | — | — | — | |
| Our_woccSupervision Protocol=Unsupervised2024.01 | 71.67 | — | — | — | — | |
| VERAVenue=CVPR’25, Supervision=Verbalized Learning, Explanation=✓, Training Data=100% / 100%, Backbone=InternVL3.5-8B2026.07 | 71.65 | — | — | — | — | |
| STG-NF All DataSupervision Protocol=Unsupervised2024.01 | 70.48 | — | — | — | — | |
| AnomalyRuler-baseImage Only=Yes, Training=No2024.07 | 69.8 | — | — | — | — | |
| URFVenue=NeurIPS’25, Supervision=Training-free, Explanation=✓, Training Data=-2026.07 | 69.02 | — | — | — | — | |
| Our_ws (RTFM)Supervision Protocol=Weakly Supervised, Features=I3D, Note=Degenerated to supervised WS2024.01 | 67.42 | — | — | — | — | |
| RTFMSupervision Protocol=Weakly Supervised, Features=I3D2024.01 | 66.83 | — | — | — | — | |
| VADTREESUPERVISION=TRAIN-FREE, EXPL.=✓2026.04 | 65.8 | — | — | — | — | |
| RTFMSUPERVISION=WEAK, EXPL.=×2026.04 | 64.94 | — | — | — | — | |
| RFTMVenue=ICCV’21, Supervision=Weak, Explanation=X, Training Data=100% / 100%2026.07 | 64.94 | — | — | — | — | |
| LAVADSUPERVISION=TRAIN-FREE, EXPL.=✓2026.04 | 64.23 | — | — | — | — | |
| LAVADVenue=CVPR’24, Supervision=Training-free, Explanation=✓, Training Data=-2026.07 | 64.23 | — | — | — | — | |
| STPromptFeature=CLIP2024.08 | 63.98 | — | — | — | — | |
| Our_wsSupervision Protocol=Unsupervised, Features=I3D, Note=Trained as a whole in framework2024.01 | 63.1 | — | — | — | — | |
| OVVADMode=Weak2023.11 | 62.94 | — | — | — | — | |
| OPVADFeature=CLIP2024.08 | 62.94 | — | — | — | — | |
| FPDMVenue=ICCV-23, Image Only=Yes, Training=Yes2024.07 | 62.7 | — | — | — | — | |
| VadCLIPFeature=CLIP2024.08 | 62.32 | — | — | — | — | |
| RTFMSupervision Protocol=Weakly Supervised, Features=C3D2024.01 | 62.3 | — | — | — | — | |
| SSMTL++Supervision Protocol=One-Class Classification2024.01 | 62.1 | — | — | — | — | |
| Georgescu et al. + anomaliesMode=Semi2023.11 | 61.3 | — | — | — | — | |
| Georgescu et al. + anomaliesFeature=None2024.08 | 61.3 | — | — | — | — | |
| RTFMMode=Weak2023.11 | 60.94 | — | — | — | — | |
| RTFMFeature=CLIP2024.08 | 60.94 | — | — | — | — | |
| DMUMode=Weak2023.11 | 59.91 | — | — | — | — | |
| DMUFeature=CLIP2024.08 | 59.91 | — | — | — | — | |
| Georgescu et al.Mode=Semi2023.11 | 59.3 | — | — | — | — | |
| BAFVenue=TPAMI-21, Image Only=No, Training=Yes2024.07 | 59.3 | — | — | — | — | |
| Georgescu et al.Feature=None2024.08 | 59.3 | — | — | — | — | |
| Georgescu et al.Supervision Protocol=One-Class Classification2024.01 | 59.3 | — | — | — | — | |
| Our_wsSupervision Protocol=Unsupervised, Features=C3D2024.01 | 59.05 | — | — | — | — | |
| AED-MADVenue=CVPR’24, Supervision=Semi, Explanation=X, Training Data=100% / 100%2026.07 | 58.5 | — | — | — | — | |
| STPAGVenue=CVPR’24, Supervision=Semi, Explanation=X, Training Data=100% / 100%2026.07 | 57.98 | — | — | — | — | |
| HOLMES-VAUSUPERVISION=INSTRU., EXPL.=✓2026.04 | 56.77 | — | — | — | — | |
| JigsawSupervision Protocol=One-Class Classification2024.01 | 56.4 | — | — | — | — | |
| Sultani et al.Supervision Protocol=Weakly Supervised, Features=I3D2024.01 | 54.12 | — | — | — | — | |
| LLaVA-1.5Venue=CVPR’24, Supervision=Training-free, Explanation=✓, Training Data=-2026.07 | 53.71 | — | — | — | — | |
| Wu et al.Mode=Weak2023.11 | 53.7 | — | — | — | — | |
| Wu et al.Feature=CLIP2024.08 | 53.7 | — | — | — | — | |
| Sultani et al.Feature=CLIP2024.08 | 53.23 | — | — | — | — | |
| Markovitz et al.Supervision Protocol=One-Class Classification2024.01 | 52 | — | — | — | — | |
| Sultani et al. (Spatial Variant)Feature=CLIP2024.08 | 51.95 | — | — | — | — | |
| Sultani et al.Mode=Weak2023.11 | 50.3 | — | — | — | — | |
| ZS CLIPVenue=ICML’21, Supervision=Training-free, Explanation=✓, Training Data=-2026.07 | 46.2 | — | — | — | — | |
| [10]Type=Object-centric, FPS=202023.06 | — | 62.1 | 86.5 | — | — | |
| [26]Type=Object-centric, FPS=512023.06 | — | 55.4 | 84.5 | — | — | |
| [27]Type=Object-centric, FPS=242023.06 | — | 61.3 | 85.6 | — | — | |
| [40]Type=Object-centric2023.06 | — | 50.7 | — | — | — | |
| [77]Type=Frame or cube level, FPS=562023.06 | — | 76.5 | 50.3 | — | — | |
| [9]Type=Frame or cube level, FPS=372023.06 | — | 68.5 | 50.3 | — | — | |
| [93]Type=Frame or cube level2023.06 | — | 62.7 | — | — | — | |
| AccI-VAD_GMMimplemented_by_authors=true, backbone=Hiera-L2024.03 | — | 66.8 | 83.2 | — | — | |
| AccI-VAD_KNNimplemented_by_authors=true, backbone=Hiera-L2024.03 | — | 65.2 | 82.5 | — | — | |
| AED-MAEType=Frame or cube level, FPS=16552023.06 | — | 58.5 | 81.4 | — | — | |
| Bertasius et al. (2021)Sampling rate=1/322022.07 | — | 68.5 | 80.3 | 0.04 | 0.05 | |
| Bertasius et al. (2021)Sampling rate=1/82022.07 | — | 64.1 | 75.4 | 0.04 | 0.05 | |
| Bertasius et al. (2021)Sampling rate=1/42022.07 | — | 61.9 | 75.4 | 0.04 | 0.06 | |
| FPDM2024.03 | — | 62.7 | — | — | — | |
| Georgescu et al. (2022)Training data=Standard2022.07 | — | 59.3 | 84.9 | 21.91 | 53.44 | |
| Georgescu et al. (2022)Training data=UBnormal2022.07 | — | 61.3 | 85.6 | 25.43 | 56.27 | |
| MSMAimplemented_by_authors=true, backbone=Hiera-L2024.03 | — | 70.3 | 85.1 | — | — | |
| MULDEregularization=true, backbone=Hiera-L2024.03 | — | 72.8 | 85.5 | — | — | |
| MULDE_beta=0regularization=false, backbone=Hiera-L2024.03 | — | 71.3 | 86 | — | — | |
| SSMTL Georgescu et al. (2021)2022.07 | — | 55.4 | 84.5 | 19.71 | 55.8 | |
| SSMTL++v12022.07 | — | 62.1 | 86.5 | 25.63 | 63.53 | |
| SSMTL++v22022.07 | — | 56 | 85.9 | 20.33 | 57.76 | |
| Sultani et al. (2018)Protocol=pre-trained2022.07 | — | 49.5 | 77.4 | — | — | |
| Sultani et al. (2018)Protocol=fine-tuned2022.07 | — | 50.3 | 76.8 | — | — | |
| UBnormal2024.03 | — | 68.5 | 80.3 | — | — |