Video Anomaly Detection on ShanghaiTech
0.9832ROC AUCCLIP-TSA
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CLIP-TSAFeature=CLIP2024.08 | 0.9832 | — | — | — | — | — | — | |
| STPromptFeature=CLIP2024.08 | 0.9781 | — | — | — | — | — | — | |
| DMUFeature=CLIP2024.08 | 0.9757 | — | — | — | — | — | — | |
| MIL-BERTFeature=I3D-RGB+Flow, Crop=12022.10 | 0.9754 | — | — | — | — | — | — | |
| VadCLIPFeature=CLIP2024.08 | 0.9749 | — | — | — | — | — | — | |
| [28]Feature=I3D-RGB, Crop=12022.10 | 0.9748 | — | — | — | — | — | — | |
| [14]Feature=VideoSwin-RGB, Crop=102022.10 | 0.9732 | — | — | — | — | — | — | |
| RTFMFeature=I3D-RGB, Crop=102022.10 | 0.9721 | — | — | — | — | — | — | |
| VigilFormerFPS=41.52026.05 | 0.9721 | — | — | — | — | — | — | |
| MSLFeature=VideoSwin [36]2024.08 | 0.972 | — | — | — | — | — | — | |
| OPVADFeature=CLIP2024.08 | 0.9698 | — | — | — | — | — | — | |
| BN-WVADFPS=22.82026.05 | 0.9681 | — | — | — | — | — | — | |
| UMILFeature=X-CLIP [41]2024.08 | 0.9678 | — | — | — | — | — | — | |
| RTFMFeature=CLIP2024.08 | 0.9676 | — | — | — | — | — | — | |
| STG-NFFPS=29.42026.05 | 0.9675 | — | — | — | — | — | — | |
| UR-DMUFPS=26.72026.05 | 0.9674 | — | — | — | — | — | — | |
| CLIP-TSAFPS=18.42026.05 | 0.9642 | — | — | — | — | — | — | |
| SSRLFeature=CLIP2024.08 | 0.9622 | — | — | — | — | — | — | |
| MGFNFPS=24.32026.05 | 0.9615 | — | — | — | — | — | — | |
| Wu et al.Feature=CLIP2024.08 | 0.9524 | — | — | — | — | — | — | |
| MistFeature=I3D-RGB, Crop=12022.10 | 0.9483 | — | — | — | — | — | — | |
| MISTFPS=27.12026.05 | 0.9483 | — | — | — | — | — | — | |
| MSL2026.05 | 0.9481 | — | — | — | — | — | — | |
| Sultani et al.Feature=CLIP2024.08 | 0.9172 | — | — | — | — | — | — | |
| RTFMFPS=31.52026.05 | 0.9151 | — | — | — | — | — | — | |
| SSMTApproach=frame & object level2022.03 | 0.902 | — | — | — | — | — | — | |
| Zaheer et al. [21]Features=C3D2020.07 | 0.8967 | — | — | — | — | — | — | |
| SSMTApproach=object-level2022.03 | 0.893 | — | — | — | — | — | — | |
| Proposed (our network + our loss)Features=C3D2020.07 | 0.8742 | — | — | — | — | — | — | |
| Sultani et al.FPS=38.22026.05 | 0.863 | — | — | — | — | — | — | |
| [17]Feature=I3D-RGB, Crop=102022.10 | 0.853 | — | — | — | — | — | — | |
| GCN-AnomalyFPS=25.62026.05 | 0.8444 | — | — | — | — | — | — | |
| Zaheer et al. [53]Features=C3D2020.07 | 0.8416 | — | — | — | — | — | — | |
| SSMTApproach=frame-level2022.03 | 0.835 | — | — | — | — | — | — | |
| BAFApproach=object-level2022.03 | 0.827 | — | — | — | 41.43 | 78.79 | — | |
| Sultani et al.Feature=CLIP, Variant=Patch-level detection2024.08 | 0.8025 | — | — | — | — | — | — | |
| OMAEApproach=object-level2022.03 | 0.7918 | — | — | — | 51.51 | 82.19 | — | |
| dummyAEApproach=object-level2022.03 | 0.787 | — | — | — | 20.7 | 44.5 | — | |
| Zhong et al.Features=C3D2020.07 | 0.7644 | — | — | — | — | — | — | |
| Multi-timescale model (Ours)2019.08 | 0.7603 | — | — | — | — | — | — | |
| VEC-AMMethod Category=Proposed2020.08 | 0.748 | — | — | — | — | — | — | |
| VEC-AMethod Category=Proposed2020.08 | 0.747 | — | — | — | — | — | — | |
| AnoPCNMethod Category=Hybrid Methods2020.08 | 0.736 | — | — | — | — | — | — | |
| MPED-RNN2019.03 | 0.734 | — | — | — | — | — | — | |
| Morais et al.2019.08 | 0.734 | — | — | — | — | — | — | |
| Skeleton-Trajectories + MPED-RNNMethod Category=Hybrid Methods2020.08 | 0.734 | — | — | — | — | — | — | |
| Prediction&ReconstructionMethod Category=Hybrid Methods2020.08 | 0.73 | — | — | — | — | — | — | |
| Liu et al.2019.03 | 0.728 | — | — | — | — | — | — | |
| Liu et al.2019.08 | 0.728 | — | — | — | — | — | — | |
| Frame-PredictionMethod Category=Frame Prediction Based2020.08 | 0.728 | — | — | — | — | — | — | |
| Abati et al.2019.08 | 0.725 | — | — | — | — | — | — | |
| PDE-AEMethod Category=Reconstruction Based2020.08 | 0.725 | — | — | — | — | — | — | |
| MNADApproach=frame-level2022.03 | 0.725 | — | — | — | — | — | — | |
| Mem-AEMethod Category=Reconstruction Based2020.08 | 0.712 | — | — | — | — | — | — | |
| MemAEApproach=frame-level2022.03 | 0.712 | — | — | — | — | — | — | |
| Conv-AE2019.03 | 0.704 | — | — | — | — | — | — | |
| Conv-AE2019.08 | 0.704 | — | — | — | — | — | — | |
| TSC SRNN2019.03 | 0.68 | — | — | — | — | — | — | |
| TSC-rRNN2019.08 | 0.68 | — | — | — | — | — | — | |
| SRNNMethod Category=Reconstruction Based2020.08 | 0.68 | — | — | — | — | — | — | |
| CAEMethod Category=Reconstruction Based2020.08 | 0.609 | — | — | — | — | — | — | |
| AccI-VAD2024.03 | — | — | 0.859 | 0.896 | — | — | — | |
| AccI-VAD_GMMimplementation=re-implemented by authors2024.03 | — | — | 0.762 | 0.829 | — | — | — | |
| AccI-VAD_KNNimplementation=re-implemented by authors2024.03 | — | — | 0.719 | 0.831 | — | — | — | |
| Action HintsSetting=zero-shot, Training on SHT=False2025.09 | — | — | — | — | — | — | 0.841 | |
| Action Hints-fullSetting=full-shot, Training on SHT=True2025.09 | — | — | — | — | — | — | 0.86 | |
| Ada-Netw/ obj.=false, Manually Cropped=true2022.09 | — | — | 0.7 | — | — | — | — | |
| AI-VADYear=20232023.09 | — | — | — | — | — | — | 0.8594 | |
| AMMC-NetMethod Category=Hybrid and Other Methods2022.04 | — | 0.737 | — | — | — | — | — | |
| AMMC-Netw/ obj.=false, Manually Cropped=true2022.09 | — | — | 0.737 | — | — | — | — | |
| AMSRCMethod Category=Proposed2022.04 | — | 0.763 | — | — | — | — | — | |
| AMSRCYear=20232023.09 | — | — | — | — | — | — | 0.766 | |
| AnoPCNMethod Category=Hybrid and Other Methods2022.04 | — | 0.736 | — | — | — | — | — | |
| BA-AED2024.03 | — | — | — | — | 41.3 | 78.8 | — | |
| BA-AED2024.03 | — | — | 0.827 | 0.893 | — | — | — | |
| BA-AED + SSPCAB2024.03 | — | — | 0.836 | 0.895 | — | — | — | |
| Background-Agnostic FrameworkYear=20202023.09 | — | — | — | — | — | — | 0.827 | |
| Background-Agnostic Framework + SSPCABYear=20222023.09 | — | — | — | — | — | — | 0.836 | |
| BAFw/ obj.=true, Manually Cropped=true2022.09 | — | — | 0.827 | 89.3 | — | — | — | |
| Bi-Directional VAD FrameworkYear=20222023.09 | — | — | — | — | — | — | 0.781 | |
| BMANw/ obj.=false, Manually Cropped=true2022.09 | — | — | 0.762 | — | — | — | — | |
| CACw/ obj.=false, Manually Cropped=true2022.09 | — | — | 0.793 | — | — | — | — | |
| CAC2024.03 | — | — | 0.793 | — | — | — | — | |
| CAE-SVM2024.03 | — | — | — | — | 20.6 | 44.5 | — | |
| CAE-SVM2024.03 | — | — | 0.787 | 0.849 | — | — | — | |
| Cloze TestYear=20202023.09 | — | — | — | — | — | — | 0.748 | |
| CMRLProtocol=WVAD, Data type (Nor/Abn)=Real/Real2025.12 | — | — | — | — | — | — | 0.976 | |
| CT-D2GAN2024.03 | — | — | 0.777 | — | — | — | — | |
| DMADYear=20232023.09 | — | — | — | — | — | — | 0.788 | |
| EVALYear=20232023.09 | — | — | — | — | — | — | 0.7663 | |
| F2PNw/ obj.=false, Manually Cropped=true2022.09 | — | — | 0.73 | — | — | — | — | |
| FFPw/ obj.=false, Manually Cropped=true2022.09 | — | — | 0.728 | — | — | — | — | |
| FPDM2024.03 | — | — | 0.786 | — | — | — | — | |
| Frame-Pred.Method Category=Prediction-Based Methods2022.04 | — | 0.728 | — | — | — | — | — | |
| Frame-Pred.2024.03 | — | — | — | — | 17 | 54.2 | — | |
| GEPCw/ obj.=false, Manually Cropped=true2022.09 | — | — | 0.761 | — | — | — | — | |
| GEPCSetting=full-shot, Training on SHT=True2025.09 | — | — | — | — | — | — | 0.761 | |
| HF22024.03 | — | — | 0.762 | — | — | — | — | |
| HF2-VADMethod Category=Hybrid and Other Methods2022.04 | — | 0.762 | — | — | — | — | — | |
| HF2-VADw/ obj.=true, Manually Cropped=true2022.09 | — | — | 0.762 | — | — | — | — |