📽️ Awesome Video Anomaly Detection

Video anomaly detection (VAD) aims to identify, understand and describe anomalous events in videos. This repository collects the latest research papers, code, datasets, seminars, utilities and related resources for VAD, updated every Friday. Like this repo😄? ⭐ it and feel free to open an issue for feedback.

Note

Welcome to nominate VAD papers or related resources via pull request! Please refer to CONTRIBUTING.md for detailed guidelines.

📌 Contents

• 📌 Contents
• 🔥 Recent Updates
• 🧪 New Setting Papers
• 📃 Weakly-supervised VAD Papers
  • Prompt Involved Papers
• 📃 Semi-supervised VAD Papers
• 📃 Skeleton-based Papers
• 📃 Fully-supervised VAD Papers
• 📑 Surveys
• 🏗️ Benchmarks
• 💽 Datasets
  • Links
  • Statistics
• 🧑🏻‍🏫 Seminars
• 👷🏻‍♂️ Evaluation Metrics
• 🛠️ Utilities
• 🔗 Related Repositories
• 🙌🏻 Acknowledgements
• 🔖 Citation

🔥 Recent Updates

Last Update: June, 2025

• ICCV '25
• CVPR '25
• ICML '25
• ICLR '25

Related Repositories:
uws4vad-wiki 🔥🔥🔥
Video-Anomaly-Detection

Papers:
ADSM
SeeKer
ADRM
MA-PDM
TAO
LEC-VAD
π-VAD
SFN-VAD
SVTA
VAU-R1
VERA \

🧪 New Setting Papers

1. [TAO] Track Any Anomalous Object: A Granular Video Anomaly Detection Pipeline    
   CVPR '25 [paper][project][code]

2. [SVTA] Towards Scalable Video Anomaly Retrieval: A Synthetic Video-Text Benchmark    
   
   arXiv '25 [paper][project]

3. [VAU-R1] Advancing Video Anomaly Understanding via Reinforcement Fine-Tuning    
   
   arXiv '25 [paper][code]

4. [VANE-Bench] VANE-Bench: Video Anomaly Evaluation Benchmark for Conversational LMMs
   
   NAACL '25 [paper][code][dataset][project]

5. [Sherlock] Sherlock: Towards Multi-scene Video Abnormal Event Extraction and Localization via a Global-local Spatial-sensitive LLM
   
   WWW '25 [paper][OpenReview]

6. [SurveillanceVQA-589K] SurveillanceVQA-589K: A Benchmark for Comprehensive Surveillance Video-Language Understanding with Large Models
   
   arXiv '25 [paper][annotation]

7. [VERA] VERA: Explainable Video Anomaly Detection via Verbalized Learning of Vision-Language Models    
   
   CVPR '25 [paper][code][project]

8. [Holmes-VAU] Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity
   
   CVPR '25 [paper][code & annotation]

9. [HAWK] HAWK: Learning to Understand Open-World Video Anomalies
   
   NeurIPS '24 [paper][code][annotation]

10. [VAR] Toward Video Anomaly Retrieval From Video Anomaly Detection: New Benchmarks and Model
    
    TIP '24 [paper][dataset]

11. [AnomalyRuler] Follow the Rules: Reasoning for Video Anomaly Detection with Large Language Models
    
    ECCV '24 [paper][code]

12. [UCA] Towards Surveillance Video-and-Language Understanding: New Dataset, Baselines, and Challenges
    
    CVPR '24 [paper][code & annotation][project]

13. [GlanceVAD] GlanceVAD: Exploring Glance Supervision for Label-efficient Video Anomaly Detection
    
    ICME '25 [paper][code & annotation]

14. [OVVAD] Open-Vocabulary Video Anomaly Detection
    
    CVPR '24 [paper][supp]

15. [LAVAD] Harnessing Large Language Models for Training-free Video Anomaly Detection
    
    CVPR '24 [paper][code][supp]

16. [CUVA] Uncovering What, Why and How: A Comprehensive Benchmark for Causation Understanding of Video Anomaly
    
    CVPR '24 [paper][code & dataset][supp]

17. [UCF-Crime-DVS] UCF-Crime-DVS: A Novel Event-Based Dataset for Video Anomaly Detection with Spiking Neural
    
    [paper][code & dataset]

18. [TDSD] TDSD: Text-Driven Scene-Decoupled Weakly Supervised Video Anomaly Detection
    
    ACM MM '24 [paper][code][OpenReview]

📃 Weakly-supervised VAD Papers

1. [ADRM] Dual‑Detector Reoptimization for Federated Weakly Supervised Video Anomaly Detection via Adaptive Dynamic Recursive Mapping    
   
   TII '25 [paper] [code] [supp]

2. [CDL] Cross-Domain Learning for Video Anomaly Detection with Limited Supervision
   
   ECCV '24 [paper]

3. [HLGAtt] Cross-Modal Fusion and Attention Mechanism for Weakly Supervised Video Anomaly Detection
   
   CVPR '24 Workshop [paper]

4. [ECU] Exploiting Completeness and Uncertainty of Pseudo Labels for Weakly Supervised Video Anomaly Detection
   
   CVPR '23 [paper][code][supp]

5. [CoMo] Look Around for Anomalies: Weakly-supervised Anomaly Detection via Context-Motion Relational Learning
   
   CVPR '23 [paper][supp]

6. [ADGCN] Graph Convolutional Label Noise Cleaner: Train a Plug-and-play Action Classifier for Anomaly Detection
   CVPR 19' [paper][code]

Prompt Involved Papers

1. [π-VAD] Just Dance with $\pi$! A Poly-modal Inductor for Weakly-supervised Video Anomaly Detection    
   
   CVPR '25 [paper]

2. [LEC-VAD] Learning Event Completeness for Weakly Supervised Video Anomaly Detection    
   
   ICML '25 [paper]

3. [Fed-WSVAD] Federated Weakly Supervised video Anomaly Detection with Multimodal Prompt
   
   AAAI '25 [paper][code]

4. [STPrompt] Weakly Supervised Video Anomaly Detection and Localization with Spatio-Temporal Prompts
   
   ACM MM '24 [paper][OpenReview]

5. [Vadclip] Vadclip: Adapting vision-language models for weakly supervised video anomaly detection
   
   AAAI '24 [paper][code]

6. [PE-MIL] Prompt-Enhanced Multiple Instance Learning for Weakly Supervised Video Anomaly Detection
   
   CVPR '24 [paper][code][supp]

7. [TPWNG] Text Prompt with Normality Guidance for Weakly Supervised Video Anomaly Detection
   
   CVPR '24 [paper][supp]

8. [PEL4VAD] Learning Prompt-Enhanced Context features for Weakly-Supervised Video Anomaly Detection
   
   TIP '24 [paper][code]

📃 Semi-supervised VAD Papers

1. [ADSM] Autoregressive Denoising Score Matching is a Good Video Anomaly Detector    
   ICCV '25 [paper][code]

2. [MA-PDM] Video Anomaly Detection with Motion and Appearance Guided Patch Diffusion Model    
   AAAI '25 [paper][code]

3. [SFN-VAD] MemoryOut: Learning Principal Features via Multimodal Sparse Filtering Network for Semi-supervised Video Anomaly Detection    
   arXiv '25 [paper][project]

4. [LPGB] Local Patterns Generalize Better for Novel Anomalies
   ICLR '25 [paper][code][OpenReview]

5. [LANP] Learning Anomalies with Normality Prior for Unsupervised Video Anomaly Detection
   
   ECCV '24 [paper]

6. [Joint-VAD] Interleaving One-Class and Weakly-Supervised Models with Adaptive Thresholding for Unsupervised Video Anomaly Detection
   
   ECCV '24 [paper][code]

7. [SSAE] Scene-Dependent Prediction in Latent Space for Video Anomaly Detection and Anticipation
   T-PAMI '24[paper][project][code][dataset]

8. [DoTA] DoTA: Unsupervised Detection of Traffic Anomaly in Driving Videos
   T-PAMI '23 [paper][code][dataset]

9. [AED-MAE] Self-Distilled Masked Auto-Encoders are Efficient Video Anomaly Detectors
   CVPR '24 [paper][code][supp]

10. [MSTL] Multi-Scale Video Anomaly Detection by Multi-Grained Spatio-Temporal Representation Learning
    
    CVPR '24 [paper][supp]

11. [MULDE] MULDE: Multiscale Log-Density Estimation via Denoising Score Matching for Video Anomaly Detection
    
    CVPR '24 [paper][code][supp]

12. [CLAP] Collaborative Learning of Anomalies with Privacy (CLAP) for Unsupervised Video Anomaly Detection: A New Baseline
    CVPR '24 [paper][code][supp]

13. [MGENet] A Multilevel Guidance-Exploration Network and Behavior-Scene Matching Method for Human Behavior Anomaly Detection
    
    ACM MM '24 [paper][code][OpenReview]

14. [MPT] Video Anomaly Detection via Progressive Learning of Multiple Proxy Tasks
    ACM MM '24 [paper][OpenReview]

📃 Skeleton-based Papers

1. [SeeKer] Sequential keypoint density estimator:an overlooked baseline of skeleton-based video anomaly detection   
   ICCV '25 [paper][code]

2. [GiCiSAD] Graph-Jigsaw Conditioned Diffusion Model for Skeleton-based Video Anomaly Detection
   WACV '25 [paper][code]

3. [PoseWatch] PoseWatch: A Transformer-based Architecture for Human-centric Video Anomaly Detection Using Spatio-temporal Pose Tokenization
   arXiv '25 [paper][code]

4. [DA-Flow] DA-Flow: Dual Attention Normalizing Flow for Skeleton-based Video Anomaly Detection
   arXiv '24 [paper]

5. [MoCoDAD] Multimodal Motion Conditioned Diffusion Model for Skeleton-based Video Anomaly Detection
   ICCV '23 [paper][code][supp]

📃 Fully-supervised VAD Papers

1. [EBB] Exploring Background-bias for Anomaly Detection in Surveillance Videos
   ACM MM 19' [paper][annotation]

2. [ALVS] ANOMALY LOCALITY IN VIDEO SURVEILLANCE
   arXiv 19' [paper][project][annotation]

📑 Surveys

1. [MSAD] Advancing Video Anomaly Detection: A Concise Review and a New Dataset
   NeurIPS '24 [paper][project]

2. [BenchRev] Unveiling the performance of video anomaly detection models — A benchmark-based review
   Intelligent Systems with Applications '23 [paper]

3. [WVAD-Review] Weakly Supervised Anomaly Detection: A Survey
   arXiv '23 [paper][repo]

4. [VAD-10] Video Anomaly Detection in 10 Years: A Survey and Outlook
   arXiv '24 [paper]

5. [GNN4TS] A Survey on Graph Neural Networks for Time Series: Forecasting, Classification, Imputation, and Anomaly Detection
   T-PAMI '24 [paper][repo]

6. [GTCNN] Graph-Time Convolutional Neural Networks: Architecture and Theoretical Analysis
   T-PAMI '23 [paper]

🏗️ Benchmarks

Supervision	Method	Publication	Visual Features	STC (AUC)	UCF (AUC)	XDV (AP)	Ave (AUC)	Cor (AUC)	UBnormal (AUC)	Ped2 (AUC)	Campus (AUC)	NWPU (AUC)	NPDI (AUC)	TAD (AUC)	Audio Features	Text Prompt	LLM Involved	Data Augumentation
Semi-supervised	LANP	ECCV '24	I3D	88.32	80.02	-	-	-	-	-	-	-	-	-	-	-	-	-
			ResNext	86.46	76.64	-	-	-	-	-	-	-	-	-	-	-	-	-
	SSAE	T-PAMI '24	-	80.5	-	-	90.2	75.8	-	-	-	75.6	-	-	-	-	-	-
	AED-MAE	CVPR '24	-	79.1	-	-	91.3	-	58.5	95.4	-	-	-	-	-	-	-	-
	MSTL	CVPR '24	I3D	87.5	80.6	-	94.3	-	-	-	70.1	-	-	-	-	-	-	-
	MULDE	CVPR '24	CLIP+Hiera	81.3	78.5	-	-	-	72.8	-	-	-	-	-	-	-	-	-
	MGEnet	MM '24	Video Swin	86.9	-	-	-	-	74.3	-	-	-	-	-	-	-	-	-
	MPT	MM '24	-	88.6	83.2	-	94.5	-	-	-	-	-	-	-	-	-	-	-
Weakly-supervised	HLGAtt	CVPR '24 Workshop	I3D	-	-	86.34	-	-	-	-	-	-	99.45	-	VGGish	-	-	-
	STPrompt	MM '24	CLIP	97.81	88.08	-	-	-	63.98	-	-	-	-	-	-	✓	-	-
	ECU	CVPR '23	I3D	-	86.22	81.43	-	-	-	-	-	-	-	91.66	VGGish	-	-	-
	CoMo	CVPR '23	I3D	97.6	86.1	81.3	89.8	-	-	-	-	-	-	-	-	-	-	-
	Vadclip	AAAI '24	CLIP	-	88.02	84.51	-	-	-	-	-	-	-	-	-	✓	-	-
	PE-MIL	CVPR '24	I3D	98.35	86.83	88.21	-	-	-	-	-	-	-	-	VGGish	✓	-	-
	ADRM	TII '25	MAEv2	97.91/97.86(Fed)	-	-	-	-	70.91/76.51(Fed)	-	-	-	-	-	-	-	-	-

💽 Datasets

Links

Dataset	Download Links	Features	Additional Annotations	Publication
ShanghaiTech Campus	BaiduYun	I3D	-	CVPR 18'
UCF-Crime	Dropbox	I3D	Full Annotation & Bounding Box Annotation	CVPR 18'
XD-Violence	OneDrive	I3D & VGGish	-	ECCV 20'
NWPU Campus	BaiduYun & Google Drive	-	-	CVPR '23 & T-PAMI '24
UBnormal	GoogleDrive	-	☑	CVPR '22
ECVA	ModelScope	-	-	CVPR '24
MSAD	Request Form	Swin & I3D	-	NeurIPS '24
TAD	GoogleDrive	-	-	TIP '21
NPDI	Contact Email	-	-	CVIU 13'
Street Scene	Page	-	-	WACV '20
Pornography-2k	Contact Email	-	-	FSI 16'

Statistics

Dataset	Citations	Year	Source	Domain	Modality	#Videos	Supervision	#Training Abnormal Videos	#Training Normal Videos	#Test Abnormal Videos	#Test Normal Videos	#Validation Normal Videos	#Validation Normal Videos	#Anomaly Types
ShanghaiTech		2018	Surveillance	Campus	Visual	437	Semi	-	330	107	-	-	-	13
							Weakly*	63	175	44	155
UCF-Crime		2018	Surveillance	Crime	Visual	1900	Semi†	-	800	140	150	-	-	13
							Weakly	810	800	140	150
NWPU Campus		2023	Surveillance	Campus	Visual	547	Semi	-	305	124	118	-	-	28
							Weakly‡	361		213
XD-Violence		2020	Film & Online	Multiple	Visual & Audio	4754	Weakly	1905	2049	500	300	-	-	7
UBnormal		2022	Virtual	Pedestrian	Visual	543	Fully	82	186	158	53	38	26	22
ECVA		2024	Online	Multiple	Visual & Audio & Text	2240	-	-	-	-	-	-	-	100
MSAD		2024	Surveillance	Multiple	Visual	720	Semi	-	360	240	120	-	-	55
							Weakly	120	360	120	120
TAD		2021	Surveillance	Traffic	Visual	500	Weakly	-	-	-	-	-	-	7
NDPI		2021	Online	Nudity	Visual	800	Weakly	-	-	-	-	-	-	1
Pornography-2k		2021	Online	Nudity	Visual	2000	Weakly	-	-	-	-	-	-	1

# : denotes "number of".
* : ShanghaiTech was initially proposed as a semi-supervised VAD dataset, and Zhong etal. later introduced its weakly supervised split.
†: UCF-Crime was originally introduced as a weakly-supervised VAD dataset, and the dataset split under semi-supervision are sourced from MULDE.
‡: NWPU Campus was initially proposed as a semi-supervised VAD dataset, and TDSD later introduced its weakly supervised split.

🧑🏻‍🏫 Seminars

1. Recent advances in anomaly detection.
   CVPR '23 Tutorial [link]

👷🏻‍♂️ Evaluation Metrics

1. AUC

Area Under Curve (AUC) of Receiver-Operating Characteristic curve (ROC) is a primary evaluation metric for VAD, measuring classification performance across all thresholds. The ROC curve plots the True Positive Rate (TPR) on the y-axis against the False Positive Rate (FPR) on the x-axis at various thresholds. AUC is computed as the integral of the ROC curve, ranging from 0 to 1, where 0.5 indicates performance approximating random guessing, and 1 denotes perfect discrimination. A higher AUC value indicates superior performance. AUC is a commonly used evaluation metric for the ShanghaiTech Campus and UCF-Crime datasets .

2. AP

Average precision (AP) summarizes a Precision-Recall (PR) curve into a single value representing the average of all precisions. It is generally understood as the approximation of the area under the PR curve. AP ranges between 0 and 1, where a perfect model has precision, recall, and AP scores of 1. The larger the metric, the better a model performs across different thresholds^[*].

3. FAR

False Alarm Rate (FAR) with a threshold of 0.5 is evaluated to measure the reliability of detection results, proposed in UCF-Crime.

🛠️ Utilities

[Video & Audio Feature Extraction]: video_features allows you to extract features from video clips, supporting a variety of modalities and extractors, i.e., S3D, R(2+1)d RGB, I3D-Net RGB + Flow, VGGish, CLIP.

🔗 Related Repositories

uws4vad-wiki: a comprehensive benchmark table for VAD datasets and methods, frequently updated.    

Video-Anomaly-Detection: a curated list of video anomaly detection papers.    

awesome-video-anomaly-detection: an awesome collection of papers and codes for video anomaly detection, updated to CVPR '22.

WSAD: a comprehensive collection and categorization of weakly supervised anomaly detection papers.

awesome anomaly detection: a curated list of awesome anomaly detection resources, including time-series anomaly detection, video-level anomaly detection, image-level anomaly detection, last updated in November 2021.

anomaly detection resources: a comprehensive resource for anomaly detection, featuring a wide range of papers on various domains, e.g., image, time-series, financial, and social media anomaly detection. It contains only a subset of materials specifically related to video anomaly detection.

🙌🏻 Acknowledgements

Special thanks to [@Zuble] for his valuable help and contributions to this project.
Thanks to [@inaomIIsfarell] | [sky836] | [rekkles2] for their nomination of VAD papers.

🔖 Citation

If you find this repository useful, please consider citing it:

@misc{awesome-vad,
    author = {Junxi Chen},
    title = {Awesome Video Anomaly Detection.},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/Junxi-Chen/Awesome-Video-Anomaly-Detection}},
    year = {2024},
}