Research map / Video / Video segmentation and tracking
Weakly supervised video anomaly detection
Detects anomalous events in surveillance video using weak labels, memory units and event completeness.
99 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Video segmentation and tracking, keyword cluster “detection · anomaly · supervised”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 2 of 6 tracked benchmarks.
31 papers in 2023–24 and 42 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 7.0% to 7.8%.
Open questions the data points to
- UCF-Crime (AUC (%)): best went from 86.98 (up to 2024) to 90.33 in 2025 — Just Dance with pi! A Poly-modal Inductor for Weakly-supervised Video Anomaly Detection (CVPR 2025 · 5 citations)
- XD-Violence (AP (%)): best went from 81.66 (up to 2024) to 88.47 in 2025 — Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · 0 citations)
- Papers report RTFM on XD-Violence (AP (%)) anywhere from 58.99 to 77.95: comparisons against it depend on whose number you take.
What looks settled
- Avenue (AUC (%)): best up to 2024 93.7, best in 2025–26 92.9 from 2 papers, at most 0.2% better, if at all.
No longer reported by papers here: CUHK Avenue (AUC, last in 2023).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Dynamic Concept Composition for Zero-Example Event Detection (AAAI 2016 · 44 citations)
- 2018 · Real-World Anomaly Detection in Surveillance Videos (CVPR 2018 · 2,159 citations)
- 2019 · Learning Regularity in Skeleton Trajectories for Anomaly Detection in Videos (CVPR 2019 · 340 citations)
- 2020 · Learning Memory-Guided Normality for Anomaly Detection (CVPR 2020 · 933 citations)
- 2021 · MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection (CVPR 2021 · 349 citations)
- 2022 · Self-Training Multi-Sequence Learning with Transformer for Weakly Supervised Video Anomaly Detection (AAAI 2022 · 279 citations)
- 2023 · MGFN: Magnitude-Contrastive Glance-and-Focus Network for Weakly-Supervised Video Anomaly Detection (AAAI 2023 · 189 citations)
- 2024 · Open-Vocabulary Video Anomaly Detection (CVPR 2024 · 90 citations)
- 2025 · Video Anomaly Detection with Motion and Appearance Guided Patch Diffusion Model (AAAI 2025 · 16 citations)
- 2026 · VAGU & GtS: LLM-Based Benchmark and Framework for Joint Video Anomaly Grounding and Understanding (AAAI 2026 · 2 citations)
Earliest papers here
- Dynamic Concept Composition for Zero-Example Event Detection (AAAI 2016 · 44 citations)
- They Are Not Equally Reliable: Semantic Event Search Using Differentiated Concept Classifiers (CVPR 2016 · 35 citations)
- Robust Complex Behaviour Modeling at 90Hz (AAAI 2016 · 0 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
ShanghaiTech · AUC (%) (higher is better)
12 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 70.5 | Learning Memory-Guided Normality for Anomaly Detection (CVPR 2020 · 933 citations) |
| 2021 | 94.83 | MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection (CVPR 2021 · 349 citations) |
| 2022 | 97.32 | Self-Training Multi-Sequence Learning with Transformer for Weakly Supervised Video Anomaly Detection (AAAI 2022 · 279 citations) |
| 2023 | 97.6 | Look Around for Anomalies: Weakly-Supervised Anomaly Detection via Context-Motion Relational Learning (CVPR 2023 · 77 citations) |
UCF-Crime · AUC (%) (higher is better)
12 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 82.3 | MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection (CVPR 2021 · 349 citations) |
| 2022 | 85.62 | Self-Training Multi-Sequence Learning with Transformer for Weakly Supervised Video Anomaly Detection (AAAI 2022 · 279 citations) |
| 2023 | 86.98 | MGFN: Magnitude-Contrastive Glance-and-Focus Network for Weakly-Supervised Video Anomaly Detection (AAAI 2023 · 189 citations) |
| 2025 | 90.33 | Just Dance with pi! A Poly-modal Inductor for Weakly-supervised Video Anomaly Detection (CVPR 2025 · 5 citations) |
CUHK Avenue · AUC (higher is better)
7 papers, 2018–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 85.1 | Future Frame Prediction for Anomaly Detection – A New Baseline (CVPR 2018 · 1,540 citations) |
| 2020 | 88.5 | Learning Memory-Guided Normality for Anomaly Detection (CVPR 2020 · 933 citations) |
| 2021 | 89.5 | Learning Normal Dynamics in Videos With Meta Prototype Network (CVPR 2021 · 222 citations) |
| 2022 | 90.3 | Comprehensive Regularization in a Bi-directional Predictive Network for Video Anomaly Detection (AAAI 2022 · 76 citations) |
XD-Violence · AP (%) (higher is better)
7 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 78.59 | Self-Training Multi-Sequence Learning with Transformer for Weakly Supervised Video Anomaly Detection (AAAI 2022 · 279 citations) |
| 2023 | 81.66 | Dual Memory Units with Uncertainty Regulation for Weakly Supervised Video Anomaly Detection (AAAI 2023 · 165 citations) |
| 2025 | 88.47 | Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · 0 citations) |
UCSD Ped2 · AUC (higher is better)
6 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 97 | Learning Memory-Guided Normality for Anomaly Detection (CVPR 2020 · 933 citations) |
| 2022 | 98.3 | Comprehensive Regularization in a Bi-directional Predictive Network for Video Anomaly Detection (AAAI 2022 · 76 citations) |
| 2025 | 99.1 | Noise-Resistant Video Anomaly Detection via RGB Error-Guided Multiscale Predictive Coding and Dynamic Memory (CVPR 2025 · 8 citations) |
Avenue · AUC (%) (higher is better)
5 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 80.87 | Bayesian Nonparametric Submodular Video Partition for Robust Anomaly Detection (CVPR 2022 · 69 citations) |
| 2023 | 93.7 | Hierarchical Semantic Contrast for Scene-Aware Video Anomaly Detection (CVPR 2023 · 95 citations) |
What has been tried
71 papers here report 813 measured changes (a component added, removed or swapped, compared with a baseline): helped 79%, no clear effect 3%, hurt 15%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add the temporal adapter (TA) | UCF-Crime · AUC | 84.79 -> 85.14 | Helped | Open-Vocabulary Video Anomaly Detection (CVPR 2024 · 90 citations) |
| add motion-gradient token weights | Avenue · micro AUC (%) | 84.0 -> 84.8 | Helped | Self-Distilled Masked Auto-Encoders are Efficient Video Anomaly Detectors (CVPR 2024 · 64 citations) |
| Add Scale-Aware Prediction Head | XD-Violence · AP | 78.73 -> 80.82 | Helped | Prompt-Enhanced Multiple Instance Learning for Weakly Supervised Video Anomaly Detection (CVPR 2024 · 44 citations) |
| Use all three proxy tasks jointly | ShanghaiTech · Micro AUC | 70.2 -> 85.1 | Helped | Multi-Scale Video Anomaly Detection by Multi-Grained Spatio-Temporal Representation Learning (CVPR 2024 · 39 citations) |
| add server knowledge accumulation (SKA) | UCF-Crime · AUC | 76.2 -> 77.1 | Helped | Collaborative Learning of Anomalies with Privacy (CLAP) for Unsupervised Video Anomaly Detection: A New Baseline (CVPR 2024 · 30 citations) |
| Add the log-density regularization term (β=0.1) | ShanghaiTech (frame-centric) · AUC-ROC micro (%) | 78.4 -> 81.3 | Helped | MULDE: Multiscale Log-Density Estimation via Denoising Score Matching for Video Anomaly Detection (CVPR 2024 · 27 citations) |
| Add temporal-spatial injection (TSI) | MoCA-Mask · mIoU | 0.366 -> 0.388 | Helped | Endow SAM with Keen Eyes: Temporal-spatial Prompt Learning for Video Camouflaged Object Detection (CVPR 2024 · 23 citations) |
| add patch diffusion to appearance and motion conditions | Ped2, Avenue, Shanghai, UBnormal · AUC (%) | 96.7 -> 98.6; 88.8 -> 91.3; 76.7 -> 79.2; 57.8 -> 63.4 | Helped | Video Anomaly Detection with Motion and Appearance Guided Patch Diffusion Model (AAAI 2025 · 16 citations) |
| add group-guided text encoding | XD-VIOLENCE anomalous videos · Top-1 accuracy | 53.42 -> 56.07 | Helped | Anomize: Better Open Vocabulary Video Anomaly Detection (CVPR 2025 · 15 citations) |
| Use scenario information in few-shot sampling | ShanghaiTech v1 · Micro AUC | 63.74 -> 68.96 | Helped | Advancing Video Anomaly Detection: A Concise Review and a New Dataset (NeurIPS 2024 · 15 citations) |
| Add FISTA and DAM to the spatial-only CSPDarknet baseline | DAUB · AP50 | 83.59 -> 95.93 | Helped | MOCID: Motion Context and Displacement Information Learning for Moving Infrared Small Target Detection (AAAI 2025 · 14 citations) |
| remove LSF, GSF, and TIM | UCF-Crime-DVS · AUC | 65.01 -> 62.78 | Hurt | UCF-Crime-DVS: A Novel Event-Based Dataset for Video Anomaly Detection with Spiking Neural Networks (AAAI 2025 · 11 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| RTFM | XD-Violence | AP (%) | 58.99 – 77.95 | 2 papers, e.g. CVPR 2025 and AAAI 2022 |
What reviewers will expect
Datasets and baselines named in the 19 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- UCF-Crime · 4 of 19 papers
- XD-Violence · 3 of 19 papers
- UBnormal · 2 of 19 papers
- Avenue · 2 of 19 papers
- ShanghaiTech Campus · 2 of 19 papers
- UCSD Ped2 · 2 of 19 papers
- GenVideo · 2 of 19 papers
- Kinetics-400 · 2 of 19 papers
Baselines
- RTFM · 3 of 19 papers
- DeMamba · 3 of 19 papers
- Sultani et al. · 2 of 19 papers
- TSA · 2 of 19 papers
- MNAD · 2 of 19 papers
- FPDM · 2 of 19 papers
- MemAE · 2 of 19 papers
- VadCLIP · 2 of 19 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 73 papers; from OpenAlex and the papers' affiliations).
- Xidian University · 3 papers
- Northwestern Polytechnical University · 3 papers
- Huazhong University of Science and Technology · 2 papers
- The University of Hong Kong · 2 papers
- The Chinese University of Hong Kong · 2 papers
- Chongqing University · 2 papers
- Beijing Jiaotong University · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- City University of Hong Kong · 2 papers
- Wuhan University · 2 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Self-supervised video representation learning (167 papers)
- Video restoration with temporal priors (159 papers)
- Sign language and physiological video analysis (99 papers)
- Spatio-temporal predictive modeling (91 papers)
- Deepfake video detection (47 papers)
Elsewhere in the topic
- Layered video decomposition and dynamics (140 papers)
- Stochastic video prediction (125 papers)
- Physics-inspired motion prediction (113 papers)
- Matching-based video object segmentation (86 papers)
- Learned optical flow estimation (73 papers)
- Learned video compression (68 papers)
- Mask propagation for unsupervised video segmentation (67 papers)
- Video semantic and panoptic segmentation (67 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).