Pith. sign in

REVIEW 4 major objections 6 minor 64 references

Autoregressive Denoising Score Matching is a Good Video Anomaly Detector

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Autoregressive denoising score matching turns a score network into a state-of-the-art video anomaly detector.

desk verdict A solid empirical VAD paper with a genuinely new iterative scoring mechanism; the theory is loose but the method works and deserves referee time. read the letter →

arxiv 2506.23282 v1 pith:MVV2R3TT submitted 2025-06-29 cs.CV

Hanwen Zhang , Congqi Cao , Qinyi Lv , Lingtong Min , Yanning Zhang This is my paper · ORCID
classification cs.CV
keywords videoanomalydetectiondenoisingscorematchingscore-basedgenerativemodeldiffusiontransformerautoregressiveinferencescene-dependentmotionweightingPSNR-basedscoring
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that the raw gradient of a learned log-density, normally a weak anomaly signal because it vanishes at local modes, can become a strong video anomaly detector if it is used autoregressively at inference. Instead of scoring a single denoising step, the method repeatedly adds intensifying Gaussian noise to its own denoised estimate, re-estimates the score, and divides the score norm by the PSNR between the denoised and original frames. The paper argues that this accumulates abnormal context across noise levels and suppresses the local-mode blind spots of likelihood-based scores. On Avenue, ShanghaiTech, and NWPU Campus, the resulting indicator sets state-of-the-art micro and macro AUCs, including a 93.2 macro AUC on ShanghaiTech.

What carries the argument

The carrying object is the noise-conditioned score transformer (NCST): a diffusion-transformer-style network that outputs the score $\nabla_{\tilde{x}} \log q_\sigma(\tilde{x})$ for patchified video frames at every noise level $\sigma_i$. It is trained by a patch-wise, motion-weighted denoising score matching objective, with scene class labels injected through adaptive layer normalization. At inference, the autoregressive denoising score matching loop of Algorithm 1 takes the denoised output from each level, re-noises it, re-scores it, and forms the ratio $\|s_\theta\| / \mathrm{PSNR}$ against the original frame; the loop is what turns the score into an accumulated anomaly signal.

What would settle it

Feed the trained model a video whose content is normal but whose rendering is out of distribution, for example clean normal frames with mild Gaussian blur or JPEG compression artifacts added after training. If those frames produce anomaly scores as high as real abnormal events, the indicator is measuring generic denoising failure, not accumulated abnormal context, and the central claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the anomaly indicator $\mathrm{score}_i(t) = \|s_\theta(\tilde{x}_t,\sigma_i)\| / \mathrm{PSNR}(\hat{x}_t,x_t)$, computed inside Algorithm 1 by setting $\tilde{x}_t = \dot{x}_t + \sigma_i\epsilon$ and $\hat{x}_t = \tilde{x}_t + \sigma_i^2 s_\theta(\tilde{x}_t,\sigma_i)$, then replacing $\dot{x}_t$ with $\hat{x}_t$ for the next level, is a better detector than the score norm alone. The score network is trained with denoising score matching on raw video patches, conditioned on scene identity and weighted by key-frame motion differences; the autoregressive loop is what the paper credits with accumulating the abnormal context and suppressing local modes. The paper reports that this design outperforms reconstruction-, prediction-, and likelihood-based counterparts on all three benchmarks, with the largest margin on ShanghaiTech's macro AUC.

Load-bearing premise

The load-bearing premise is that for normal frames the trained network's denoised output stays close to the original frame, so anomalies appear as a score-norm spike or a PSNR drop rather than as a generic failure to denoise any unfamiliar input.

Editorial extensions

If this is right

  • If ADSM is right, raw-pixel score matching can outperform methods that rely on pretrained latent features, since the latent-space version of the same model drops substantially on ShanghaiTech and NWPU Campus.
  • The motion-weighting and scene-conditioning ablations each add several macro-AUC points, so the score itself benefits from video-specific structure rather than only from generative mode coverage.
  • The PSNR aggregation is the largest single ablation gain, implying that combining a likelihood signal with an appearance-reconstruction signal is complementary at inference.
  • The 130M-parameter NCST scores an 8-frame sequence in under 20 milliseconds, keeping the method compatible with real-time surveillance once object tracking is included.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same autoregressive re-noising trick may transfer to other out-of-distribution detection settings where the score norm alone is known to be blind near local modes; that is a testable extension the paper does not run.
  • The repeated $\hat{x}_t \to \tilde{x}_t \to \hat{x}_t$ loop resembles a short reverse-diffusion trajectory, so the accumulated score could be interpreted as a cheap proxy for how far the iterated denoiser drifts from the original frame; this drift interpretation is ours, not stated in the paper.
  • Because the indicator flags any PSNR drop relative to the original frame, one should expect it to be sensitive to non-semantic corruptions such as compression noise or sensor artifacts in normal footage; this would be worth measuring before deployment.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes ADSM, a video anomaly detection method that trains a noise-conditioned score transformer (NCST) with scene-conditioned and motion-weighted denoising score matching, then applies an autoregressive re-noising procedure at inference that combines the score norm with a PSNR-based reconstruction fidelity term. Experiments on Avenue, ShanghaiTech, and NWPU Campus report state-of-the-art micro and macro AUCs, and the code is released.

Significance. If the results are reproducible, the paper makes a strong empirical contribution: it introduces the first score-based transformer for video anomaly detection and demonstrates consistent SOTA gains across three benchmarks, including the large-scale NWPU Campus dataset, with both micro and macro AUCs. The reproduction of likelihood-based baselines (MSMA, MULDE) under the same settings in Table 3 and the component-wise ablations in Table 4 are valuable additions. However, the central claim that the score-matching mechanism itself, rather than the PSNR reconstruction term or noise injection, drives the gains requires additional evidence and clarification.

major comments (4)
  1. [Sec. 3.2, Sec. 4.1, Algorithm 1] The noise schedule is described inconsistently. Section 3.2 defines {sigma_i} as a geometric sequence satisfying sigma_L / sigma_{L-1} = ... > 1, while Section 4.1 says training samples sigma from a log-uniform distribution on [0.001, 1.0] and evaluation uses L = 20 evenly spaced noise levels. Algorithm 1 uses sigma_i without specifying the sequence. These choices lead to different score objectives and different inference dynamics. The paper should state the exact schedule used for training and inference, justify it, and report sensitivity to the number of levels L.
  2. [Eq. (8), Algorithm 1, Eq. (10)] The motion-weighted DSM objective in Eq. (8) means s_theta is not an unbiased estimator of the score of the normal-data distribution, because the weights omega_j depend on the input x. Yet Algorithm 1 applies the Tweedie-style update x_hat = x_tilde + sigma_i^2 s_theta(x_tilde, sigma_i), and Eq. (10) uses PSNR(x_hat, x_t) as a normality measure. For low-motion patches, which are downweighted in training, the denoising update may be biased, so the PSNR term could reflect denoising bias rather than anomaly content. The paper should provide an ablation that isolates the score-norm term from the PSNR term, and ideally show that s_theta trained with Eq. (8) still yields accurate denoising on normal validation frames.
  3. [Table 4, Sec. 4.4] The textual description of the ablations does not match the reported numbers. The text claims that the appearance term gives the largest gain (5.7% micro, 5.8% macro), but the adjacent comparison that yields roughly this gain (row 4 to row 6) also includes the motion component; the isolated appearance gain (row 5 to row 6) is 2.4% micro and 3.3% macro. Please clarify the exact component configuration of each row and re-report the incremental contributions consistently.
  4. [Algorithm 1, Sec. 3.5] The claim that autoregressive re-noising 'accumulates the abnormal context' is not substantiated. Feeding the denoised output back into the noise-adding process may also accumulate denoising errors, especially for out-of-distribution inputs. The ablation in Table 4 (rows 1 vs 2) shows an empirical improvement, but it does not identify whether the benefit arises from context accumulation, from ensembling over multiple noise levels, or from implicit multi-scale reconstruction. A controlled analysis (e.g., comparing feed-forward multi-noise scoring with the autoregressive version, and tracking error accumulation on normal vs anomalous sequences) would support the mechanism explanation.
minor comments (6)
  1. [Sec. 4.2] The statement 'the qualitative analysis in Sec. 3.5 confirms the validity' is a cross-reference error; Section 3.5 is the method description, not a qualitative analysis.
  2. [Fig. 3 caption] The abbreviation definitions are misordered: 'MLP, MHA, and FFP stand for the multi-layer perception layer, the feed-forward neural network, and the multi-head attention layer, respectively' should map MLP to multi-layer perceptron, MHA to multi-head attention, and FFN (not FFP) to feed-forward network.
  3. [Table 1, Sec. 4.2] No error bars or significance tests are reported. The claim in Section 4.2 that results fluctuate 'less than 1%' is not backed by any experimental protocol; please report variance over multiple runs or seeds.
  4. [Algorithm 1] The use of x_dot for the re-noised variable is nonstandard and easy to confuse with a time derivative; suggest renaming to something like x_cur.
  5. [Eq. (10)] It is unclear how PSNR is computed for an 8-frame sequence; specify whether it is the average of per-frame PSNRs or computed over the concatenated sequence, and state the pixel-value range used for the MAX constant.
  6. [Table 3] The reproduction details for MSMA and MULDE under 'our settings' are not provided; please include a short description (e.g., same object-cropping, same noise levels, same number of epochs) to make the comparison reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the ADSM anomaly indicator is a constructed heuristic trained by denoising score matching and evaluated on external benchmarks; the self-citations are not load-bearing.

full rationale

The paper's derivation chain is not circular. The anomaly indicator in Eq. (10) is a hand-designed combination of a score norm and a PSNR reconstruction term; the score network is trained with the motion-weighted denoising score-matching objective in Eq. (8), and the update x_hat = x_tilde + sigma_i^2 s_theta is the standard Tweedie denoising step. The autoregressive loop in Algorithm 1 re-noises the previous denoised estimate rather than fitting any test-set parameter, and no fitted value or test label is renamed as a prediction. The benchmark claims in Table 1 are external comparisons against public datasets. The motion-weighted objective does mean s_theta is not exactly the true score of the normal-data distribution, but that is a correctness or robustness concern about whether the PSNR term measures normality, not circularity: the indicator is not defined in terms of its own output or of the reported AUCs. Self-citations to the authors' SSAE [8] and NWPU Campus dataset [6] appear as related work and dataset sources, but the ADSM formulation does not depend on an unverified claim from those papers. No uniqueness theorem is imported from the authors, and no ansatz is smuggled in via self-citation. The self-reported uncertainty about random noise in Sec. 4.2 is a reliability caveat, not a circular step.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No new physical entities or latent quantities are introduced beyond a learned score network and its hyperparameters. The central claim rests on standard score-matching theory plus several dataset- and application-specific heuristic choices, which are listed above.

free parameters (4)
  • Noise level range [sigma_1, sigma_L] = [0.001, 1.0]
    Chosen by hand for training and evaluation; central to the score matching objective and the L iterative steps.
  • Number of noise levels L = 20 at inference (training samples a log-uniform level)
    L controls the number of autoregressive iterations and score accumulations; no sensitivity analysis is given.
  • PSNR pixel value range (MAX) = unspecified
    PSNR is defined only up to a MAX constant, and the choice changes every score_i(t) value; the paper does not state it.
  • Final aggregation weights for Si(t) = unspecified
    The text says 'weigh and average these anomaly scores Si(t)' but gives no weights or formula.
assumptions (5)
  • standard math Score matching objective (Eq. 2) estimates the score of noise-perturbed data
    Standard result from Vincent (2011) and Song & Ermon (2019), used as the training foundation.
  • domain assumption Anomalies are out-of-distribution with low likelihood under the normal data distribution
    The core premise of likelihood-based VAD, stated in Sec. 1 and Fig. 1.
  • domain assumption Scene label y (camera ID) is available for both training and test sequences
    The scene-conditioning component in Sec. 3.3 requires a known scene class for every input clip.
  • ad hoc to paper Score norm with PSNR denominator is a valid anomaly indicator
    Eq. 10 combines score norm and PSNR without a derivation showing this separates normal from abnormal under the learned score.
  • ad hoc to paper Autoregressive re-noising accumulates abnormal context without accumulating error
    Algorithm 1 iterates denoising; the paper asserts this accumulates anomaly context but provides no stability or error analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Autoregressive Denoising Score Matching is a Good Video Anomaly Detector." pith.science (2026). https://pith.science/paper/MVV2R3TT

@misc{pith2026250623282,
  author       = {Pith},
  title        = {Pith review of: Autoregressive Denoising Score Matching is a Good Video Anomaly Detector},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MVV2R3TT}},
  note         = {Machine review of arXiv:2506.23282}
}
read the original abstract

Video anomaly detection (VAD) is an important computer vision problem. Thanks to the mode coverage capabilities of generative models, the likelihood-based paradigm is catching growing interest, as it can model normal distribution and detect out-of-distribution anomalies. However, these likelihood-based methods are blind to the anomalies located in local modes near the learned distribution. To handle these ``unseen" anomalies, we dive into three gaps uniquely existing in VAD regarding scene, motion and appearance. Specifically, we first build a noise-conditioned score transformer for denoising score matching. Then, we introduce a scene-dependent and motion-aware score function by embedding the scene condition of input sequences into our model and assigning motion weights based on the difference between key frames of input sequences. Next, to solve the problem of blindness in principle, we integrate unaffected visual information via a novel autoregressive denoising score matching mechanism for inference. Through autoregressively injecting intensifying Gaussian noise into the denoised data and estimating the corresponding score function, we compare the denoised data with the original data to get a difference and aggregate it with the score function for an enhanced appearance perception and accumulate the abnormal context. With all three gaps considered, we can compute a more comprehensive anomaly indicator. Experiments on three popular VAD benchmarks demonstrate the state-of-the-art performance of our method.

Figures

Figures reproduced from arXiv: 2506.23282 by the authors.

Figure 1
Figure 1. An illustration of the local modes of anomalies. We [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An overview of the proposed method. We implement autoregressive denoising score matching via a novel noise-conditioned [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. A diagram of our NCST block architecture used in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: An overview of the proposed autoregressive denoising [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: A diagram of visualization on NWPU Campus dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 46 canonical work pages

  1. [1]

    Ub- normal: New benchmark for supervised open-set video anomaly detection

    Andra Acsintoae, Andrei Florescu, Mariana-Iuliana Georgescu, Tudor Mare, Paul Sumedrea, Radu Tudor Ionescu, Fahad Shahbaz Khan, and Mubarak Shah. Ub- normal: New benchmark for supervised open-set video anomaly detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 20143–20153, 2022. 6

    work page 2022

  2. [2]

    All are worth words: A vit backbone for diffusion models

    Fan Bao, Shen Nie, Kaiwen Xue, Yue Cao, Chongxuan Li, Hang Su, and Jun Zhu. All are worth words: A vit backbone for diffusion models. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 22669–22679, 2023. 2, 3, 7

    work page 2023

  3. [3]

    One transformer fits all distributions in multi-modal diffu- sion at scale

    Fan Bao, Shen Nie, Kaiwen Xue, Chongxuan Li, Shi Pu, Yaole Wang, Gang Yue, Yue Cao, Hang Su, and Jun Zhu. One transformer fits all distributions in multi-modal diffu- sion at scale. InInternational Conference on Machine Learn- ing, pages 1692–1717. PMLR, 2023. 3, 7

    work page 2023

  4. [4]

    Ssmtl++: Revisiting self- supervised multi-task learning for video anomaly detection

    Antonio Barbalau, Radu Tudor Ionescu, Mariana-Iuliana Georgescu, Jacob Dueholm, Bharathkumar Ramachandra, Kamal Nasrollahi, Fahad Shahbaz Khan, Thomas B Moes- lund, and Mubarak Shah. Ssmtl++: Revisiting self- supervised multi-task learning for video anomaly detection. Computer Vision and Image Understanding , 229:103656,

    work page

  5. [5]

    Appearance-motion memory consistency net- work for video anomaly detection

    Ruichu Cai, Hao Zhang, Wen Liu, Shenghua Gao, and Zhifeng Hao. Appearance-motion memory consistency net- work for video anomaly detection. In Proceedings of the AAAI conference on artificial intelligence , pages 938–946,

    work page

  6. [6]

    A new comprehensive benchmark for semi-supervised video anomaly detection and anticipation

    Congqi Cao, Yue Lu, Peng Wang, and Yanning Zhang. A new comprehensive benchmark for semi-supervised video anomaly detection and anticipation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 20392–20401, 2023. 1, 2, 4, 6

    work page 2023

  7. [7]

    Context recov- ery and knowledge retrieval: A novel two-stream framework for video anomaly detection

    Congqi Cao, Yue Lu, and Yanning Zhang. Context recov- ery and knowledge retrieval: A novel two-stream framework for video anomaly detection. IEEE Transactions on Image Processing, 2024. 1, 2, 6

    work page 2024

  8. [8]

    Scene-dependent prediction in latent space for video anomaly detection and anticipation.IEEE transactions on pattern analysis and machine intelligence, 2024

    Congqi Cao, Hanwen Zhang, Yue Lu, Peng Wang, and Yan- ning Zhang. Scene-dependent prediction in latent space for video anomaly detection and anticipation.IEEE transactions on pattern analysis and machine intelligence, 2024. 1, 2, 4, 5, 6

    work page 2024

Show all 64 references
  1. [9]

    Anomaly detection: A survey

    Varun Chandola, Arindam Banerjee, and Vipin Kumar. Anomaly detection: A survey. ACM computing surveys (CSUR), 41(3):1–58, 2009. 1

    2009

  2. [10]

    Clustering driven deep autoencoder for video anomaly de- tection

    Yunpeng Chang, Zhigang Tu, Wei Xie, and Junsong Yuan. Clustering driven deep autoencoder for video anomaly de- tection. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceed- ings, Part XV 16, pages 329–345. Springer, 2020. 2

    2020

  3. [11]

    Mmtracking: Openmm- lab video perception toolbox and benchmark

    MMTracking Contributors. Mmtracking: Openmm- lab video perception toolbox and benchmark. In https://github.com/open-mmlab/mmtracking, 2020. 6, 8

    2020

  4. [12]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. 3, 4

    2010 arXiv

  5. [13]

    Anomaly detection with bidirectional con- sistency in videos

    Zhiwen Fang, Jiafei Liang, Joey Tianyi Zhou, Yang Xiao, and Feng Yang. Anomaly detection with bidirectional con- sistency in videos. IEEE transactions on neural networks and learning systems, 33(3):1079–1092, 2020. 2

    2020

  6. [14]

    Multimodal motion con- ditioned diffusion model for skeleton-based video anomaly detection

    Alessandro Flaborea, Luca Collorone, Guido Maria D’Amely Di Melendugno, Stefano D’Arrigo, Bardh Prenkaj, and Fabio Galasso. Multimodal motion con- ditioned diffusion model for skeleton-based video anomaly detection. In Proceedings of the IEEE/CVF International Conference on Co...

    2023

  7. [15]

    Anomaly detection in video via self- supervised and multi-task learning

    Mariana-Iuliana Georgescu, Antonio Barbalau, Radu Tu- dor Ionescu, Fahad Shahbaz Khan, Marius Popescu, and Mubarak Shah. Anomaly detection in video via self- supervised and multi-task learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognitio...

    2021

  8. [16]

    A background-agnostic framework with adversarial training for abnormal event detection in video

    Mariana Iuliana Georgescu, Radu Tudor Ionescu, Fa- had Shahbaz Khan, Marius Popescu, and Mubarak Shah. A background-agnostic framework with adversarial training for abnormal event detection in video. IEEE transactions on pattern analysis and machine intelligence, 44(9):4505–4523,

  9. [17]

    Memorizing normality to detect anomaly: Memory-augmented deep autoencoder for unsupervised anomaly detection

    Dong Gong, Lingqiao Liu, Vuong Le, Budhaditya Saha, Moussa Reda Mansour, Svetha Venkatesh, and Anton van den Hengel. Memorizing normality to detect anomaly: Memory-augmented deep autoencoder for unsupervised anomaly detection. In Proceedings of the IEEE/CVF inter- national con...

  10. [18]

    Learning temporal reg- ularity in video sequences

    Mahmudul Hasan, Jonghyun Choi, Jan Neumann, Amit K Roy-Chowdhury, and Larry S Davis. Learning temporal reg- ularity in video sequences. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 733–742, 2016. 1, 2

    2016

  11. [19]

    Normalizing flows for human pose anomaly detection

    Or Hirschorn and Shai Avidan. Normalizing flows for human pose anomaly detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13545– 13554, 2023. 1, 2, 6

    2023

  12. [20]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 1, 2

    2020

  13. [21]

    Estimation of non- normalized statistical models by score matching

    Aapo Hyv ¨arinen and Peter Dayan. Estimation of non- normalized statistical models by score matching. Journal of Machine Learning Research, 6(4), 2005. 3

    2005

  14. [22]

    Object-centric auto-encoders and dummy anomalies for abnormal event detection in video

    Radu Tudor Ionescu, Fahad Shahbaz Khan, Mariana-Iuliana Georgescu, and Ling Shao. Object-centric auto-encoders and dummy anomalies for abnormal event detection in video. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 7842–7851, 20...

    2019

  15. [23]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

    arXiv

  16. [24]

    Variational abnormal be- havior detection with motion consistency

    Jing Li, Qingwang Huang, Yingjun Du, Xiantong Zhen, Shengyong Chen, and Ling Shao. Variational abnormal be- havior detection with motion consistency. IEEE Transac- tions on Image Processing, 31:275–286, 2021. 1, 2, 6

    2021

  17. [25]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceeding...

    2014

  18. [26]

    Key frame extraction from mpeg video stream

    Guozhu Liu and Junming Zhao. Key frame extraction from mpeg video stream. In 2010 Third International Symposium on Information Processing, pages 423–427. IEEE, 2010. 5

    2010

  19. [27]

    A kernelized stein discrepancy for goodness-of-fit tests

    Qiang Liu, Jason Lee, and Michael Jordan. A kernelized stein discrepancy for goodness-of-fit tests. In International conference on machine learning , pages 276–284. PMLR,

  20. [28]

    Fu- ture frame prediction for anomaly detection–a new baseline

    Wen Liu, Weixin Luo, Dongze Lian, and Shenghua Gao. Fu- ture frame prediction for anomaly detection–a new baseline. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 6536–6545, 2018. 1, 2, 6

    2018

  21. [29]

    A hybrid video anomaly detection framework via memory-augmented flow reconstruction and flow-guided frame prediction

    Zhian Liu, Yongwei Nie, Chengjiang Long, Qing Zhang, and Guiqing Li. A hybrid video anomaly detection framework via memory-augmented flow reconstruction and flow-guided frame prediction. In Proceedings of the IEEE/CVF interna- tional conference on computer vision , pages 13588–13597,

  22. [30]

    Abnormal event detec- tion at 150 fps in matlab

    Cewu Lu, Jianping Shi, and Jiaya Jia. Abnormal event detec- tion at 150 fps in matlab. In Proceedings of the IEEE inter- national conference on computer vision , pages 2720–2727,

  23. [31]

    Learnable locality-sensitive hashing for video anomaly de- tection

    Yue Lu, Congqi Cao, Yifan Zhang, and Yanning Zhang. Learnable locality-sensitive hashing for video anomaly de- tection. IEEE Transactions on Circuits and Systems for Video Technology, 33(2):963–976, 2022. 2, 6

    2022

  24. [32]

    A revisit of sparse coding based anomaly detection in stacked rnn framework

    Weixin Luo, Wen Liu, and Shenghua Gao. A revisit of sparse coding based anomaly detection in stacked rnn framework. In Proceedings of the IEEE international conference on com- puter vision, pages 341–349, 2017. 2, 6

    2017

  25. [33]

    Fu- ture frame prediction network for video anomaly detection

    Weixin Luo, Wen Liu, Dongze Lian, and Shenghua Gao. Fu- ture frame prediction network for video anomaly detection. IEEE transactions on pattern analysis and machine intelli- gence, 44(11):7505–7520, 2021. 2

    2021

  26. [34]

    Learning normal dynamics in videos with meta prototype network

    Hui Lv, Chen Chen, Zhen Cui, Chunyan Xu, Yong Li, and Jian Yang. Learning normal dynamics in videos with meta prototype network. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 15425–15434, 2021. 1, 2, 6

    2021

  27. [35]

    Latte: Latent diffusion transformer for video generation

    Xin Ma, Yaohui Wang, Gengyun Jia, Xinyuan Chen, Zi- wei Liu, Yuan-Fang Li, Cunjian Chen, and Yu Qiao. Latte: Latent diffusion transformer for video generation. arXiv preprint arXiv:2401.03048, 2024. 2, 4, 7

    2024 arXiv

  28. [36]

    Multi- scale score matching for out-of-distribution detection

    Ahsan Mahmood, Junier Oliva, and Martin Styner. Multi- scale score matching for out-of-distribution detection. arXiv preprint arXiv:2010.13132, 2020. 1, 7

    2010 arXiv

  29. [37]

    Deep multi-scale video prediction beyond mean square error

    Michael Mathieu, Camille Couprie, and Yann LeCun. Deep multi-scale video prediction beyond mean square error. arXiv preprint arXiv:1511.05440, 2015. 5

    2015 arXiv

  30. [38]

    Mulde: Multiscale log- density estimation via denoising score matching for video anomaly detection

    Jakub Micorek, Horst Possegger, Dominik Narnhofer, Horst Bischof, and Mateusz Kozinski. Mulde: Multiscale log- density estimation via denoising score matching for video anomaly detection. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , ...

    2024

  31. [39]

    Dit-3d: Exploring plain diffusion transformers for 3d shape generation

    Shentong Mo, Enze Xie, Ruihang Chu, Lanqing Hong, Matthias Niessner, and Zhenguo Li. Dit-3d: Exploring plain diffusion transformers for 3d shape generation. Advances in neural information processing systems, 36:67960–67971,

  32. [40]

    Anomaly detec- tion in video sequence with appearance-motion correspon- dence

    Trong-Nguyen Nguyen and Jean Meunier. Anomaly detec- tion in video sequence with appearance-motion correspon- dence. In Proceedings of the IEEE/CVF international con- ference on computer vision, pages 1273–1283, 2019. 1, 2

    2019

  33. [41]

    Learning memory-guided normality for anomaly detection

    Hyunjong Park, Jongyoun Noh, and Bumsub Ham. Learning memory-guided normality for anomaly detection. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14372–14381, 2020. 1, 2, 6

    2020

  34. [42]

    Pytorch: An im- perative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An im- perative style, high-performance deep learning library. Ad- vances in Neural Information Processing Systems,...

    2019

  35. [43]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 4195–4205,

  36. [44]

    A survey of single-scene video anomaly detection

    Bharathkumar Ramachandra, Michael J Jones, and Ranga Raju Vatsavai. A survey of single-scene video anomaly detection. IEEE transactions on pattern analysis and machine intelligence, 44(5):2293–2312, 2020. 1, 4, 5

    2020

  37. [45]

    Attribute-based representa- tions for accurate and interpretable video anomaly detection

    Tal Reiss and Yedid Hoshen. Attribute-based representa- tions for accurate and interpretable video anomaly detection. arXiv preprint arXiv:2212.00789, 2022. 6

    2022 arXiv

  38. [46]

    Self-distilled masked auto-encoders are efficient video anomaly detectors

    Nicolae-C Ristea, Florinel-Alin Croitoru, Radu Tudor Ionescu, Marius Popescu, Fahad Shahbaz Khan, Mubarak Shah, et al. Self-distilled masked auto-encoders are efficient video anomaly detectors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognitio...

    2024

  39. [47]

    Multi-timescale trajectory predic- tion for abnormal human activity detection

    Royston Rodrigues, Neha Bhargava, Rajbabu Velmurugan, and Subhasis Chaudhuri. Multi-timescale trajectory predic- tion for abnormal human activity detection. In Proceedings of the IEEE/CVF winter conference on applications of com- puter vision, pages 2626–2634, 2020. 1, 2, 6

    2020

  40. [48]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 3, 7

    2022

  41. [49]

    Learning normal patterns via adversarial attention-based autoencoder for abnormal event detection in videos

    Hao Song, Che Sun, Xinxiao Wu, Mei Chen, and Yunde Jia. Learning normal patterns via adversarial attention-based autoencoder for abnormal event detection in videos. IEEE Transactions on Multimedia, 22(8):2138–2148, 2019. 2

    2019

  42. [50]

    Generative modeling by esti- mating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by esti- mating gradients of the data distribution. Advances in neural information processing systems, 32, 2019. 1, 2, 3, 4

    2019

  43. [51]

    Score-based generative modeling through stochastic differential equa- tions

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. arXiv preprint arXiv:2011.13456, 2020. 1, 2, 3

    2011 arXiv

  44. [52]

    Hierarchical semantic contrast for scene-aware video anomaly detection

    Shengyang Sun and Xiaojin Gong. Hierarchical semantic contrast for scene-aware video anomaly detection. In Pro- ceedings of the IEEE/cvf conference on computer vision and pattern recognition, pages 22846–22856, 2023. 4

    2023

  45. [53]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017. 2, 4

    2017

  46. [54]

    A connection between score matching and denoising autoencoders

    Pascal Vincent. A connection between score matching and denoising autoencoders. Neural computation, 23(7):1661– 1674, 2011. 3, 4

    2011

  47. [55]

    Video anomaly detection by solving decoupled spatio-temporal jigsaw puzzles

    Guodong Wang, Yunhong Wang, Jie Qin, Dongming Zhang, Xiuguo Bao, and Di Huang. Video anomaly detection by solving decoupled spatio-temporal jigsaw puzzles. In Eu- ropean Conference on Computer Vision , pages 494–511. Springer, 2022. 1, 2

    2022

  48. [56]

    A deep one-class neural network for anomalous event detection in complex scenes

    Peng Wu, Jing Liu, and Fang Shen. A deep one-class neural network for anomalous event detection in complex scenes. IEEE transactions on neural networks and learning systems, 31(7):2609–2622, 2019. 2

    2019

  49. [57]

    Feature prediction diffusion model for video anomaly detection

    Cheng Yan, Shiyu Zhang, Yang Liu, Guansong Pang, and Wenjun Wang. Feature prediction diffusion model for video anomaly detection. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 5527–5537,

  50. [58]

    Video event restoration based on keyframes for video anomaly detection

    Zhiwei Yang, Jing Liu, Zhaoyang Wu, Peng Wu, and Xiaotao Liu. Video event restoration based on keyframes for video anomaly detection. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 14592–14601, 2023. 2, 5

    2023

  51. [59]

    Representation alignment for generation: Training diffu- sion transformers is easier than you think

    Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Representation alignment for generation: Training diffu- sion transformers is easier than you think. arXiv preprint arXiv:2410.06940, 2024. 7

    arXiv

  52. [60]

    Old is gold: Redefining the adversarially learned one-class classifier training paradigm

    Muhammad Zaigham Zaheer, Jin-ha Lee, Marcella Astrid, and Seung-Ik Lee. Old is gold: Redefining the adversarially learned one-class classifier training paradigm. In Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14183–14193, 2020. 1, 2, 6

    2020

  53. [61]

    Bytetrack: Multi-object tracking by associating every detection box

    Yifu Zhang, Peize Sun, Yi Jiang, Dongdong Yu, Fucheng Weng, Zehuan Yuan, Ping Luo, Wenyu Liu, and Xinggang Wang. Bytetrack: Multi-object tracking by associating every detection box. In European conference on computer vision, pages 1–21. Springer, 2022. 6, 8

    2022

  54. [62]

    Graph convolutional label noise cleaner: Train a plug-and-play action classifier for anomaly detection

    Jia-Xing Zhong, Nannan Li, Weijie Kong, Shan Liu, Thomas H Li, and Ge Li. Graph convolutional label noise cleaner: Train a plug-and-play action classifier for anomaly detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1237–1246,

  55. [63]

    Spatial–temporal convolutional neural networks for anomaly detection and localization in crowded scenes

    Shifu Zhou, Wei Shen, Dan Zeng, Mei Fang, Yuanwang Wei, and Zhijiang Zhang. Spatial–temporal convolutional neural networks for anomaly detection and localization in crowded scenes. Signal Processing: Image Communication, 47:358– 368, 2016. 2

    2016

  56. [64]

    Motion-aware feature for improved video anomaly detection

    Yi Zhu and Shawn Newsam. Motion-aware feature for improved video anomaly detection. arXiv preprint arXiv:1907.10211, 2019. 2

    1907 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.