Pith. sign in

REVIEW 4 major objections 5 minor 67 references

Glance, Scrutinize, and Think: Advancing Video Anomaly Detection from Training-Free to Agentic Reasoning

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read The paper reports that a single tool-augmented multimodal model can both explain what anomaly happened and localize when it occurred at 148 frames per second, closing the trade-off that previously split video anomaly detection into two…

desk verdict A serious benchmark and a plausible RL recipe, but the temporal-grounding claim collapses under the paper's own metric. read the letter →

arxiv 2608.11260 v1 pith:WOIWBKY7 submitted 2026-08-07 cs.AI cs.CV

Shibo Gao , Peipei Yang , Xu-Yao Zhang , Linlin Huang This is my paper · ORCID
classification cs.AIcs.CV
keywords VideoAnomalyDetectionUnderstandingTemporalGroundingAgenticReinforcementLearningTool-AugmentedReasoningTraining-FreeFrameworkVAGU-TJeAUG
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

Video anomaly detection is split between systems that know when an anomaly occurs but not what it is, and language-model systems that say what happened but cannot say when. The paper argues this split is a symptom of missing unified reasoning and proposes a global-to-local loop—glance across the whole video, scrutinize suspicious segments, and iterate when the guess is wrong—as the common answer. It realizes the loop twice: a training-free pipeline called Glance then Scrutinize (GtS), which lifts a direct video-QA baseline from a JeAUG score of 2.28 to 4.04, and a tool-augmented agentic model trained by cold-start supervised fine-tuning followed by reinforcement learning, reported at 7.35 on anomaly understanding and 5.91 on JeAUG while running at 148 frames per second. The supporting resources are a new benchmark, VAGU-T, with 7,567 human-validated videos and tool-calling reasoning traces, and a new joint metric, JeAUG, that couples semantic quality with temporal precision. A sympathetic reading is that the when–what dissociation is not an inherent trade-off but a trainable capability gap.

What carries the argument

The machinery is the chain-of-tool-thought itself, a single reasoning loop that replaces exhaustive processing: a think block records the model's temporal hypothesis, a crop-video tool returns densely resampled frames from that window, and the model decides whether to answer or re-enter the loop with a corrected window. That loop is trained end-to-end with a joint reward whose three terms supervise semantic accuracy, output format, and temporal precision, with the temporal term defined as IoU between the last tool-call window and ground truth—a choice that makes the only rewarded localization one the model actually inspected. GtS instantiates the same loop without training by using dual textual guidance and CLIP encoders to build a smoothed temporal anomaly curve, then feeding top segments to VQA and VTG models. The benchmark VAGU-T supplies the traces that make the loop learnable, and the JeAUG metric $\min(\gamma F(\mathrm{IoU}),1)\cdot \mathrm{Score}_{\mathrm{A.U.}}$ is the measurement surface on which all headline comparisons are read.

What would settle it

Take a random subset of VAGU-T, have independent raters mark anomaly intervals and rate anomaly descriptions, and compare their judgments to JeAUG's assumed human-consensus IoU of 0.7 and to its score progression; also recompute the head-to-head comparisons with standard metrics used separately (temporal mAP/IoU and ROUGE/BLEU). If JeAUG does not track the independent raters, or if the agentic model's advantage over frame/segment-wise baselines disappears on the standard metrics, the central claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that the apparent trade-off between temporal grounding and semantic understanding in video anomaly detection dissolves when the detector is built around a global-to-local inspection loop. In the training-free version, the loop is assembled from frozen components: a language model turns captions, category lists, and phrase banks into static and dynamic prompts, CLIP-style encoders score each frame or segment against those prompts to produce a smoothed anomaly curve, and high-scoring windows are handed to question-answering and temporal-grounding models for refinement. In the agentic version, the entire loop lives inside one multimodal large language model: the model looks at globally sampled frames, forms a temporal hypothesis, invokes a native video-cropping tool to inspect that window densely, and either commits to an answer or corrects the window and looks again, with the last tool window serving as the temporal localization. Training proceeds in two stages—cold-start supervised fine-tuning on reverse-engineered tool-calling traces, then reinforcement learning with reward $R = R_{\mathrm{acc}} + R_{\mathrm{format}} + R_{\mathrm{time}}$, where $R_{\mathrm{time}}$ is the IoU of the final inspected window against the human annotation. On the VAGU-T evaluation set the paper reports that the agentic model reaches the best accuracy among all compared methods (A.U. 7.35, JeAUG 5.91) at 148 FPS, and that the reinforcement-learning stage contributes the bulk of the joint-metric gain over supervised fine-tuning.

Load-bearing premise

The load-bearing premise is that the new JeAUG metric measures what practitioners care about: its grounding curve and video-length compensation were hand-fitted to match a small internal human study, and the paper provides no independent validation against established anomaly metrics or outside raters.

Editorial extensions

If this is right

  • If the reported numbers are reproducible, a single roughly 7-billion-parameter multimodal model can perform real-time video anomaly detection with both semantic explanations and temporal grounding, removing the need for exhaustive frame- and segment-wise captioning.
  • The reinforcement-learning gain over supervised fine-tuning (JeAUG 4.48 to 5.91) implies that outcome-based rewards elicit self-correction and targeted inspection that trace imitation alone does not teach.
  • The training-free GtS result (same-backbone JeAUG 2.28 to 4.04) implies that coarse textual hypotheses can direct scrutiny well enough to be immediately deployable wherever off-the-shelf VLMs are available.
  • Because VAGU-T's evaluation videos are held out from both SFT and RL, the claimed advantage of the agentic model is a test-set finding, not a training-set artifact, if the trace-construction pipeline is as described.
  • The JeAUG operating points (JeAUG at least 3, FPS at least 30) give the field a concrete target for what counts as an acceptable joint anomaly-detection system, allowing different methods to be compared on a single combined scale.

Reading between the lines

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

  • The same glance-scrutinize-think loop could transfer to other open-ended monitoring problems with no predefined query—wildlife surveillance, industrial fault detection, or security footage review—where the anomaly type and location must both be discovered autonomously.
  • A testable extension is to vary the maximum number of tool rounds and the density of resampled frames; the paper's per-category results suggest the agentic model's largest advantages would appear on short, subtle events such as shoplifting and traffic violations.
  • Because GtS is training-free and already improves grounding, it could be used as a noisy annotator or trace generator to bootstrap agentic training in new domains without fresh human annotation; the paper does not itself test this loop.
  • The grounding function $F(\mathrm{IoU})$ inside JeAUG could be reused as a dense reward during training rather than only an evaluation score, potentially aligning the optimized objective more directly with the reported metric.
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 / 5 minor

Summary. The paper addresses the joint problem of temporal grounding and semantic understanding in video anomaly detection (VAD), which it frames as a "when-what" dissociation in existing methods. It proposes two approaches: GtS, a training-free pipeline that uses static and dynamic textual guidance to perform coarse-to-fine anomaly localization and understanding with off-the-shelf VLMs, and a tool-augmented agentic VAD method in which a multimodal LLM learns, via cold-start supervised fine-tuning followed by GRPO-style reinforcement learning, to invoke a video-cropping tool and self-correct its temporal hypotheses. To support training and evaluation, the authors extend their prior VAGU benchmark into VAGU-T, adding chain-of-thought tool-calling traces, and introduce the JeAUG metric, which jointly combines an LLM-judged understanding score with a hand-fitted grounding function of temporal IoU. Experiments on VAGU-T report that GtS substantially improves over direct training-free baselines, and that the Agentic RL model achieves the best overall accuracy (A.U. 7.35, JeAUG 5.91) at 148 FPS.

Significance. If the reported results are taken at face value, the agentic method would be a meaningful advance: a single tool-augmented multimodal model that simultaneously delivers semantic anomaly understanding, temporal grounding, and real-time inference. The paper's strengths include a clearly specified two-stage training pipeline, ablations that support the incremental contribution of each reward term, a large human-validated benchmark with QA pairs and tool-calling traces, and a level of methodological detail that makes the approach reproducible in principle. The significance is, however, conditional on the validity of the author-defined JeAUG metric and on the actual raw temporal-grounding accuracy of the models, neither of which is demonstrated in the current manuscript.

major comments (4)
  1. [Sec. 6.2, Eq. (17), Eq. (18), Fig. 8, Table 5] The reported grounding factor of 0.804 for Agentic RL is, under the paper's own metric, consistent with very imprecise temporal localization, not precise grounding. For the stated average video length of 2,716 frames, Eq. (18) gives gamma ≈ 1.25, so the grounding factor 0.804 implies F(IoU) ≈ 0.64. In Eq. (17), this value corresponds to floor(10*IoU) = 1, i.e., a mean raw IoU between 0.1 and 0.2, far below the human-consensus anchor of 0.7 discussed in Sec. 6.2. Even if all evaluation videos were much shorter (gamma close to 1), the implied IoU would be only about 0.2. Moreover, because min(gamma*F(IoU),1) saturates at IoU ≈ 0.3 for long videos, JeAUG cannot reward any temporal-precision improvement above this level. The headline JeAUG 5.91 and the 'precise temporal grounding' claim are therefore not supported by the reported evidence. The authors should report raw IoU (and ideally mAP at IoU thresholds such as 0.3/0.5) separately from the A.U. score, and should re-examine the saturation behavior of JeAUG.
  2. [Sec. 6.1–6.3] The validity of JeAUG itself is not established. The grounding function F(IoU) and the length-compensation factor gamma are manually designed with constants (0.63, 0.5, 0.25, etc.) that are asserted to match human judgment, but the supporting human study is described only verbally: 10 participants on 50 videos, with the claim that even the lowest pairwise IoU was about 0.7. No agreement statistics, no confidence intervals, and no correlation between JeAUG and human ratings are reported, and there is no external validation against established VAD metrics or independent raters. Since all headline comparisons in Table 5 are reported in JeAUG or A.U., the central claims rest on a metric whose validity is asserted rather than demonstrated. The authors should provide quantitative human-alignment evidence and show that the main conclusions are robust to alternative metric choices (e.g., raw IoU, AUC, or thresholded variants).
  3. [Table 5, Sec. 7.2] All headline results come from a single evaluation pass with no error bars, no multiple seeds, and no significance tests. This is particularly important for the claimed superiority of Agentic RL over SFT (A.U. 7.35 vs. 6.62, JeAUG 5.91 vs. 4.48) and over the frame/segment-wise baselines, since RL training is stochastic and FPS measurements depend on implementation. The authors should report mean and standard deviation over at least three independent runs (or otherwise quantify variability), and perform a simple significance test (e.g., paired bootstrap) for the main comparisons in Table 5.
  4. [Sec. 5.4, Eq. (15), Sec. 6, Eq. (16)] The temporal grounding reward R_time in Eq. (15) is computed as the IoU between the final tool-calling window and the ground-truth interval, and the grounding component of JeAUG in Eq. (16) uses the same ground-truth interval through F(IoU). The RL objective therefore directly optimizes the grounding factor that is later reported in the evaluation metric. This alignment by construction does not by itself invalidate the results, but it means the reported JeAUG gains are exactly what the training procedure was designed to maximize. Independent validation of JeAUG and a separate reporting of raw IoU are necessary to determine whether the gain reflects genuine temporal-precision improvement or only metric-specific saturation.
minor comments (5)
  1. [Sec. 3.3, Eq. (1)] The length-adaptive probability P_multi uses an undefined clip notation and the thresholds L_min and L_max are never given numerical values; please specify them and clarify the formula.
  2. [Sec. 4.1, Eq. (4)] The Savitzky-Golay smoothing formula refers to coefficients q_p and a normalization Q that are not defined in the text; please make the notation explicit or cite the standard filter definition.
  3. [Table 4, Sec. 7.1] Inference FPS is reported without specifying the hardware and precision used for the trained models, whereas the training hardware is listed; this makes the FPS comparisons across methods hard to interpret.
  4. [Sec. 3.3] The description of the grounding annotation states that annotators reached consensus 'through IoU-based aggregation,' but no aggregation rule or agreement threshold is given; a brief specification would improve reproducibility.
  5. [Sec. 6.1, Table 2] The coefficient-of-variation comparison in Table 2 supports the claim that A.U. is more stable than ROUGE/BLEU/METEOR on this data, but the table does not report correlation with human judgments, which is the more relevant validity check for a new metric.

Circularity Check

1 steps flagged · score 6.0 of 10

The reported grounding factor 0.804 is generated by the paper's own hand-fitted F(IoU) and gamma, and by the paper's equations it corresponds to raw IoU of only roughly 0.1-0.2, so the claimed temporal precision is an artifact of the metric rather than independent evidence of precise localization.

  1. fitted input called prediction [Sec. 6.2 (Eqs. 16-18), Sec. 6.3, Sec. 7.4 (Fig. 8)]
    "for grounding we adopt the grounding factor γ·F(IoU) from JeAUG as the evaluation criterion... The Agentic RL model improves both dimensions further (7.35 and 0.804), with the grounding gain directly attributable to the temporal reward R_time that supervises where the model chooses to look."

    The grounding factor is not raw IoU but the hand-fitted F of Eq. 17 multiplied by gamma of Eq. 18. For the stated average video length of 2,716 frames, Eq. 18 gives gamma≈1.25, so the reported 0.804 implies F(IoU)≈0.64. Eq. 17's floor term then gives floor(10·IoU)=1, i.e., raw IoU in [0.1,0.2), far below the paper's own human-consensus IoU of 0.7. The metric's floor of 0.5 for IoU<0.1 and the length-compensation factor 1.25 manufacture the apparent grounding score by construction. The claim that Agentic RL attains precise temporal grounding is therefore an artifact of the metric's fitted constants, not an independent measurement of localization accuracy.

full rationale

The paper's main quantitative claims are evaluated only on the authors' VAGU-T benchmark with the authors' JeAUG metric. That alone is not circular: the benchmark annotations are human-validated and the evaluation set is held out from training. The central circularity is narrower: the grounding half of JeAUG is a hand-fitted function F(IoU) with a floor at 0.5 for IoU<0.1 and a saturation at IoU=0.7, multiplied by a length-compensation factor gamma that reaches 1.25 at the benchmark's mean video length. Using gamma·F(IoU) as evidence of temporal grounding makes the reported 0.804 for Agentic RL consistent with raw IoU of only 0.1-0.2, which contradicts the paper's own human-consensus standard of 0.7. The headline JeAUG 5.91 follows directly from A.U. 7.35 times this inflated grounding factor. No uniqueness theorem is imported, and the RL training itself uses raw IoU (Eq. 15), so the trained model may genuinely improve localization; but the paper's evidence for the 'when' side of its central claim is defined by the metric it designed, making the precision claim partially circular. The A.U. side and the FPS measurements are independent and not circular.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central claims rest on a set of hand-tuned hyperparameters in both the GtS pipeline and the JeAUG metric, plus several domain assumptions about LLM judges, CLIP similarity, and transferability of tool-augmented RL. No genuinely new physical or conceptual entities are introduced.

free parameters (8)
  • alpha (GtS) = 0.4
    Weight between static and dynamic similarity curves in Eq. (4), tuned on 100 held-out videos as described in Sec. 7.1.
  • tau (GtS) = mean of all peak values of the similarity curve
    Peak selection threshold in Eq. (6), set from held-out videos and used to filter candidate anomaly peaks.
  • theta (GtS) = (total frame number)/12
    Minimum inter-peak distance in Eq. (6), derived from the 100-video tuning set.
  • eta (GtS) = 1/20
    Window size proportion relative to video length T in Eq. (7), tuned on held-out videos.
  • F(IoU) constants (JeAUG) = 0.63, 0.7, 0.5, 10
    Manually set to make the grounding score saturate at the human-consensus IoU of 0.7, with a 0.5 floor for failed grounding but successful understanding (Eq. 17).
  • gamma constants (JeAUG) = 0.25, 100, 0.03, 100
    Video-length compensation curve shape in Eq. (18), hand-designed to grow from 1 and saturate at 1.25 for long videos.
  • trace acceptance IoU threshold = 0.3
    Minimum IoU for retaining a generated tool-calling trace in Sec. 3.3; filters out traces whose final tool window does not overlap the ground-truth interval.
  • RL length threshold for R_acc=0
    Answer length threshold mentioned in Sec. 5.4 for flagging length-hacking answers, but not numerically specified.
assumptions (4)
  • domain assumption LLM-as-a-Judge (GPT-4o) produces valid anomaly-description quality scores.
    The A.U. score and R_acc reward rely on an external LLM judge; no validation against human ratings is reported for the judge itself.
  • domain assumption Human IoU consensus of approximately 0.7 is an appropriate saturation point for grounding quality.
    Based on a small internal study (10-15 participants) reported in Sec. 6, with no public data or statistical details.
  • domain assumption CLIP and Video-CLIP cross-modal similarities can localize anomaly-relevant segments via text prompts.
    Standard assumption in zero-shot VAD, adopted without task-specific validation beyond the reported experiments.
  • domain assumption Tool-augmented GRPO training transfers from general video QA to query-free VAD.
    The method builds on VITAL/LongVT results for QA; the paper adapts it to VAD but does not compare against alternative training regimes on an external benchmark.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Glance, Scrutinize, and Think: Advancing Video Anomaly Detection from Training-Free to Agentic Reasoning." pith.science (2026). https://pith.science/paper/WOIWBKY7

@misc{pith2026260811260,
  author       = {Pith},
  title        = {Pith review of: Glance, Scrutinize, and Think: Advancing Video Anomaly Detection from Training-Free to Agentic Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WOIWBKY7}},
  note         = {Machine review of arXiv:2608.11260}
}
read the original abstract

Video Anomaly Detection (VAD) aims to identify anomalous events and localize their temporal intervals. Existing approaches exhibit a "when-what" dissociation: traditional DNN-based methods localize when anomalies occur but lack semantic understanding, whereas LLM-based methods explain what happens but neglect precise temporal grounding. We attribute this to the absence of a unified reasoning paradigm. Inspired by how humans inspect surveillance videos - glancing globally to form temporal hypotheses, scrutinizing suspicious segments, and thinking iteratively to correct errors - we study this global-to-local paradigm from two perspectives. We first propose Glance then Scrutinize (GtS), a training-free framework using static and dynamic textual guidance for coarse-to-fine anomaly grounding and understanding, balancing accuracy and speed. To break the ceiling imposed by frozen external modules, we further propose a tool-augmented agentic VAD method, where a multimodal large language model learns to invoke a video cropping tool, inspect densely resampled frames, and self-correct mislocalized hypotheses, via cold-start supervised fine-tuning followed by reinforcement learning with a joint answer-grounding reward. For training and evaluation, we extend our prior VAGU benchmark into VAGU-T (Video Anomaly Grounding, Understanding, and Thinking), comprising 7,567 real-world videos over 21 anomaly categories with human-validated grounding, explanations, QA pairs, and chain-of-thought tool-calling traces. We further introduce JeAUG, a metric jointly evaluating semantic interpretability and temporal precision. Experiments show that GtS substantially surpasses training-free baselines, while the agentic model delivers both higher accuracy and faster inference.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 38 canonical work pages

  1. [1]

    In: Proceedings of the International Conference on Image and Graphics (ICIG), pp

    Gao, S., Gong, J., Yang, P., Liang, C., Huang, L.: A stable long- term tracking method for group-housed pigs. In: Proceedings of the International Conference on Image and Graphics (ICIG), pp. 238–249 (2023)

    work page 2023

  2. [2]

    arXiv preprint arXiv:2507.21649 (2025)

    Gao, S., Yang, P., Guo, H., Liu, Y., Chen, Y., Li, S., Zhu, H., Xu, J., Zhang, X.-Y., Huang, L.: The evolution of video anomaly detection: A unified framework from dnn to mllm. arXiv preprint arXiv:2507.21649 (2025)

    arXiv 2025

  3. [3]

    In: European Conference on Computer Vision, pp

    Lu, Y., Yu, F., Reddy, M.K.K., Wang, Y.: Few-shot scene-adaptive anomaly detection. In: European Conference on Computer Vision, pp. 125–141 (2020)

    work page 2020

  4. [4]

    In: European Conference on Computer Vision, pp

    Wang, G., Wang, Y., Qin, J., Zhang, D., Bao, X., Huang, D.: Video anomaly detection by solving decoupled spatio-temporal jigsaw puzzles. In: European Conference on Computer Vision, pp. 494–511 (2022)

    work page 2022

  5. [5]

    arXiv preprint arXiv:2302.05160 (2023)

    Zhou, H., Yu, J., Yang, W.: Dual memory units with uncertainty reg- ulation for weakly supervised video anomaly detection. arXiv preprint arXiv:2302.05160 (2023)

    arXiv 2023

  6. [6]

    arXiv preprint arXiv:2308.11681 (2023)

    Wu, P., Zhou, X., Pang, G., Zhou, L., Yan, Q., Wang, P., Zhang, Y.: Vadclip: Adapting vision-language models for weakly supervised video anomaly detection. arXiv preprint arXiv:2308.11681 (2023)

    arXiv 2023

  7. [7]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Acsintoae, A., Florescu, A., Georgescu, M.-I., Mare, T., Sumedrea, P., Ionescu, R.T., Khan, F.S., Shah, M.: Ubnormal: New benchmark for supervised open-set video anomaly detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 20143–20153 (2022) Springer Nature 2021 LATEX template 30Glance, Scrutinize, and Think

    work page 2022

  8. [8]

    In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Du, H., Zhang, S., Xie, B., Nan, G., Zhang, J., Xu, J., Liu, H., Leng, S., Liu, J., Fan, H.,et al.: Uncovering what why and how: A comprehensive benchmark for causation understanding of video anomaly. In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 18793–18803 (2024)

    work page 2024

Show all 67 references
  1. [9]

    Advances in Neural Information Processing Systems37, 139751–139785 (2024)

    Tang, J., Lu, H., Wu, R., Xu, X., Ma, K., Fang, C., Guo, B., Lu, J., Chen, Q., Chen, Y.: Hawk: Learning to understand open-world video anomalies. Advances in Neural Information Processing Systems37, 139751–139785 (2024)

    2024

  2. [10]

    arXiv preprint arXiv:2412.01095 (2024)

    Ye, M., Liu, W., He, P.: Vera: Explainable video anomaly detec- tion via verbalized learning of vision-language models. arXiv preprint arXiv:2412.01095 (2024)

    2024 arXiv

  3. [11]

    arXiv preprint arXiv:2406.12235 (2024)

    Zhang, H., Xu, X., Wang, X., Zuo, J., Han, C., Huang, X., Gao, C., Wang, Y., Sang, N.: Holmes-vad: Towards unbiased and explainable video anomaly detection via multi-modal llm. arXiv preprint arXiv:2406.12235 (2024)

    2024 arXiv

  4. [12]

    arXiv preprint arXiv:2401.05702 (2024)

    Lv, H., Sun, Q.: Video anomaly detection and explanation via large language models. arXiv preprint arXiv:2401.05702 (2024)

    2024 arXiv

  5. [13]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Zanella, L., Menapace, W., Mancini, M., Wang, Y., Ricci, E.: Harnessing large language models for training-free video anomaly detection. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 18527–18536 (2024)

    2024

  6. [14]

    In: ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp

    Gao, S., Yang, P., Huang, L.: Suvad: Semantic understanding based video anomaly detection using mllm. In: ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 1–5 (2025). IEEE

    2025

  7. [15]

    In: International Conference on Machine Learning, pp

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J.,et al.: Learning transfer- able visual models from natural language supervision. In: International Conference on Machine Learning, pp. 8748–8763 (2021)

    2021

  8. [16]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Zhang, H., Gu, X., Li, J., Ma, C., Bai, S., Zhang, C., Zhang, B., Zhou, Z., He, D., Tang, Y.: Thinking with videos: Multimodal tool-augmented reinforcement learning for long video reasoning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,...

    2026

  9. [17]

    In: Proceedings of the IEEE/CVF Conference on Springer Nature 2021 LATEX template Glance, Scrutinize, and Think31 Computer Vision and Pattern Recognition, pp

    Yang, Z., Wang, S., Zhang, K., Wu, K., Leng, S., Zhang, Y., Li, B., Qin, C., Lu, S., Li, X.,et al.: Longvt: Incentivizing” thinking with long videos” via native tool calling. In: Proceedings of the IEEE/CVF Conference on Springer Nature 2021 LATEX template Glance, Scrutinize, ...

    2026

  10. [18]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Gao, S., Yang, P., Liu, Y., Chen, Y., Zhu, H., Zhang, X.-Y., Huang, L.: Vagu & gts: Llm-based benchmark and framework for joint video anomaly grounding and understanding. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 40, pp. 4167–4175 (2026). https://...

    2026 doi

  11. [19]

    In: Proceedings of the Workshop on Text Summarization Branches Out, 2004 (2004)

    Chin-Yew, L.: Rouge: A package for automatic evaluation of summaries. In: Proceedings of the Workshop on Text Summarization Branches Out, 2004 (2004)

    2004

  12. [20]

    In: Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pp

    Papineni, K., Roukos, S., Ward, T., Zhu, W.-J.: Bleu: a method for automatic evaluation of machine translation. In: Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pp. 311–318 (2002)

    2002

  13. [21]

    In: Proceedings of the Acl Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation And/or Summarization, pp

    Banerjee, S., Lavie, A.: Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. In: Proceedings of the Acl Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation And/or Summarization, pp. 65–72 (2005)

    2005

  14. [22]

    IEEE Access8, 88170–88176 (2020)

    Dong, F., Zhang, Y., Nie, X.: Dual discriminator generative adversarial network for video anomaly detection. IEEE Access8, 88170–88176 (2020)

    2020

  15. [23]

    In: Proceedings of the 25th ACM International Conference on Multimedia, pp

    Zhao, Y., Deng, B., Shen, C., Liu, Y., Lu, H., Hua, X.-S.: Spatio-temporal autoencoder for video anomaly detection. In: Proceedings of the 25th ACM International Conference on Multimedia, pp. 1933–1941 (2017)

    2017

  16. [24]

    In: Proceedings of the IEEE Inter- national Conference on Computer Vision, pp

    Astrid, M., Zaheer, M.Z., Lee, S.-I.: Synthetic temporal anomaly guided end-to-end video anomaly detection. In: Proceedings of the IEEE Inter- national Conference on Computer Vision, pp. 207–214 (2021)

    2021

  17. [25]

    In: IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp

    Wu, S., Moore, B.E., Shah, M.: Chaotic invariants of lagrangian particle trajectories for anomaly detection in crowded scenes. In: IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 2054–2060 (2010). IEEE

    2010

  18. [26]

    In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, pp

    Wang, J., Cherian, A.: Gods: Generalized one-class discriminative sub- spaces for anomaly detection. In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, pp. 8201–8211 (2019)

    2019

  19. [27]

    IEEE transactions on neural networks and learning systems31(7), 2609–2622 (2019)

    Wu, P., Liu, J., Shen, F.: A deep one-class neural network for anomalous event detection in complex scenes. IEEE transactions on neural networks and learning systems31(7), 2609–2622 (2019)

    2019

  20. [28]

    In: ICASSP 2025 – 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (2025)

    Gao, S., Yang, P., Huang, L.: Semi-supervised video anomaly detection Springer Nature 2021 LATEX template 32Glance, Scrutinize, and Think with compact deformable 3d convolution. In: ICASSP 2025 – 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (IC...

    2025

  21. [29]

    In: Proceedings of the Asian Conference on Computer Vision, pp

    Gao, S., Yang, P., Huang, L.: Scene-adaptive svad based on multi-modal action-based feature extraction. In: Proceedings of the Asian Conference on Computer Vision, pp. 2471–2488 (2024)

    2024

  22. [30]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Zaheer, M.Z., Mahmood, A., Khan, M.H., Segu, M., Yu, F., Lee, S.-I.: Generative cooperative learning for unsupervised video anomaly detec- tion. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 14744–14754 (2022)

    2022

  23. [31]

    In: 2023 IEEE International Conference on Image Processing (ICIP), pp

    Joo, H.K., Vo, K., Yamazaki, K., Le, N.: Clip-tsa: Clip-assisted tem- poral self-attention for weakly-supervised video anomaly detection. In: 2023 IEEE International Conference on Image Processing (ICIP), pp. 3230–3234 (2023)

    2023

  24. [32]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Ding, C., Pang, G., Shen, C.: Catching both gray and black swans: Open- set supervised anomaly detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 7388–7398 (2022)

    2022

  25. [33]

    In: European Conference on Computer Vision, pp

    Zhu, Y., Bao, W., Yu, Q.: Towards open set video anomaly detection. In: European Conference on Computer Vision, pp. 395–412 (2022)

    2022

  26. [34]

    arXiv preprint arXiv:2310.12790 (2023)

    Zhu, J., Ding, C., Tian, Y., Pang, G.: Anomaly heterogeneity learning for open-set supervised anomaly detection. arXiv preprint arXiv:2310.12790 (2023)

    2023 arXiv

  27. [35]

    arXiv preprint arXiv:2406.10326 (2024)

    Bharadwaj, R., Gani, H., Naseer, M., Khan, F.S., Khan, S.: Vane-bench: Video anomaly evaluation benchmark for conversational lmms. arXiv preprint arXiv:2406.10326 (2024)

    2024 arXiv

  28. [36]

    arXiv preprint arXiv:2503.04504 (2025)

    Ahn, S., Jo, Y., Lee, K., Kwon, S., Hong, I., Park, S.: Anyanomaly: Zero-shot customizable video anomaly detection with lvlm. arXiv preprint arXiv:2503.04504 (2025)

    2025

  29. [37]

    In: European Conference on Computer Vision, pp

    Yang, Y., Lee, K., Dariush, B., Cao, Y., Lo, S.-Y.: Follow the rules: reason- ing for video anomaly detection with large language models. In: European Conference on Computer Vision, pp. 304–322 (2024). Springer

    2024

  30. [38]

    In: Proceedings of the 64th Annual Meeting of the Association for Computa- tional Linguistics (Volume 1: Long Papers), pp

    Song, S., Wang, C., Lang, Y., Yang, Y., Zhang, X.-Y.: Lbllm: Lightweight binarization of large language models via three-stage distillation. In: Proceedings of the 64th Annual Meeting of the Association for Computa- tional Linguistics (Volume 1: Long Papers), pp. 35470–35484. ...

    2026 doi

  31. [39]

    In: Findings of the Association for Computational Linguistics: ACL 2025, pp

    Song, S., Wang, C., Wang, R.-Q., Yang, Y., Zhang, X.-Y.: Achieving binary weight and activation for llms using post-training quantization. In: Findings of the Association for Computational Linguistics: ACL 2025, pp. 8782–8795 (2025)

    2025

  32. [40]

    arXiv preprint arXiv:2505.15966 (2025)

    Wang, H., Su, A., Ren, W., Lin, F., Chen, W.: Pixel reasoner: Incentiviz- ing pixel-space reasoning with curiosity-driven reinforcement learning. arXiv preprint arXiv:2505.15966 (2025)

    2025 arXiv

  33. [41]

    arXiv preprint arXiv:2505.14362 (2025)

    Zheng, Z., Yang, M., Hong, J., Zhao, C., Xu, G., Yang, L., Shen, C., Yu, X.: Deepeyes: Incentivizing” thinking with images” via reinforcement learning. arXiv preprint arXiv:2505.14362 (2025)

    2025 arXiv

  34. [42]

    In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pp

    Yang, Z., Song, J., Song, S., Pang, W., Yuan, Y.: Mermaid: Multi- perspective self-reflective agents with generative augmentation for emo- tion recognition. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pp. 24650–24666 (2025)

    2025

  35. [43]

    arXiv preprint arXiv:2510.23473 (2025)

    Wang, S., Jin, J., Wang, X., Song, L., Fu, R., Wang, H., Ge, Z., Lu, Y., Cheng, X.: Video-thinker: Sparking” thinking with videos” via reinforcement learning. arXiv preprint arXiv:2510.23473 (2025)

    2025

  36. [44]

    arXiv preprint arXiv:2506.13654 (2025)

    Tian, S., Wang, R., Guo, H., Wu, P., Dong, Y., Wang, X., Yang, J., Zhang, H., Zhu, H., Liu, Z.: Ego-r1: Chain-of-tool-thought for ultra-long egocentric video reasoning. arXiv preprint arXiv:2506.13654 (2025)

    2025 arXiv

  37. [45]

    Advances in Neural Information Processing Systems38, 129243–129273 (2026)

    Fan, S., Cui, J., Guo, M.-H., Yang, S.: Tool-augmented spatiotemporal reasoning for streamlining video question answering task. Advances in Neural Information Processing Systems38, 129243–129273 (2026)

    2026

  38. [46]

    Nature645(8081), 633–638 (2025)

    Guo, D., Yang, D., Zhang, H., Song, J., Wang, P., Zhu, Q., Xu, R., Zhang, R., Ma, S., Bi, X.,et al.: Deepseek-r1 incentivizes reasoning in llms through reinforcement learning. Nature645(8081), 633–638 (2025)

    2025

  39. [47]

    arXiv preprint arXiv:2402.03300 (2024)

    Shao, Z., Wang, P., Zhu, Q., Xu, R., Song, J., Bi, X., Zhang, H., Zhang, M., Li, Y., Wu, Y., et al.: Deepseekmath: Pushing the limits of mathemat- ical reasoning in open language models. arXiv preprint arXiv:2402.03300 (2024)

    2024 arXiv

  40. [48]

    arXiv preprint arXiv:2503.06749 (2025)

    Huang, W., Jia, B., Zhai, Z., Cao, S., Ye, Z., Zhao, F., Xu, Z., Tang, X., Hu, Y., Lin, S.: Vision-r1: Incentivizing reasoning capability in multimodal large language models. arXiv preprint arXiv:2503.06749 (2025)

    2025 arXiv

  41. [49]

    arXiv preprint arXiv:2503.07365 (2025)

    Meng, F., Du, L., Liu, Z., Zhou, Z., Lu, Q., Fu, D., Han, T., Shi, B., Springer Nature 2021 LATEX template 34Glance, Scrutinize, and Think Wang, W., He, J., et al.: Mm-eureka: Exploring the frontiers of multi- modal reasoning with rule-based reinforcement learning. arXiv prepr...

    2025 arXiv

  42. [50]

    arXiv preprint arXiv:2504.07615 (2025)

    Shen, H., Liu, P., Li, J., Fang, C., Ma, Y., Liao, J., Shen, Q., Zhang, Z., Zhao, K., Zhang, Q., et al.: Vlm-r1: A stable and generalizable r1-style large vision-language model. arXiv preprint arXiv:2504.07615 (2025)

    2025 arXiv

  43. [51]

    arXiv preprint arXiv:2512.00743 (2025)

    Lyu, Q., Chen, Z., Wang, C., Shi, H., Gao, S., Piao, R., Zeng, Y., Si, J., Ding, F., Li, J., Lau, C.P., Wang, W.: Multi-grpo: Multi-group advantage estimation for text-to-image generation with tree-based trajectories and multiple rewards. arXiv preprint arXiv:2512.00743 (2025)

    2025

  44. [52]

    arXiv preprint arXiv:2504.21277 (2025)

    Zhou, G., Qiu, P., Chen, C., Wang, J., Yang, Z., Xu, J., Qiu, M.: Rein- forced mllm: A survey on rl-based reasoning in multimodal large language models. arXiv preprint arXiv:2504.21277 (2025)

    2025 arXiv

  45. [53]

    Advances in Neural Information Processing Systems38, 99114–99137 (2026)

    Feng, K., Gong, K., Li, B., Guo, Z., Wang, Y., Peng, T., Wu, J., Zhang, X., Wang, B., Yue, X.: Video-r1: Reinforcing video reasoning in mllms. Advances in Neural Information Processing Systems38, 99114–99137 (2026)

    2026

  46. [54]

    Advances in neural information processing systems38, 4350–4376 (2026)

    Wang, Q., Yu, Y., Yuan, Y., Mao, R., Zhou, T.: Videorft: Incentivizing video reasoning capability in mllms via reinforced fine-tuning. Advances in neural information processing systems38, 4350–4376 (2026)

    2026

  47. [55]

    Advances in Neural Information Processing Systems38, 83330–83364 (2026)

    Wang, Y., Wang, Z., Xu, B., Du, Y., Lin, K., Xiao, Z., Yue, Z., Ju, J., Zhang, L., Yang, D.,et al.: Time-r1: Post-training large vision language model for temporal video grounding. Advances in Neural Information Processing Systems38, 83330–83364 (2026)

    2026

  48. [56]

    arXiv preprint arXiv:2504.06958 (2025)

    Li, X., Yan, Z., Meng, D., Dong, L., Zeng, X., He, Y., Wang, Y., Qiao, Y., Wang, Y., Wang, L.: Videochat-r1: Enhancing spatio-temporal perception via reinforcement fine-tuning. arXiv preprint arXiv:2504.06958 (2025)

    2025 arXiv

  49. [57]

    Advances in Neural Information Processing Systems38, 172842–172870 (2026)

    Chen, Y., Huang, W., Shi, B., Hu, Q., Ye, H., Zhu, L., Liu, Z., Molchanov, P., Kautz, J., Qi, X.,et al.: Scaling rl to long videos. Advances in Neural Information Processing Systems38, 172842–172870 (2026)

    2026

  50. [58]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Sultani, W., Chen, C., Shah, M.: Real-world anomaly detection in surveil- lance videos. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 6479–6488 (2018)

    2018

  51. [59]

    In: Computer Vision–ECCV 2020: 16th European Confer- ence, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXX 16, pp

    Wu, P., Liu, J., Shi, Y., Sun, Y., Shao, F., Wu, Z., Yang, Z.: Not only look, but also listen: Learning multimodal violence detection under weak supervision. In: Computer Vision–ECCV 2020: 16th European Confer- ence, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXX 16, p...

    2020

  52. [60]

    arXiv preprint arXiv:2507.06261 (2025)

    Comanici, G., Bieber, E., Schaekermann, M., Pasupat, I., Sachdeva, N., Dhillon, I., Blistein, M., Ram, O., Zhang, D., Rosen, E., et al.: Gem- ini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arXiv pre...

    2025 arXiv

  53. [61]

    arXiv preprint arXiv:2410.21276 (2024)

    Hurst, A., Lerer, A., Goucher, A.P., Perelman, A., Ramesh, A., Clark, A., Ostrow, A., Welihinda, A., Hayes, A., Radford, A., et al.: Gpt-4o system card. arXiv preprint arXiv:2410.21276 (2024)

    2024 arXiv

  54. [62]

    arXiv preprint arXiv:2503.20783 (2025)

    Liu, Z., Chen, C., Li, W., Qi, P., Pang, T., Du, C., Lee, W.S., Lin, M.: Understanding r1-zero-like training: A critical perspective. arXiv preprint arXiv:2503.20783 (2025)

    2025 arXiv

  55. [63]

    arXiv preprint arXiv:2407.21783 (2024)

    Grattafiori, A., Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Letman, A., Mathur, A., Schelten, A., Vaughan, A., et al.: The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

    2024 arXiv

  56. [64]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Ren, S., Yao, L., Li, S., Sun, X., Hou, L.: Timechat: A time-sensitive multimodal large language model for long video understanding. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 14313–14323 (2024)

    2024

  57. [65]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Huang, B., Wang, X., Chen, H., Song, Z., Zhu, W.: Vtimellm: Empower llm to grasp video moments. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 14271–14280 (2024)

    2024

  58. [66]

    arXiv preprint arXiv:2306.05424 (2023)

    Maaz, M., Rasheed, H., Khan, S., Khan, F.S.: Video-chatgpt: Towards detailed video understanding via large vision and language models. arXiv preprint arXiv:2306.05424 (2023)

    2023 arXiv

  59. [67]

    arXiv preprint arXiv:2409.14485 (2024)

    Shu, Y., Liu, Z., Zhang, P., Qin, M., Zhou, J., Liang, Z., Huang, T., Zhao, B.: Video-xl: Extra-long vision language model for hour-scale video understanding. arXiv preprint arXiv:2409.14485 (2024)

    2024 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.