Pith. sign in

REVIEW 4 major objections 5 minor 16 references

ADOPD: Reference-Privileged On-Policy Distillation for MLLM-Based Industrial Anomaly Detection

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read ADOPD internalizes reference comparison into a query-only model, gaining 6.14 points on MMAD.

desk verdict A clean training-recipe paper with a real mechanism and an unvalidated proxy; worth refereeing, not desk-rejecting. read the letter →

arxiv 2608.09789 v1 pith:YBB5G4O5 submitted 2026-08-10 cs.CV

Jingtai He , Shiyuan Meng , Wenchao Meng , Qinmin Yang This is my paper · ORCID
classification cs.CV
keywords industrialanomalydetectionmultimodallargelanguagemodelon-policydistillationknowledgereferencecomparisonprivilegedinformationzero-shotMMADbenchmark
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

ADOPD asks whether an industrial anomaly detector based on a multimodal language model can absorb the benefit of comparing a query image against a normal reference during training, so that no reference is needed at deployment. The paper's answer is yes: a teacher that sees both a matched and a deliberately mismatched reference supervises a student that sees only the query image, and the student learns to inspect fine-grained defects on its own. On the MMAD benchmark, the resulting query-only student reaches 77.31% average zero-shot accuracy, improving its Qwen3-VL-4B backbone by 6.14 points and beating the one-shot reference-conditioned version by 2.64 points. The gain concentrates in comparison-sensitive subtasks such as anomaly discrimination and defect localization, which is the signature of learned reference comparison rather than generic supervision.

What carries the argument

The load-bearing component is the contrastive reference support score $s_g = \frac{1}{T_g}\sum_{t=1}^{T_g}[\log p_T^\mathrm{mat}(y_t) - \log p_T^\mathrm{mis}(y_t)]$, the average teacher log-likelihood gap between the matched and mismatched reference views on a student-generated rollout. This score is standardized within the group of rollouts for the same prompt to produce the sequence-level weight $w_g = s_g[1 + \mathrm{clip}((s_g - \mu)/\sigma, -\epsilon_w, \epsilon_w)]$, which scales the matched-reference token advantage $A_g^t = \mathrm{sg}[\log p_T^\mathrm{mat}(y_t) - \log p_S(y_t)]$ before a policy-gradient update. The matched view sets the learning direction; the mismatched view calibrates how strongly to trust that direction without ever reversing it, because the weight is detached and non-negative.

What would settle it

Train ADOPD with the mismatched reference replaced by a matched reference from a different instance of the same product class, keeping the prompt and crop procedure identical; if accuracy stays at the matched/mismatched level, the support score is not measuring reference-validity discrimination.

Watch

Extended reading notes

Core claim

The paper establishes that the benefits of inference-time reference comparison can be internalized into model parameters through a distillation scheme that separates what to learn from how strongly to learn it. A frozen reference-aware teacher evaluates student-generated rollouts under two asymmetric views: a matched regional reference that preserves the valid query–reference relation, and a mismatched reference that deliberately breaks it. The matched-view teacher–student log-ratio gives the token-level learning direction, while the matched–mismatched likelihood gap provides a sequence-level weight that calibrates how much each rollout should contribute. The student, which sees only the query image and question, thereby learns an anomaly inspection strategy that at zero-shot inference outperforms the same backbone with explicit one-shot reference access on the MMAD benchmark.

Load-bearing premise

The contrastive support score $s_g$ truly measures whether the teacher's preference is grounded in the query–reference comparison; if it mainly tracks crop artifacts, prompt wording, or language priors, the weighting rescales rollouts for the wrong reason.

Editorial extensions

If this is right

  • Zero-shot MLLM anomaly detection can match or exceed explicit one-shot reference use on average accuracy, while removing retrieval and reference processing at inference time.
  • The training signal is not generic visual supervision: gains cluster in anomaly discrimination, defect localization, and defect analysis, and regional reference crops outperform full-image references.
  • On-policy distillation with a reference-privileged teacher beats off-policy SFT on the same 6K-record dataset, and the contrastive reference weight adds a further gain over uncalibrated OPD.
  • The method transfers to a 2B student and to an EMA self-teacher, though the gains are larger with a 4B student and an external 32B teacher.

Reading between the lines

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

  • Because ADOPD improves recall sharply (+23.95 points over one-shot) at some precision cost, the learned strategy may be a more aggressive detector; future work could tune the reference weight to trade false alarms against missed defects.
  • The matched/mismatched contrast suggests a general recipe for privileged-information distillation: any teacher-side condition whose removal should change the answer can calibrate update strength, and text-only or object-detection privileged signals could be tried outside industrial anomaly detection.
  • The support score could be evaluated explicitly against human-labeled reference usefulness, or by ablations that corrupt only the reference region while keeping the prompt identical, to test whether the weight tracks genuine reference validity rather than low-level artifacts.
  • Since ADOPD still depends on paired images and defect annotations, automatically constructing weak references, such as anomaly-free patches, is a direct next step the paper itself flags.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 proposes ADOPD, a reference-privileged on-policy distillation method for MLLM-based industrial anomaly detection. A query-only student samples rollouts, and a frozen reference-conditioned teacher scores the same tokens under a matched reference, producing a token-level advantage, while a mismatched reference provides a sequence-level contrastive weight. The weighted policy-gradient objective is trained on 6,000 paired Real-IAD records. On MMAD, ADOPD reports 77.31% zero-shot average accuracy, improving the Qwen3-VL-4B backbone by 6.14 points and exceeding the backbone's one-shot setting by 2.64 points. Ablations compare SFT, uncalibrated OPD, reference granularity, and teacher construction.

Significance. If the central claim is fully supported, ADOPD would be a notable demonstration that the benefit of inference-time reference comparison can be amortized into model parameters, with a zero-shot student surpassing a one-shot teacher-conditioned baseline. The OPD derivation in Section 2 and Appendix A is internally consistent, and the matched/mismatched contrastive design is well motivated. However, the paper's mechanistic conclusion rests on an unvalidated proxy, the sequence-level support score s_g, and the experimental evidence is reported without variance estimates. These issues make the current version prematurely conclusive.

major comments (4)
  1. [§3.2, Eqs. (9)–(10), Table 2] The central mechanistic claim that ADOPD's 2.36-point gain over uncalibrated OPD comes from reference-specific weighting is not yet supported because s_g is never validated as a measure of genuine query–reference grounding. s_g is a teacher log-likelihood gap on student-sampled tokens; it can be inflated by crop artifacts, red-bounding-box cues, or language priors that are unrelated to the validity of the reference relation. The within-prompt standardization in Eq. (10) removes constant offsets but not token-dependent confounds. A minimal check would be a within-prompt permutation test: randomly permute the weights w_g across the G rollouts of the same prompt and show that the calibration gain in Table 2 degrades or disappears. Additionally, s_g could be correlated with rollout quality, for example by comparing each rollout against the reference-conditioned teacher's answer or against the audit metadata answer. Without such a test, the observed gain could come from an unrelated mechanism, and the paper's central interpretation remains an assertion.
  2. [§3.2, Eq. (10), sentence 'the weight w_g is detached and non-negative'] The claim that w_g is non-negative and therefore never reverses the matched-reference OPD direction is not implied by Eq. (10). Since w_g = s_g [1 + clip((s_g − μ)/σ, −ε_w, ε_w)] and the bracketed factor is non-negative for ε_w ≤ 1, w_g becomes negative whenever s_g is negative. The paper does not report whether s_g is always positive in the 60 training steps. If negative s_g occurs, the sequence weight reverses the token-level advantage, contradicting the stated design. The authors should either use a strictly positive transformation of s_g, such as |s_g| or exp(s_g), or provide empirical evidence that s_g is always non-negative in their runs.
  3. [§4.3, Table 4, '4B EMA' row] The EMA-teacher ablation does not isolate teacher construction. The student policy is query-only, so an EMA copy of the student cannot provide the matched/mismatched reference-conditioned distributions used in ADOPD unless the same reference conditioning is also fed to the EMA model. If the EMA teacher is evaluated on student rollouts without any reference input, then this row changes both the teacher construction and the availability of privileged references, and the comparison to the 32B reference-aware teacher is confounded. The text should state explicitly what inputs the EMA teacher receives; otherwise the 74.96 vs. 77.31 comparison cannot be attributed to teacher construction.
  4. [§4.1–§4.3, Tables 1–4] All quantitative claims are based on single runs with no variance estimates. Several headline differences are small relative to typical MLLM training noise: ADOPD exceeds Reason-IAD by 0.60 points in Table 1, SFT exceeds uncalibrated OPD by 0.75 points in Table 2, and the random-crop vs. no-image contrast in Table 3 is 0.71 points. With a single run, these differences may not be statistically reliable. The authors should report mean and standard deviation over at least three seeds, or provide a significance test, for the main comparison and the ablations that support the core claims.
minor comments (5)
  1. [§3.1 and Appendix B, Table 5] The teacher input protocol is inconsistent: Section 3.1 states that the teacher receives either a matched or a mismatched reference, while Appendix B and Table 5 show a teacher raw prompt containing both <teacher image> and <teacherneg image>. Please clarify whether the two reference views are implemented by replacing the image in a single prompt or by masking one of the two images, and whether the teacher ever sees both references simultaneously.
  2. [Abstract and §4.2] The phrase 'outperforming its one-shot setting by 2.64 points' is ambiguous because ADOPD does not have a one-shot setting; the comparison is against the Qwen3-VL-4B backbone evaluated one-shot. Please rephrase to 'outperforming the backbone's one-shot setting'.
  3. [§3.2, Eq. (9)] The statement 'as no answer is involved, s_g is a verifier-free proxy' is misleading: s_g is computed over the student-generated rollout, which includes answer tokens. The intended meaning is presumably that no ground-truth answer label is used. Please clarify the wording.
  4. [Figure 1(a)] The per-model annotations '+3.31', '+3.56', '+1.25', and '+5.18' are not explained in the caption or text. It should be stated what these deltas represent and whether they are one-shot minus zero-shot accuracy differences.
  5. [References] Several references are cited to arXiv preprints dated 2026; this is acceptable for a preprint, but the list should be checked for consistency with the ICLR-style formatting, particularly the use of 'LLaV A-NeXT' and the VeRL citation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central MMAD result is external, no evaluation label enters training or weighting, and no load-bearing argument reduces to a self-citation.

full rationale

The paper's claimed derivation is self-contained against external evidence. The training objective (Eq. 12) combines token-level matched-reference teacher–student log-ratio advantages (Eq. 8) with a sequence-level weight computed from within-prompt matched-versus-mismatched teacher log-likelihood gaps (Eqs. 9–10). None of these quantities is fitted to, or defined in terms of, the MMAD benchmark labels; evaluation is zero-shot on MMAD using the trained student without references, and the training data (6K Real-IAD pairs) are disjoint from the evaluation benchmarks. The teacher is a frozen external model (Qwen3-VL-32B), so the matched/mismatched contrast is not a self-citation or a uniqueness assumption. The contrastive reference support score s_g is a heuristic proxy, but an unvalidated proxy is a correctness risk, not a circularity: the paper does not define s_g in terms of the outcome it predicts, and the 2.36-point gain over uncalibrated OPD is an external empirical comparison, not a construction identity. The only self-citation (Meng et al., 2024, MoEAD) is contextual background for the IAD task and carries no load-bearing argument. No equation or fitted parameter is renamed as a prediction, and no uniqueness theorem is imported from the authors' prior work. Therefore the derivation does not reduce to its inputs.

Assumptions & free parameters 11 free parameters · 5 assumptions · 0 invented entities

The method's contribution is a training recipe built on standard OPD calculus plus two domain assumptions about reference semantics and transfer. The free parameters are hand-set hyperparameters, none fitted to the MMAD benchmark. The riskiest unpaid input is the assumption that the teacher's matched-mismatched likelihood gap tracks genuine reference-specific support.

free parameters (11)
  • epsilon_w = 1
    Symmetric clipping bound in Eq. 10 for the group-relative reference weight; set by hand with no sensitivity analysis.
  • G (rollouts per prompt) = 8
    Number of student rollouts sampled per prompt in Algorithm 1; controls the mean/std calibration and update variance.
  • beta (EMA decay) = 0.95
    Decay coefficient for the EMA teacher in the OPSD ablation (Table 4); not used in the default frozen-teacher configuration.
  • learning_rate = 1e-6
    Default optimizer learning rate for student updates.
  • training_steps = 60
    Number of optimization steps; no convergence or early-stopping analysis is provided.
  • batch_size = 64 (mini-batch 16)
    Training batch and mini-batch sizes; affect the number of rollouts per update.
  • mask_binarization_threshold = 128
    Threshold used in Appendix B to binarize defect masks; determines which pixels count as defects.
  • min_defect_component_size = 16 pixels
    Minimum connected component size for a defect region to be retained in training data; filters out small anomalies.
  • max_defect_regions = 4
    At most four defect regions are retained per image and montage-organized; limits teacher reference complexity.
  • crop_short_side = 336 pixels
    Privileged regional crop resize; controls the visual resolution of teacher references.
  • max_camera_views_per_instance = 2
    Dataset construction limit to increase instance diversity; affects training distribution.
assumptions (5)
  • standard math Sampled-token reverse-KL gradient identity (Eqs. 13-22) is unbiased for the fixed-prefix gradient when the prefix is stop-gradient.
    Used to justify the token-level advantage in Eq. 8; the Appendix derivation is standard score-function calculus.
  • domain assumption The matched-mismatched teacher log-likelihood gap s_g measures reference-specific support.
    Section 3.2 states s_g is a verifier-free proxy; no independent evidence links s_g to actual reference usage rather than query cues or language priors.
  • domain assumption Matched and mismatched references constructed from ground-truth masks and opposite anomaly states correctly preserve or break the query-reference relation.
    Appendix B builds these crops from defect annotations; the method depends on annotation quality and the availability of paired normal/defective images from the same instance and camera.
  • domain assumption Training on Real-IAD transfers to the four MMAD datasets under zero-shot inference.
    The central benchmark result assumes the distilled inspection strategy survives the distribution shift from Real-IAD to MVTec-AD, VisA, MVTec-LOCO, and GoodsAD.
  • domain assumption The frozen Qwen3-VL-32B teacher's reference-conditioned preferences are a reliable supervision source for the 4B student.
    The whole distillation relies on the teacher's matched-reference distribution being more useful than the student's; no teacher accuracy or failure analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ADOPD: Reference-Privileged On-Policy Distillation for MLLM-Based Industrial Anomaly Detection." pith.science (2026). https://pith.science/paper/YBB5G4O5

@misc{pith2026260809789,
  author       = {Pith},
  title        = {Pith review of: ADOPD: Reference-Privileged On-Policy Distillation for MLLM-Based Industrial Anomaly Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YBB5G4O5}},
  note         = {Machine review of arXiv:2608.09789}
}
read the original abstract

Industrial anomaly detection (IAD) requires identifying fine-grained deviations from normal visual patterns. Multimodal large language models (MLLMs) can improve recognition accuracy by comparing query images with references at inference time, but these benefits rely on additional retrieval and processing. We investigate whether the benefits of reference comparison can instead be internalized in the model parameters. Access to references during training allows a reference-aware teacher to supervise a query-only student. However, the teacher may favor plausible responses based on query cues or language priors rather than valid visual information. We propose ADOPD, a reference-privileged on-policy distillation framework. The teacher evaluates student-generated rollouts under matched and mismatched references. The matched-reference teacher-to-student log-ratio defines the token-level learning direction, specifying what the student should learn. The likelihood gap between the two reference views estimates reference-specific support and calibrates the sequence-level weight. ADOPD achieves 77.31% average accuracy on the MMAD benchmark under zero-shot inference, improving the Qwen3-VL-4B backbone by 6.14 points and outperforming its one-shot setting by 2.64 points. Experiments show that ADOPD learns a fine-grained anomaly inspection strategy from reference comparison. The project will be available at https://github.com/withTai/ADOPD.

Figures

Figures reproduced from arXiv: 2608.09789 by the authors.

Figure 1
Figure 1. Reference-conditioned inference to reference-privileged training. (a) A normal ref￾erence improves all four tested MLLM backbones. (b) Reference-privileged off-policy distilla￾tion trains the student by minimizing the cross-entropy loss on fixed teacher-generated targets. (c) Reference-privileged on-policy distillation evaluates student-generated prefixes with the reference￾aware teacher and updates the student thro… view at source ↗
Figure 2
Figure 2. Overview of ADOPD. For each prompt x, the student πS samples G rollouts. A frozen teacher πT evaluates the same tokens with matched and mismatched references. The matched￾reference teacher–student log-probability ratio gives the token-level advantage A g t . The teacher like￾lihood ratio between the two reference views gives the sequence-level magnitude w g to form Abg t . For each rollout, the frozen teacher πT rec… view at source ↗
Figure 3
Figure 3. MMAD results for ADOPD and Qwen3-VL-4B baselines. (a) Aggregate accuracy, re￾call, precision, and F1 under zero-shot and one-shot inference. (b) Dataset-wise accuracy, recall, precision, and F1 under zero-shot and one-shot inference [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 4 canonical work pages

  1. [1]

    Qwen3-vl technical report.arXiv preprint arXiv:2511.21631, 2025a

    Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, et al. Qwen3-vl technical report.arXiv preprint arXiv:2511.21631, 2025a. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2.5-vl technical report.arXiv pr...

    arXiv

  2. [6]

    Ad-copilot: A vision-language assistant for industrial anomaly detection via visual in-context comparison.arXiv preprint arXiv:2603.13779,

    Xi Jiang, Yue Guo, Jian Li, Yong Liu, Bin-Bin Gao, Hanqiu Deng, Jun Liu, Heng Zhao, Chengjie Wang, and Feng Zheng. Ad-copilot: A vision-language assistant for industrial anomaly detection via visual in-context comparison.arXiv preprint arXiv:2603.13779,

    arXiv

  3. [7]

    Visual contrastive self-distillation.arXiv preprint arXiv:2607.21556,

    Yijun Liang, Yunjie Tian, Yijiang Li, Yuqi Jia, Furong Huang, Tianyi Zhou, and Di Fu. Visual contrastive self-distillation.arXiv preprint arXiv:2607.21556,

    arXiv

  4. [8]

    Improved baselines with visual instruction tuning

    10 Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 26296–26306, 2024a. Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. Llava-next: Improved reasoning, ocr, and world kno...

    work page 2024

  5. [9]

    URLhttps://thinkingmachines.ai/blog/ on-policy-distillation/

    doi: 10.64434/tml.20251026. URLhttps://thinkingmachines.ai/blog/ on-policy-distillation/. Shiyuan Meng, Wenchao Meng, Qihang Zhou, Shizhong Li, Weiye Hou, and Shibo He. Moead: A parameter-efficient model for multi-class anomaly detection. InEuropean Conference on Com- puter Vision, pp. 345–361. Springer,

    doi:10.64434/tml.20251026

  6. [10]

    Agentiad: Tool-augmented single- agent for industrial anomaly detection.arXiv preprint arXiv:2512.13671,

    Junwen Miao, Penghui Du, Yi Liu, Yu Wang, and Yan Wang. Agentiad: Tool-augmented single- agent for industrial anomaly detection.arXiv preprint arXiv:2512.13671,

    arXiv

  7. [11]

    Privileged information distillation for language models.arXiv preprint arXiv:2602.04942,

    Emiliano Penaloza, Dheeraj Vattikonda, Nicolas Gontier, Alexandre Lacoste, Laurent Charlin, and Massimo Caccia. Privileged information distillation for language models.arXiv preprint arXiv:2602.04942,

    arXiv

  8. [12]

    Hybridflow: A flexible and efficient rlhf framework.arXiv preprint arXiv: 2409.19256,

    Guangming Sheng, Chi Zhang, Zilingfeng Ye, Xibin Wu, Wang Zhang, Ru Zhang, Yanghua Peng, Haibin Lin, and Chuan Wu. Hybridflow: A flexible and efficient rlhf framework.arXiv preprint arXiv: 2409.19256,

    arXiv

Show all 16 references
  1. [13]

    Self-distilled rlvr.arXiv preprint arXiv:2604.03128, 2026a

    Chenxu Yang, Chuanyu Qin, Qingyi Si, Minghui Chen, Naibin Gu, Dingyu Yao, Zheng Lin, Weiping Wang, Jiaqi Wang, and Nan Duan. Self-distilled rlvr.arXiv preprint arXiv:2604.03128, 2026a. Wenkai Yang, Weijie Liu, Ruobing Xie, Kai Yang, Saiyong Yang, and Yankai Lin. Learning beyon...

    arXiv

  2. [14]

    Pku-goodsad: A supermarket goods dataset for unsupervised anomaly detection and segmentation.IEEE Robotics and Automation Letters, 9 (3):2008–2015,

    Jian Zhang, Runwei Ding, Miaoju Ban, and Linhui Dai. Pku-goodsad: A supermarket goods dataset for unsupervised anomaly detection and segmentation.IEEE Robotics and Automation Letters, 9 (3):2008–2015,

    2008

  3. [15]

    Anomalyclip: Object-agnostic prompt learning for zero-shot anomaly detection

    Qihang Zhou, Guansong Pang, Yu Tian, Shibo He, and Jiming Chen. Anomalyclip: Object-agnostic prompt learning for zero-shot anomaly detection. InInternational Conference on Learning Rep- resentations, volume 2024, pp. 49705–49737,

    2024

  4. [16]

    Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models.arXiv preprint arXiv:2504.10479,

    Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shenglong Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Jie Shao, et al. Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models.arXiv preprint arXiv:2504.10479,

    arXiv

  5. [2022]

    Anomalyr1: A grpo-based end-to-end mllm for industrial anomaly detection.arXiv preprint arXiv:2504.11914,

    Yuhao Chao, Jie Liu, Jie Tang, and Gangshan Wu. Anomalyr1: A grpo-based end-to-end mllm for industrial anomaly detection.arXiv preprint arXiv:2504.11914,

    arXiv

  6. [2024]

    Glm-4.5 v and glm-4.1 v-thinking: Towards versatile multi- modal reasoning with scalable reinforcement learning.arXiv preprint arXiv:2507.01006,

    Wenyi Hong, Wenmeng Yu, Xiaotao Gu, Guo Wang, Guobing Gan, Haomiao Tang, Jiale Cheng, Ji Qi, Junhui Ji, Lihang Pan, et al. Glm-4.5 v and glm-4.1 v-thinking: Towards versatile multi- modal reasoning with scalable reinforcement learning.arXiv preprint arXiv:2507.01006,

    arXiv

  7. [2025]

    Towards explainable industrial anomaly detection via knowledge-guided latent reasoning.arXiv preprint arXiv:2602.09850, 2026a

    Peng Chen, Chao Huang, Yunkang Cao, Chengliang Liu, Wei Wang, Wenqiang Wang, Mingbo Yang, Li Shen, Wenqi Ren, and Xiaochun Cao. Towards explainable industrial anomaly detection via knowledge-guided latent reasoning.arXiv preprint arXiv:2602.09850, 2026a. Shuimu Chen, Jing Jin,...

    arXiv

  8. [2026]

    Anomalygpt: Detecting industrial anomalies using large vision-language models

    Zhaopeng Gu, Bingke Zhu, Guibo Zhu, Yingying Chen, Ming Tang, and Jinqiao Wang. Anomalygpt: Detecting industrial anomalies using large vision-language models. InProceedings of the AAAI conference on artificial intelligence, volume 38, pp. 1932–1940,

    1932

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.