Pith. sign in

REVIEW 3 major objections 7 minor 1 cited by

MCL-AD: Multimodal Collaboration Learning for Zero-Shot 3D Anomaly Detection

T0 review · 3 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper proposes MCL-AD, a zero-shot 3D anomaly detection framework that fuses point clouds, RGB images, and text prompts; it reports image-level AUROC of 89.0% and point-level AUROC of 97.7% on MVTec3D-AD.

desk verdict The reported zero-shot SOTA is invalidated by the paper's own protocol: training on test-split ground-truth masks and tuning fusion weights on the same test set, so the gains are partly fitted, not predicted. read the letter →

arxiv 2509.10282 v1 pith:XPBSEVS5 submitted 2025-09-12 cs.CV cs.LG

Gang Li , Tianjiao Chen , Mingle Zhou , Min Li , Delong Han , Jin Wan This is my paper · ORCID
classification cs.CVcs.LG
keywords zero-shot3DanomalydetectionmultimodallearningpromptpointcloudRGBimageCLIPMVTec3D-ADEyecandies
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

Zero-shot 3D anomaly detection, finding defects in objects never seen during training, usually reads only point clouds, but the paper argues that the task is better solved by making three modalities talk to one another: point-cloud geometry, RGB appearance, and text semantics. The proposed MCL-AD renders point clouds into depth views, pushes RGB and depth features through a frozen CLIP encoder, and learns separate normal/anomaly text prompts for each branch. A multimodal contrastive loss aligns the two branches on semantic state, and a collaborative modulation mechanism reweights their score maps at test time so texture defects and geometric defects reinforce each other. On MVTec3D-AD the framework reports 89.0% image-level AUROC and 97.7% point-level AUROC, above the point-cloud-only MVP-PCLIP baseline, and it also improves on Eyecandies and in cross-dataset transfer. The claimed payoff is an industrial inspection method that adapts to novel objects without category-specific training.

What carries the argument

The load-bearing object is the object-agnostic decoupled text prompt (ODTP): two pairs of learnable prompt templates, one pair for RGB and one for point clouds, with the class label replaced by the generic word 'object' so the prompts carry no category bias. The multimodal contrastive loss (MCL) makes these prompts interoperable by treating the point-cloud prompt as an anchor, pulling the RGB prompt of the same semantic state closer and pushing the opposite state away. The collaborative modulation mechanism (CMM) is the inference-side counterpart: it fuses the two branches' anomaly score maps as $\frac{\eta\,G_\sigma(M_{\mathrm{rgb}}) + (2-\eta)\,G_\sigma(M_{\mathrm{point}})}{2}$ with a dataset-dependent weight. Multi-view rendering and inverse rendering supply the shared feature space, since a frozen CLIP encoder sees only 2D views.

What would settle it

Run MCL-AD with the same architecture and hyperparameters but without access to any ground-truth masks, replacing the Focal, Dice, and cross-entropy losses in Eqs. (20), (23), (26), and (28) with normal-only objectives, and measure I-AUROC and P-AUROC on MVTec3D-AD. If the scores fall below the reported 89.0% and 97.7% and approach MVP-PCLIP's 71.9% and 87.5%, the gain is explained by test-set supervision rather than by multimodal collaboration.

Watch

Extended reading notes

Core claim

The paper's central claim is that zero-shot 3D anomaly detection is a multimodal alignment problem, not a point-cloud-only problem. MCL-AD renders each point cloud into nine depth views, extracts RGB and depth features with a frozen CLIP image encoder, and learns four text-prompt embeddings: normal and anomalous prompts for the RGB branch, and normal and anomalous prompts for the point-cloud branch. An object-agnostic decoupled prompt template removes category priors, and a multimodal contrastive loss uses the point-cloud prompt as anchor to pull the same-state RGB prompt near while pushing the opposite-state prompt away. At inference, a collaborative modulation mechanism reweights the two branches' score maps with a balance weight. The authors report that this recipe lifts MVTec3D-AD image-level AUROC from 71.9% (MVP-PCLIP) to 89.0% and point-level AUROC from 87.5% to 97.7%, with consistent gains on Eyecandies and in cross-dataset transfer.

Load-bearing premise

The load-bearing premise is that training on the test splits of MVTec3D-AD and Eyecandies, including the ground-truth anomaly masks used in the loss functions, counts as zero-shot because no object-specific normal training split is used; if test-set annotations are labeled training data, the reported gains and the zero-shot claim stand or fall with that allowance.

Editorial extensions

If this is right

  • If the reported numbers are right, point-cloud-only zero-shot 3D detectors are leaving a large margin on the table, and fusing RGB and text at the prompt level becomes the effective baseline.
  • The object-agnostic decoupled prompts should transfer to unseen object categories without per-category prompt engineering, which is what an inspection system facing novel parts needs.
  • The cross-dataset results imply a model can be tuned on one industrial 3D dataset and deployed on another with no target-domain annotation, at a modest accuracy cost.
  • The dataset-dependent fusion weight $\eta$ shows that no single static fusion rule dominates, so adaptive fusion is a necessary component for practical deployment.

Reading between the lines

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

  • Retrain MCL-AD with the ground-truth masks removed from the local losses in Eqs. (20) and (26) and compare; the drop in AUROC would quantify how much of the 'zero-shot' gain comes from test-set supervision rather than multimodal collaboration.
  • The same decoupled-prompt idea could be tested on cheaper 2.5D inputs such as stereo depth or surface normals, where geometry cues are available without a full point-cloud rendering pipeline.
  • A fair cross-method comparison would hold the supervision budget constant; the baselines compared here are not described as training on test-set anomaly masks, so without that constraint the reported margins are not directly attributable to the architecture.
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

3 major / 7 minor

Summary. The paper proposes MCL-AD, a multimodal framework for zero-shot 3D anomaly detection that combines frozen CLIP encoders for RGB images and multi-view rendered depth images with learnable text prompts, a multimodal contrastive loss, and a collaborative modulation mechanism for inference-time fusion. Training optimizes point-cloud and RGB local/global losses plus the contrastive loss; testing uses the fused branch outputs. The authors report state-of-the-art results on MVTec3D-AD and Eyecandies under both within-dataset and cross-dataset settings, with image-level AUROC improving from 71.9% (MVP-PCLIP) to 89.0% and point-level AUROC from 87.5% to 97.7% on MVTec3D-AD.

Significance. If the experimental protocol were valid, the reported gains would be substantial and practically relevant for zero-shot 3D anomaly detection. The paper offers a clearly described architecture, a sensible motivation for combining RGB and point-cloud modalities, and extensive ablations over prompt design and fusion hyperparameters. These strengths, however, do not offset the core problem: the method is trained on the ground-truth anomaly annotations of the test splits, and key hyperparameters are selected using the same test-set metrics. Consequently, the reported numbers do not demonstrate zero-shot generalization, and the central claim of the paper is unsupported.

major comments (3)
  1. [IV-A, Eqs. (20), (23), (26), (28)] The manuscript states that 'the training sets of both datasets only consist of normal samples, so we exclusively use their test sets, which contain point-level annotations for point clouds and pixel-level annotations for images.' The local and global losses in Eqs. (20), (23), (26), and (28) directly supervise the model with the ground-truth masks M_i^(k), M_i and labels G_i^(k), G_i. This means the method uses anomaly annotations from the evaluation data during training, contradicting the abstract's claim of detecting defects 'without relying on labeled training data.' Under this protocol, Tables I and II measure supervised fitting on the test set, not zero-shot generalization.
  2. [III-D, IV-C6, Tables VII and VIII] The fusion weight η and the contrastive-loss weight λ3 are chosen by optimizing the same evaluation metrics on the MVTec3D-AD test set: λ3 is set to 0.8 'based on the experiment in Section IV-C6,' and η is set to 0.8 from the ablation in Table VII. This test-set-based selection of hyperparameters means part of the reported gain is fitted to the evaluation data. A credible zero-shot evaluation would fix these values before seeing the test set, and the paper provides no evidence that the reported results survive such a protocol.
  3. [IV-A, Evaluation Setting] The within-dataset evaluation protocol is underspecified. The text says the model is 'evaluated on test objects that belong to different categories but come from the same dataset,' but it does not state which categories are used for training the prompts versus held out for testing, nor whether the split is per category or per object. Without this information, the within-dataset numbers in Tables I and II are not reproducible, and it is impossible to assess whether the method generalizes across categories or merely memorizes category-specific test annotations.
minor comments (7)
  1. [Eqs. (24) and (29)] The definitions of L_point and L_rgb are written as L_point = F_local^p + L_global^p and L_rgb = F_local^rgb + L_global^rgb, where F_local^p and F_local^rgb are similarity maps rather than scalar losses; please clarify how these terms are reduced to scalar loss values.
  2. [Eq. (35)] In the formula for Score_point, the global feature f_global^rgb appears in the numerator, which appears to be a typo for f_global^point; please correct.
  3. [IV-C1, Table IV] The text states that 'the P-AUROC improving from 60.4% to 97.7%,' but in Table IV the value 60.4 is the I-AUROC of the baseline row; the P-AUROC of that row is not reported. Please correct the description.
  4. [IV-A, Datasets] The sentence 'the training sets of both datasets only consist of normal samples, so we exclusively use their test sets' is confusing because the method is actually trained on the test sets; please rephrase to state precisely what data are used for training and evaluation.
  5. [References [13] and [46]] References [13] and [46] refer to the same paper (AnomalyCLIP); please merge them or cross-reference them to avoid duplication.
  6. [Table I] The pixel-level rows contain incomplete entries such as '(-, )' and appear misaligned; please reformat the table for clarity.
  7. [IV-A, Implementation details] The rotation angles for rendering are given as five values for the X axis and four values for the Y axis, which does not obviously produce the stated 9 views; please clarify how the nine viewpoints are generated.

Circularity Check

2 steps flagged · score 6.0 of 10

The zero-shot SOTA claim is unsubstantiated: MCL-AD trains its learnable prompts on the annotated test splits and selects eta and lambda3 by optimizing the reported metrics on the same test set, so the predictions reduce to supervised fits.

  1. fitted input called prediction [Section IV-A; Eqs. (20), (23), (26), (28)]
    "In our ZS-3D anomaly detection task, the training sets of both datasets only consist of normal samples, so we exclusively use their test sets, which contain point-level annotations for point clouds and pixel-level annotations for images."

    The abstract defines the setting as zero-shot, i.e. 'without relying on labeled training data.' The model's learnable text prompts are optimized with the ground-truth masks and labels M_i^(k), M_i, G_i^(k), G_i in Eqs. (20), (23), (26), and (28), and those same test splits are used to report Tables I and II. The reported I-AUROC, AP, P-AUROC, and AU-PRO are therefore scores of a model fitted to the evaluation data, not zero-shot predictions; the 'prediction' is computed after the model has seen the very annotations it is supposed to predict.

  2. fitted input called prediction [Section III-D Eq. (30); Section IV-C5 Table VII; Section IV-C6 Table VIII]
    "while λ3 is set to 0.8 based on the experiment in Section IV-C6 ... As shown in Tab. VII, the model achieves the best performance when η = 0.8 ... with the best performance observed when λ3 is set to 0.8."

    Both η and λ3 are selected by scanning values and reading the I-AUROC, AP, P-AUROC, and AUPRO numbers in Tables VII and VIII, all computed on the MVTec3D-AD test set. The chosen values are then hard-coded into the fusion equations (36)-(37) and the total loss (30), and the final tables report the resulting peak numbers. Thus part of the reported 'state-of-the-art' performance is an artifact of test-set metric optimization rather than a parameter-free zero-shot prediction.

full rationale

The paper contains two concrete circular reductions, both of the 'fitted input called prediction' form. First, Section IV-A states that the method 'exclusively use[s] their test sets' for training, and the loss equations directly supervise the learnable prompts with test-set ground-truth masks and labels; the same test splits then generate the headline numbers. Second, the hyperparameters η and λ3 are explicitly chosen by ablations on the MVTec3D-AD test set (Tables VII and VIII), so the reported best scores are partially fitted to the evaluation metric. These are not merely self-citation or naming issues: the paper's central claim of zero-shot detection 'without relying on labeled training data' is invalidated by its own protocol. I did not find load-bearing self-citation circularity; the cited AnomalyCLIP and MVP-PCLIP baselines are external prior work, and the architectural modules (MPLM, CMM) could in principle be evaluated under a clean protocol. But as written, the within-dataset and even cross-dataset numbers are obtained from a model whose prompts and fusion weights were set using test-set annotations and test-set metric optimization. A correct zero-shot protocol would fix prompts and hyperparameters before any exposure to the target test split; the paper provides no evidence the reported 89.0 I-AUROC / 97.7 P-AUROC survive that protocol.

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

The central claim rests on a training protocol that uses ground-truth anomaly masks from the test splits, plus hyperparameters tuned on the same test set. No new physical entities are introduced; the framework relies on standard CLIP encoders and hand-designed rendering and fusion choices.

free parameters (4)
  • eta (fusion weight) = 0.8 (on MVTec3D-AD)
    Selected via test-set ablation (Tab. VII); the value for Eyecandies and cross-dataset settings is not reported.
  • lambda3 (multimodal contrastive loss weight) = 0.8
    Selected via test-set ablation (Tab. VIII); lambda1 and lambda2 are fixed to 1.
  • Prompt hyperparameters = prompt length 14, depth 4, insertion layer 9
    Tuned via ablations in Fig. 5 using test-set performance on MVTec3D-AD.
  • MCL margin
    The triplet loss margin in Eq. (16) is not specified in the paper, so the loss is not fully defined.
assumptions (4)
  • ad hoc to paper Ground-truth anomaly masks from the test splits are available for training the prompt learning losses (Eqs. 20 and 26).
    The paper states in Sec. IV-A that it exclusively uses the test sets, which contain anomaly annotations, for model training; this conflicts with the zero-shot framing and is not a standard assumption for zero-shot anomaly detection.
  • domain assumption CLIP frozen encoders provide aligned embeddings for depth-rendered point clouds and RGB images with text prompts.
    The method relies on CLIP's pretrained visual-text alignment to transfer to anomaly detection.
  • domain assumption Inverse rendering maps 2D view features back to 3D without losing anomaly information.
    Used in Eqs. 19 and 34 for point cloud localization; no validation that this mapping preserves all anomaly details.
  • domain assumption The multi-view depth rendering with 9 fixed viewpoints captures all relevant geometric anomalies.
    Rendering parameters are fixed by hand; no analysis of viewpoint coverage.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MCL-AD: Multimodal Collaboration Learning for Zero-Shot 3D Anomaly Detection." pith.science (2026). https://pith.science/paper/XPBSEVS5

@misc{pith2026250910282,
  author       = {Pith},
  title        = {Pith review of: MCL-AD: Multimodal Collaboration Learning for Zero-Shot 3D Anomaly Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XPBSEVS5}},
  note         = {Machine review of arXiv:2509.10282}
}
read the original abstract

Zero-shot 3D (ZS-3D) anomaly detection aims to identify defects in 3D objects without relying on labeled training data, making it especially valuable in scenarios constrained by data scarcity, privacy, or high annotation cost. However, most existing methods focus exclusively on point clouds, neglecting the rich semantic cues available from complementary modalities such as RGB images and texts priors. This paper introduces MCL-AD, a novel framework that leverages multimodal collaboration learning across point clouds, RGB images, and texts semantics to achieve superior zero-shot 3D anomaly detection. Specifically, we propose a Multimodal Prompt Learning Mechanism (MPLM) that enhances the intra-modal representation capability and inter-modal collaborative learning by introducing an object-agnostic decoupled text prompt and a multimodal contrastive loss. In addition, a collaborative modulation mechanism (CMM) is proposed to fully leverage the complementary representations of point clouds and RGB images by jointly modulating the RGB image-guided and point cloud-guided branches. Extensive experiments demonstrate that the proposed MCL-AD framework achieves state-of-the-art performance in ZS-3D anomaly detection.

Figures

Figures reproduced from arXiv: 2509.10282 by the authors.

Figure 1
Figure 1. Comparison between existing CLIP-based ZS-3D anomaly detection methods and the proposed MCL-AD framework. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The framework of the proposed MCL-AD. (a) During training, the Multimodal Prompt Learning Mechanism (MPLM) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visualization results on the MVTec3D-AD and Eyecandies datasets. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visualization of the effectiveness of the key module. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Ablation Study on Textual Prompt Hyperparameters. (a) is the length of normal learnable prompt, (b) is the length of [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Back to Point: Exploring Point-Language Models for Zero-Shot 3D Anomaly Detection

    cs.CV 2026-03 conditional novelty 6.0 of 10

    Direct point-language alignment with multi-granularity patches and learnable geometric descriptors yields strong zero-shot 3D anomaly localization on Real3D-AD and Anomaly-ShapeNet, though object-level scores lag PointAD.

Reference graph

Works this paper leans on

49 extracted references · 43 canonical work pages · cited by 1 Pith paper

  1. [1]

    Mvtec AD - A comprehensive real-world dataset for unsupervised anomaly detection,

    P. Bergmann, M. Fauser, D. Sattlegger, and C. Steger, “Mvtec AD - A comprehensive real-world dataset for unsupervised anomaly detection,” inCVPR. Computer Vision Foundation / IEEE, 2019, pp. 9592–9600

    work page 2019

  2. [2]

    Revisiting reverse distillation for anomaly detection,

    T. D. Tien, A. T. Nguyen, N. H. Tran, T. D. Huy, S. T. M. Duong, C. D. T. Nguyen, and S. Q. H. Truong, “Revisiting reverse distillation for anomaly detection,” in CVPR. IEEE, 2023, pp. 24 511–24 520

    work page 2023

  3. [3]

    Uninformed students: Student-teacher anomaly detec- tion with discriminative latent embeddings,

    P. Bergmann, M. Fauser, D. Sattlegger, and C. Steger, “Uninformed students: Student-teacher anomaly detec- tion with discriminative latent embeddings,” inCVPR. Computer Vision Foundation / IEEE, 2020, pp. 4182– 4191

    work page 2020

  4. [4]

    Explainable deep one-class classification,

    P. Liznerski, L. Ruff, R. A. Vandermeulen, B. J. Franks, M. Kloft, and K. M ¨uller, “Explainable deep one-class classification,” inICLR. OpenReview.net, 2021. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13

    work page 2021

  5. [5]

    Ex- plainable deep few-shot anomaly detection with deviation networks,

    G. Pang, C. Ding, C. Shen, and A. van den Hengel, “Ex- plainable deep few-shot anomaly detection with deviation networks,”CoRR, vol. abs/2108.00462, 2021

    arXiv 2021

  6. [6]

    Deep learning for anomaly detection: A review,

    G. Pang, C. Shen, L. Cao, and A. van den Hengel, “Deep learning for anomaly detection: A review,”ACM Comput. Surv., vol. 54, no. 2, pp. 38:1–38:38, 2022

    work page 2022

  7. [7]

    Towards total recall in industrial anomaly detection,

    K. Roth, L. Pemula, J. Zepeda, B. Sch ¨olkopf, T. Brox, and P. V . Gehler, “Towards total recall in industrial anomaly detection,” inCVPR. IEEE, 2022, pp. 14 298– 14 308

    work page 2022

  8. [8]

    Anomaly detection via reverse distillation from one-class embedding,

    H. Deng and X. Li, “Anomaly detection via reverse distillation from one-class embedding,” inCVPR. IEEE, 2022, pp. 9727–9736

    work page 2022

Show all 49 references
  1. [9]

    Winclip: Zero-/few-shot anomaly clas- sification and segmentation,

    J. Jeong, Y . Zou, T. Kim, D. Zhang, A. Ravichandran, and O. Dabeer, “Winclip: Zero-/few-shot anomaly clas- sification and segmentation,” inCVPR. IEEE, 2023, pp. 19 606–19 616

    2023

  2. [10]

    Pushing the limits of fewshot anomaly detection in industry vision: Graphcore,

    G. Xie, J. Wang, J. Liu, Y . Jin, and F. Zheng, “Pushing the limits of fewshot anomaly detection in industry vision: Graphcore,” inICLR. OpenReview.net, 2023

    2023

  3. [11]

    Anomaly detection under distribution shift,

    T. Cao, J. Zhu, and G. Pang, “Anomaly detection under distribution shift,” inICCV. IEEE, 2023, pp. 6488–6500

    2023

  4. [12]

    Anomalygpt: Detecting industrial anomalies using large vision-language models,

    Z. Gu, B. Zhu, G. Zhu, Y . Chen, M. Tang, and J. Wang, “Anomalygpt: Detecting industrial anomalies using large vision-language models,” inAAAI. AAAI Press, 2024, pp. 1932–1940

    2024

  5. [14]

    Registration based few-shot anomaly detection,

    C. Huang, H. Guan, A. Jiang, Y . Zhang, M. W. Spratling, and Y . Wang, “Registration based few-shot anomaly detection,” inECCV (24), ser. Lecture Notes in Computer Science, vol. 13684. Springer, 2022, pp. 303–319

    2022

  6. [15]

    A unified model for multi-class anomaly detection,

    Z. You, L. Cui, Y . Shen, K. Yang, X. Lu, Y . Zheng, and X. Le, “A unified model for multi-class anomaly detection,” inNeurIPS, 2022

    2022

  7. [16]

    When and how does known class help discover unknown ones? provable understanding through spectral analysis,

    Y . Sun, Z. Shi, Y . Liang, and Y . Li, “When and how does known class help discover unknown ones? provable understanding through spectral analysis,” inICML, ser. Proceedings of Machine Learning Research, vol. 202. PMLR, 2023, pp. 33 014–33 043

    2023

  8. [17]

    Self-supervised pseudo multi-class pre-training for unsupervised anomaly detection and segmentation in medical images,

    Y . Tian, F. Liu, G. Pang, Y . Chen, Y . Liu, J. W. Verjans, R. Singh, and G. Carneiro, “Self-supervised pseudo multi-class pre-training for unsupervised anomaly detection and segmentation in medical images,”Medical Image Anal., vol. 90, p. 102930, 2023

    2023

  9. [18]

    A survey on visual anomaly detec- tion: Challenge, approach, and prospect,

    Y . Cao, X. Xu, J. Zhang, Y . Cheng, X. Huang, G. Pang, and W. Shen, “A survey on visual anomaly detec- tion: Challenge, approach, and prospect,”CoRR, vol. abs/2401.16402, 2024

    2024 arXiv

  10. [19]

    Easynet: An easy network for 3d industrial anomaly detection,

    R. Chen, G. Xie, J. Liu, J. Wang, Z. Luo, J. Wang, and F. Zheng, “Easynet: An easy network for 3d industrial anomaly detection,” inACM Multimedia. ACM, 2023, pp. 7038–7046

    2023

  11. [20]

    The mvtec 3d-ad dataset for unsupervised 3d anomaly de- tection and localization,

    P. Bergmann, X. Jin, D. Sattlegger, and C. Steger, “The mvtec 3d-ad dataset for unsupervised 3d anomaly de- tection and localization,” inVISIGRAPP (5: VISAPP). SCITEPRESS, 2022, pp. 202–213

    2022

  12. [21]

    Simplenet: A simple network for image anomaly detection and localization,

    Z. Liu, Y . Zhou, Y . Xu, and Z. Wang, “Simplenet: A simple network for image anomaly detection and localization,” inCVPR. IEEE, 2023, pp. 20 402–20 411

    2023

  13. [22]

    DSR - A dual subspace re-projection network for surface anomaly detection,

    V . Zavrtanik, M. Kristan, and D. Skocaj, “DSR - A dual subspace re-projection network for surface anomaly detection,” inECCV (31), ser. Lecture Notes in Computer Science, vol. 13691. Springer, 2022, pp. 539–554

    2022

  14. [23]

    Prototypical residual networks for anomaly detection and localization,

    H. Zhang, Z. Wu, Z. Wang, Z. Chen, and Y . Jiang, “Prototypical residual networks for anomaly detection and localization,” inCVPR. IEEE, 2023, pp. 16 281– 16 291

    2023

  15. [24]

    Back to the feature: classical 3d features are (almost) all you need for 3d anomaly detection,

    E. Horwitz and Y . Hoshen, “Back to the feature: classical 3d features are (almost) all you need for 3d anomaly detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 2968–2977

    2023

  16. [25]

    Multimodal industrial anomaly detection via hybrid fusion,

    Y . Wang, J. Peng, J. Zhang, R. Yi, Y . Wang, and C. Wang, “Multimodal industrial anomaly detection via hybrid fusion,” inCVPR. IEEE, 2023, pp. 8032–8041

    2023

  17. [26]

    Shape- guided dual-memory learning for 3d anomaly detection,

    Y . Chu, C. Liu, T. Hsieh, H. Chen, and T. Liu, “Shape- guided dual-memory learning for 3d anomaly detection,” inICML, ser. Proceedings of Machine Learning Re- search, vol. 202. PMLR, 2023, pp. 6185–6194

    2023

  18. [27]

    To- wards zero-shot point cloud anomaly detection: A multi- view projection framework,

    Y . Cheng, Y . Cao, G. Xie, Z. Lu, and W. Shen, “To- wards zero-shot point cloud anomaly detection: A multi- view projection framework,”CoRR, vol. abs/2409.13162, 2024

    2024 arXiv

  19. [28]

    Beyond dents and scratches: Logical constraints in unsupervised anomaly detection and local- ization,

    P. Bergmann, K. Batzner, M. Fauser, D. Sattlegger, and C. Steger, “Beyond dents and scratches: Logical constraints in unsupervised anomaly detection and local- ization,”Int. J. Comput. Vis., vol. 130, no. 4, pp. 947– 969, 2022

    2022

  20. [29]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervision,” in ICML, ser. Proceedings of Machine Learning Research, vol. 139...

    2021

  21. [30]

    Self-supervised predictive convolutional attentive block for anomaly de- tection,

    N. Ristea, N. Madan, R. T. Ionescu, K. Nasrollahi, F. S. Khan, T. B. Moeslund, and M. Shah, “Self-supervised predictive convolutional attentive block for anomaly de- tection,” inCVPR. IEEE, 2022, pp. 13 566–13 576

    2022

  22. [31]

    Pointad: Comprehending 3d anomalies from points and pixels for zero-shot 3d anomaly detection,

    Q. Zhou, J. Yan, S. He, W. Meng, and J. Chen, “Pointad: Comprehending 3d anomalies from points and pixels for zero-shot 3d anomaly detection,” inNeurIPS, 2024

    2024

  23. [32]

    Ms- flow: Multiscale flow-based framework for unsupervised anomaly detection,

    Y . Zhou, X. Xu, J. Song, F. Shen, and H. T. Shen, “Ms- flow: Multiscale flow-based framework for unsupervised anomaly detection,”IEEE Trans. Neural Networks Learn. Syst., vol. 36, no. 2, pp. 2437–2450, 2025

    2025

  24. [33]

    The eyecandies dataset for unsuper- vised multimodal anomaly detection and localization,

    L. Bonfiglioli, M. Toschi, D. Silvestri, N. Fioraio, and D. D. Gregorio, “The eyecandies dataset for unsuper- vised multimodal anomaly detection and localization,” inACCV (5), ser. Lecture Notes in Computer Science, vol. 13845. Springer, 2022, pp. 459–475

    2022

  25. [34]

    Anomaly detection in 3d point clouds using deep geometric descriptors,

    P. Bergmann and D. Sattlegger, “Anomaly detection in 3d point clouds using deep geometric descriptors,” inWACV. IEEE, 2023, pp. 2612–2622

    2023

  26. [35]

    Asymmetric student-teacher networks for industrial anomaly detection,

    M. Rudolph, T. Wehrbein, B. Rosenhahn, and B. Wandt, JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14 “Asymmetric student-teacher networks for industrial anomaly detection,” inWACV. IEEE, 2023, pp. 2591– 2601

    2021

  27. [36]

    Towards scalable 3d anomaly detection and localization: A benchmark via 3d anomaly synthesis and A self- supervised learning network,

    W. Li, X. Xu, Y . Gu, B. Zheng, S. Gao, and Y . Wu, “Towards scalable 3d anomaly detection and localization: A benchmark via 3d anomaly synthesis and A self- supervised learning network,” inCVPR. IEEE, 2024, pp. 22 207–22 216

    2024

  28. [37]

    Cheating depth: Enhancing 3d surface anomaly detection via depth simulation,

    V . Zavrtanik, M. Kristan, and D. Skocaj, “Cheating depth: Enhancing 3d surface anomaly detection via depth simulation,” inWACV. IEEE, 2024, pp. 2153–2161

    2024

  29. [38]

    Learning to prompt for vision-language models,

    K. Zhou, J. Yang, C. C. Loy, and Z. Liu, “Learning to prompt for vision-language models,”Int. J. Comput. Vis., vol. 130, no. 9, pp. 2337–2348, 2022

    2022

  30. [39]

    Maple: Multi-modal prompt learning,

    M. U. Khattak, H. A. Rasheed, M. Maaz, S. H. Khan, and F. S. Khan, “Maple: Multi-modal prompt learning,” inCVPR. IEEE, 2023, pp. 19 113–19 122

    2023

  31. [40]

    Multimodal industrial anomaly detection by crossmodal feature mapping,

    A. Costanzino, P. Z. Ramirez, G. Lisanti, and L. D. Stefano, “Multimodal industrial anomaly detection by crossmodal feature mapping,” inCVPR. IEEE, 2024, pp. 17 234–17 243

    2024

  32. [41]

    A zero-/few-shot anomaly classification and segmentation method for CVPR 2023 V AND workshop challenge tracks 1&2: 1st place on zero-shot AD and 4th place on few-shot AD,

    X. Chen, Y . Han, and J. Zhang, “A zero-/few-shot anomaly classification and segmentation method for CVPR 2023 V AND workshop challenge tracks 1&2: 1st place on zero-shot AD and 4th place on few-shot AD,” CoRR, vol. abs/2305.17382, 2023

    2023 arXiv

  33. [42]

    Segment any anomaly without training via hy- brid prompt regularization,

    Y . Cao, X. Xu, C. Sun, Y . Cheng, Z. Du, L. Gao, and W. Shen, “Segment any anomaly without training via hy- brid prompt regularization,”CoRR, vol. abs/2305.10724, 2023

    2023 arXiv

  34. [43]

    Complementary pseudo multimodal feature for point cloud anomaly detection,

    Y . Cao, X. Xu, and W. Shen, “Complementary pseudo multimodal feature for point cloud anomaly detection,” Pattern Recognit., vol. 156, p. 110761, 2024

    2024

  35. [44]

    Masked autoencoders for point cloud self- supervised learning,

    Y . Pang, W. Wang, F. E. H. Tay, W. Liu, Y . Tian, and L. Yuan, “Masked autoencoders for point cloud self- supervised learning,” inECCV (2), ser. Lecture Notes in Computer Science, vol. 13662. Springer, 2022, pp. 604–621

    2022

  36. [45]

    Pointclip V2: prompting CLIP and GPT for powerful 3d open-world learning,

    X. Zhu, R. Zhang, B. He, Z. Guo, Z. Zeng, Z. Qin, S. Zhang, and P. Gao, “Pointclip V2: prompting CLIP and GPT for powerful 3d open-world learning,” inICCV. IEEE, 2023, pp. 2639–2650

    2023

  37. [46]

    Anoma- lyclip: Object-agnostic prompt learning for zero-shot anomaly detection,

    Q. Zhou, G. Pang, Y . Tian, S. He, and J. Chen, “Anoma- lyclip: Object-agnostic prompt learning for zero-shot anomaly detection,” inICLR. OpenReview.net, 2024

    2024

  38. [47]

    Zero-shot learning on 3d point cloud objects and beyond,

    A. Cheraghian, S. Rahman, T. F. Chowdhury, D. Camp- bell, and L. Petersson, “Zero-shot learning on 3d point cloud objects and beyond,”Int. J. Comput. Vis., vol. 130, no. 10, pp. 2364–2384, 2022

    2022

  39. [48]

    Progres- sive boundary guided anomaly synthesis for industrial anomaly detection,

    Q. Chen, H. Luo, H. Gao, C. Lv, and Z. Zhang, “Progres- sive boundary guided anomaly synthesis for industrial anomaly detection,”IEEE Trans. Circuits Syst. Video Technol., vol. 35, no. 2, pp. 1193–1208, 2025

    2025

  40. [49]

    Normal im- age guided segmentation framework for unsupervised anomaly detection,

    P. Xing, Y . Sun, D. Zeng, and Z. Li, “Normal im- age guided segmentation framework for unsupervised anomaly detection,”IEEE Trans. Circuits Syst. Video Technol., vol. 34, no. 6, pp. 4639–4652, 2024

    2024

  41. [50]

    Learning global-local correspondence with semantic bottleneck for logical anomaly detection,

    H. Yao, W. Yu, W. Luo, Z. Qiang, D. Luo, and X. Zhang, “Learning global-local correspondence with semantic bottleneck for logical anomaly detection,”IEEE Trans. Circuits Syst. Video Technol., vol. 34, no. 5, pp. 3589– 3605, 2024

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.