Pith. sign in

REVIEW 3 major objections 5 minor 41 references

DeCo: Zero-Shot Industrial Anomaly Generation through Decoupling and Recoupling

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read DeCo claims that splitting anomaly structure from product texture, then recombining them, makes zero-shot synthetic industrial defects realistic enough to train downstream detectors.

desk verdict Solid, clearly-written anomaly generation paper with a genuinely new LoRA routing mechanism, but the headline SOTA claim is under-cut by missing comparisons to the closest reference-based baselines and test-set hyperparameter tuning. read the letter →

arxiv 2608.07904 v2 pith:6BUKX4WI submitted 2026-08-08 cs.CV

classification cs.CV
keywords zero-shotanomalygenerationindustrialinspectionsynthesisdiffusionmodelsLoRAcompositiondecouplingandrecouplingdetectionMVTecAD
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that synthetic industrial defects look wrong because existing methods either describe the defect in words or copy it from a reference image without separating the defect's geometry from the product it came from. DeCo's answer is to split the job: first decouple the defect's structure from the source product and bind it to a dedicated anomaly token in a diffusion model, then explicitly recouple that structure with the normal texture of a target product. The paper reports that detectors trained on DeCo-generated images improve pixel-level average precision by 5.1 points on MVTec AD and 8.2 points on VisA over existing zero-shot generation, and that its anomaly localization reaches 68.0 P-AP and 64.5 P-F1 on MVTec AD. If correct, this would make zero-shot synthetic defects usable as training data for industrial inspection without collecting real images of each product's failures.

What carries the argument

Dual-Routing Flow (DR-Flow) with AP-Routing and QK-Routing. AP-Routing separates the abnormal text token from the product token so each is transformed only by its own LoRA; QK-Routing computes image queries with the anomaly LoRA and image keys and values with the frozen product LoRA, forcing the anomaly LoRA to encode structure rather than texture. Product-Invariant Flow (PI-Flow) computes a residual velocity by subtracting the product-only prediction, and Product Compatibility Correction (PCC) extrapolates the final prediction away from the normal-product prediction. Together these mechanisms decide what is learned by which adapter and where the learned defect is injected during generation.

What would settle it

Run the same DeCo pipeline with the routing reversed, putting the anomaly LoRA on the key and value projections and the product LoRA on the query projection, on the same MVTec reference-target pairs; if reversed routing matches or exceeds the reported 68.0 P-AP, the claimed query-controls-structure role is not doing the required work.

Watch

Extended reading notes

Core claim

DeCo establishes text-structure binding in Stable Diffusion 3 by training two LoRA adapters with complementary roles. A frozen Product LoRA encodes the normal product into the key and value projections, while a trainable anomaly LoRA acts only on the query projection, on the premise that queries determine spatial layout while keys and values carry appearance. Dual-Routing Flow (DR-Flow) routes the abnormal text token exclusively through the anomaly LoRA and the product token through the product LoRA; Product-Invariant Flow subtracts the product-only velocity as a residual target so the anomaly token cannot absorb source-product texture. At inference the two LoRAs are merged and Product Compatibility Correction extrapolates the denoising velocity away from the product-only prediction, which the paper says compensates for mismatches when an anomaly learned on one product is transplanted to another. The paper's evidence is that downstream U-Net detectors trained on DeCo images beat all compared zero-shot and synthesis baselines on MVTec AD, VisA, and Real-IAD.

Load-bearing premise

The load-bearing premise is that in the image-generation model's attention mechanism, one projection sets where things go while the other two set how things look, and that this split survives for abstract industrial defects; if that assumption is wrong for defects, the anomaly adapter can silently copy the source product's texture.

Editorial extensions

If this is right

  • A defect structure learned from one set of products can be transplanted to a different product category with no real defect images of the target, because structure and texture are bound to separate tokens.
  • Downstream detectors gain most at the pixel level, where synthetic-data quality matters most: the paper reports +5.1 P-AP on MVTec AD and +8.2 P-AP on VisA.
  • The same separation makes generated anomalies more controllable: changing the product LoRA changes the surface while the anomaly token keeps its geometry.
  • Because the method needs only normal images of the target product, cold-start inspection for new product lines becomes feasible without waiting for real defects to be collected.

Reading between the lines

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

  • If the query/key/value role split holds for abstract industrial defects, the same decouple-recouple recipe could transfer other non-semantic visual structures, such as tool marks, wear patterns, or biological textures, without a text description of the anomaly.
  • The paper's full-rank anomaly LoRA result suggests that when a diffusion model lacks semantic priors for a target concept, adapter capacity, not just routing, is the binding constraint, so low-rank LoRA composition methods may systematically underperform on abstract industrial structures.
  • Product Compatibility Correction is an inference-time compatibility dial; a testable extension would be to make its strength per anomaly-product pair, or learned, rather than a global constant.
  • A stronger decoupling test would be to remove reference masks at inference and check whether the anomaly token still localizes the defect structure, since masks are currently borrowed from the reference set.
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 / 5 minor

Summary. The paper introduces DeCo, a zero-shot industrial anomaly generation method that uses a reference product's anomaly images (with masks) to synthesize anomalies on a target product for which only normal images are available. DeCo trains a frozen Product LoRA (P-LoRA) and a trainable Anomaly LoRA (A-LoRA). Anomaly acquisition is guided by Dual-Routing Flow (DR-Flow), which splits text tokens via AP-Routing and restricts the A-LoRA to the query projection via QK-Routing, and by Product-Invariant Flow (PI-Flow), which subtracts a product-only velocity to define a residual anomaly flow. At inference the two LoRAs are injected jointly, and Product Compatibility Correction (PCC) extrapolates the velocity prediction to compensate for weak anomaly visibility. The paper reports generation-quality metrics on MVTec AD and downstream anomaly-detection results on MVTec AD, VisA, and Real-IAD, claiming a new state of the art, including a 5.1% P-AP gain over AnoStyler on MVTec AD and an 8.2% P-AP gain on VisA.

Significance. If the reported results are substantiated, DeCo would be a practically useful contribution: it addresses a real cold-start problem, the method is described with complete equations, each component is ablated, the code is promised to be available, and the cross-dataset generalization experiments (VisA and Real-IAD) use fixed hyperparameters without retuning. The structural rather than semantic binding of abstract anomalies is a sensible idea and is more general than the semantic-level decoupling used in prior LoRA-composition work. However, the central claims of state-of-the-art performance and of clean product-invariant decoupling are currently supported by incomplete comparative evidence and by hyperparameters selected on the same benchmark used for the headline numbers.

major comments (3)
  1. [Sec. 2.1 and Sec. 5.1, Tables 1–3 and 12] The two reference-based zero-shot anomaly generation methods discussed in Section 2.1, AnomalyControl [14] and AnomaMagic [19], are never included in any quantitative comparison. Since DeCo is itself a reference-based method that consumes reference anomaly images, these are among its closest competitors. Without them, the paper's claim of a 'new state-of-the-art' is not established; the reported gains might be against a weaker subset of baselines. Please add these methods under the same downstream detection protocol, or explicitly justify their exclusion with concrete reasons.
  2. [Sec. 5.1, Tables 9–10] The two main hyperparameters of DeCo, λ_PI = 0.1 and ω = 2, are selected by maximizing downstream metrics on MVTec AD, and the headline MVTec results in Tables 2 and 4 are obtained with these same values. This makes the reported 68.0% P-AP partly a fitted value on the test set. The VisA and Real-IAD results with fixed hyperparameters are a valuable independent check, but they do not certify the MVTec improvement. Please report a hyperparameter-selection procedure that does not use the test set (for example, selection on a validation split), or provide per-seed means and variances across repeated runs, or otherwise quantify the sensitivity of Table 2 to the chosen values.
  3. [Sec. 4.3, Eq. (12)] The definition v_PI = v_t − û_p assumes that the residual after subtracting the product-only flow is exactly the 'pure anomaly' and is product-invariant. This is the central assumption behind the PI-Flow objective in Eq. (13), but the paper provides only qualitative visual evidence (Fig. 3) that the residual excludes product-specific appearance. Since the decoupling claim is load-bearing for the method, please provide a direct test: for example, measure the similarity between v_PI and the product-specific flow û_p across different product pairs, or evaluate whether an A-LoRA trained with v_PI transfers anomalies across products without leaking source-product texture in a quantitative metric.
minor comments (5)
  1. [References [8] and [9]] References [8] and [9] appear to be the same paper (Duan et al., AAAI 2023); please remove the duplicate or replace one with the intended different source.
  2. [Table 5] The entries '30 + k×11' and '30 + k×12' in Table 5 do not define k; please state that k is the number of anomaly categories/types used for training the A-LoRA.
  3. [Sec. 4.1, paragraph after Eq. (6)] The sentence 'A-LoRA,stands for Anomaly LoRA' has a typo (missing space after the comma); please correct it.
  4. [Table 1 and Sec. 5.1] The metric 'IC-L' is introduced as 'IC-LPIPS' in Section 5.1 but never defined in the table caption; please add a one-line definition so Table 1 is self-contained.
  5. [Appendix A, 'Architecture and Implementation Details'] The term 'full-rank A-LoRA' is unusual because LoRA is defined as a low-rank adaptation in Eq. (2); please clarify that full rank here means the update matrices have rank equal to the projection dimension, and state the memory and storage implications of this choice.

Circularity Check

1 steps flagged · score 4.0 of 10

MVTec headline gain is partly a tuned value; no equation-level circularity, and VisA/Real-IAD provide independent checks.

  1. fitted input called prediction [Sec. 5.5 (Tables 9 and 10) and Sec. 5.3 headline MVTec result]
    "Setting λPI = 0.1 effectively limits this incorrect binding, ensuring the token captures the anomaly structure and achieving the best P-AP (68.0%). ... Setting ω = 2 effectively pushes the prediction away from the product-only prediction v̂p. This successfully compensates for the incompatibility, ensures clear anomaly generation, and achieves the highest downstream metrics (98.0% I-AUC and 68.0% P-AP). Therefore, we adopt ω = 2 as the optimal default."

    The paper tunes λPI and ω on the MVTec AD benchmark itself, choosing the values that give the best P-AP, and then uses the resulting 68.0% P-AP / 5.3% margin as the headline evidence that DeCo 'outperforms' AnoStyler. The MVTec improvement is therefore the maximum of the tuning grid, not an independent prediction; the same metric was used to select the hyperparameters. The VisA and Real-IAD experiments reuse these values without re-tuning, providing an out-of-sample check, so the circularity is partial rather than total.

full rationale

The derivation chain of DeCo is largely self-contained: DR-Flow (Eqs. 9-11) and PI-Flow (Eqs. 12-13) define training objectives from the flow-matching loss, and the Q/K/V role assignment is imported from external papers [3,4,34], not from the authors' own prior work. The only self-citation (SEAS, ref [5]) is a few-shot baseline and is not load-bearing for the central claim. No equation reduces to a fitted parameter by definition, and the downstream detection on real MVTec/VisA/Real-IAD anomalies is an external benchmark. The one partial circularity is the selection of λPI and ω on the MVTec test set (Tabs. 9-10) before reporting the 5.1% MVTec gain; the VisA and Real-IAD fixed-hyperparameter results mitigate this. The absence of AnomalyControl and AnomaMagic from the quantitative tables is a completeness-of-evidence concern, but it is a comparison-coverage issue, not circularity.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The method relies on several domain assumptions about attention-role separation and the residual flow target. The two load-bearing free parameters (lambda_PI, omega) are tuned on the MVTec benchmark. No fundamentally new physical or conceptual entities are introduced; 'abnormal token' and 'anomaly structure' are operational concepts rather than invented entities.

free parameters (5)
  • PI-Flow loss weight lambda_PI = 0.1
    Tuned via ablation on MVTec AD (Tab. 9); selected for best P-AP.
  • PCC scale omega = 2
    Tuned via ablation on MVTec AD (Tab. 10); selected for best P-AP.
  • P-LoRA rank = 64
    Chosen; Appendix B shows full-rank A-LoRA is better but P-LoRA rank is not varied.
  • A-LoRA rank = full rank
    Chosen; Appendix B shows rank 64 degrades P-AP, so full rank is adopted.
  • CFG scale = 5.0
    Standard classifier-free guidance scale, not central to the claim.
assumptions (6)
  • domain assumption In SD3 self-attention, Q controls spatial structure while K/V provide texture
    Stated in Sec 3 and used in QK-Routing (Eq. 10).
  • domain assumption P-LoRA trained on normal images produces a velocity that solely reconstructs the normal product
    Sec 4.3: 'Since P-LoRA is fixed and trained on normal images, v_p is a constant velocity that solely reconstructs the normal product.'
  • ad hoc to paper The residual v_PI = v_t - v_p represents pure anomaly structure
    Defined in Eq. 12 and used as training target for A-LoRA; no proof that the subtraction removes all product texture.
  • domain assumption The abnormal text token <anomaly> can bind an abstract anomaly structure without semantic priors
    Core to DR-Flow; paper argues industrial anomalies lack semantic priors, so structural binding is needed.
  • domain assumption Reference ground-truth masks are available and correctly localize anomalies
    Masks used in L_DR (Eq. 11), L_PI (Eq. 13), and inference blending (Eq. 17).
  • domain assumption Freezing P-LoRA during A-LoRA training and merging LoRAs at inference preserves each concept
    Hybrid injection Eq. 15 assumes additive LoRA weights remain valid.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeCo: Zero-Shot Industrial Anomaly Generation through Decoupling and Recoupling." pith.science (2026). https://pith.science/paper/6BUKX4WI

@misc{pith2026260807904,
  author       = {Pith},
  title        = {Pith review of: DeCo: Zero-Shot Industrial Anomaly Generation through Decoupling and Recoupling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6BUKX4WI}},
  note         = {Machine review of arXiv:2608.07904}
}
read the original abstract

Industrial anomaly inspection is severely hindered by the scarcity of real anomalous data. Zero-shot industrial anomaly generation addresses this by generating anomalies on specific products without requiring any of their real anomalous images. However, existing methods suffer from two critical limitations, i.e., inaccurate anomaly information acquisition and uncontrolled anomaly-product fusion. To overcome these challenges, we propose DeCo, which decouples the anomaly structure from its source product, and explicitly recouples it with the normal textures of the target product. During anomaly information acquisition, Dual-Routing Flow (DR-Flow) binds the texture-invariant anomaly structure to an abnormal token, while a parallel constraint, Product-Invariant Flow (PI-Flow), prevents the abnormal token from binding the source product. During anomaly-product fusion, we propose a hybrid injection to recouple the acquired anomaly structure with the target product, and Product Compatibility Correction (PCC) to compensate for the incompatibility between the acquired anomaly structure and the product. Extensive experiments demonstrate that DeCo establishes a new state-of-the-art. Training downstream detection models on our generated data yields massive pixel AP improvements of 5.1% on MVTec AD and 8.2% on VisA. Code is available at https://github.com/HUST-SLOW/DeCo.

Figures

Figures reproduced from arXiv: 2608.07904 by the authors.

Figure 1
Figure 1. (a) Text-based methods face a semantic gap when using fixed text de￾scriptions, yielding unrealistic patterns for complex anomalies. (b) Reference-based methods struggle to disentangle the anomaly from the original product, causing un￾wanted textures to leak into the target generation. (c) Our method (Ours) explicitly decouples the anomaly structure from its original product, and seamlessly recouples it with the spe… view at source ↗
Figure 2
Figure 2. Training pipeline of DeCo. It employs a Dual-LoRA design with a frozen Product LoRA (blue, pre-trained on the left) and a trainable Anomaly LoRA (green). DR-Flow (Middle) binds the anomaly structure to the abnormal token via AP￾Routing and QK-Routing. PI-Flow (Right) acts as a constraint to prevent the ab￾normal token from directly inheriting product features from the source reference. of T . We acquire anomaly info… view at source ↗
Figure 3
Figure 3. (I) PI-Flow. (II) Without PI-Flow, the generated anomalies may carry over the source product’s appearance. (K) and value (V ) provide the visual texture. By computing KI and VI using only the frozen P-LoRA, we limit the available textures to the normal product. At the same time, we compute QI using the trainable A-LoRA. Through this design, the trainable A-LoRA extracts the reference anomaly’s structure into QI . Wi… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Zero-shot anomaly generation using the same reference anomaly. (a) Across different products, the anomaly varies in visibility. (b) With PCC to address the in￾compatibility, the anomaly can be generated well. variation arises because the two LoRAs are trained independe…
Figure 5
Figure 5. Figure 5: Comparison of generated anomaly images. with results summarized in Tab. 2. By establishing text-structure binding via DR-Flow, our method achieves 68.0% in P-AP and 64.5% in P-F1, outperform￾ing the zero-shot method AnoStyler by 5.3% and 3.8%. It also obtains a PRO sco…
Figure 6
Figure 6. Figure 6: Qualitative results of generated anomalies. [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]
Figure 7
Figure 7. Figure 7: More generated images. The layout follows [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]
Figure 8
Figure 8. Figure 8: More generated images. The layout follows [PITH_FULL_IMAGE:figures/full_fig_p023_8.png]
Figure 9
Figure 9. Figure 9: More generated images. The layout follows [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 10
Figure 10. Figure 10: More generated images. The layout follows [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 32 canonical work pages

  1. [14]

    arXiv preprint arXiv:2412.06510 (2024)

    He, S., Liu, L., Shu, X., Wang, B., Feng, Y., Zhao, S.: Anomalycontrol: Learning cross-modal semantic features for controllable anomaly synthesis. arXiv preprint arXiv:2412.06510 (2024)

  2. [19]

    arXiv preprint arXiv:2511.10020 (2025)

    Jiang, Y., Luo, W., Zhang, H., Chen, Q., Yao, H., Shen, W., Cao, Y.: Anomagic: Crossmodal prompt-driven zero-shot anomaly generation. arXiv preprint arXiv:2511.10020 (2025)

  3. [1]

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

    Avrahami, O., Lischinski, D., Fried, O.: Blended diffusion for text-driven editing of natural images. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18208–18218 (2022)

  4. [2]

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

    Bergmann, P., Fauser, M., Sattlegger, D., Steger, C.: Mvtec ad–a comprehen- sive real-world dataset for unsupervised anomaly detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9592– 9600 (2019)

  5. [3]

    In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision

    Cao, M., Wang, X., Qi, Z., Shan, Y., Qie, X., Zheng, Y.: Masactrl: Tuning-free mu- tual self-attention control for consistent image synthesis and editing. In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision. pp. 22560– 22570 (2023)

  6. [4]

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

    Chung, J., Hyun, S., Heo, J.P.: Style injection in diffusion: A training-free approach for adapting large-scale diffusion models for style transfer. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 8795– 8805 (2024)

  7. [5]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025)

    Dai, Z., Zeng, S., Liu, H., Li, X., Xue, F., Zhou, Y.: Seas: Few-shot industrial anomaly image generation with separation and sharing fine-tuning. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025)

  8. [6]

    arXiv preprint arXiv:1708.04552 (2017)

    DeVries, T., Taylor, G.W.: Improved regularization of convolutional neural net- works with cutout. arXiv preprint arXiv:1708.04552 (2017)

Show all 41 references
  1. [7]

    Dong, J., Liang, W., Li, H., Zhang, D., Cao, M., Ding, H., Khan, S.H., Shah- baz Khan, F.: How to continually adapt text-to-image diffusion models for flexible customization? Advances in Neural Information Processing Systems37, 130057– 130083 (2024)

  2. [8]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Duan, Y., Hong, Y., Niu, L., Zhang, L.: Few-shot defect image generation via defect-aware feature manipulation. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 37, pp. 571–578 (2023)

  3. [9]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Duan, Y., Hong, Y., Niu, L., Zhang, L.: Few-shot defect image generation via defect-aware feature manipulation. In: Proceedings of the AAAI Conference on Artificial Intelligence. pp. 571–578 (2023)

  4. [10]

    In: Forty-first International Conference on Machine Learning (2024)

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Pines, D., Sauer, A., Tja, M., et al.: Scaling rectified flow transformers for high- resolution image synthesis. In: Forty-first International Conference on Machine Learning (2024)

  5. [11]

    In: European Conference on Computer Vision

    Frenkel, Y., Vinker, Y., Shamir, A., Cohen-Or, D.: Implicit style-content separation using b-lora. In: European Conference on Computer Vision. pp. 181–198. Springer (2024) 16 S. Zeng et al

  6. [12]

    Advances in Neural Information Processing Systems36, 15890–15902 (2023)

    Gu, Y., Wang, X., Wu, J.Z., Shi, Y., Chen, Y., Fan, Z., Xiao, W., Zhao, R., Chang, S., Wu, W., et al.: Mix-of-show: Decentralized low-rank adaptation for multi-concept customization of diffusion models. Advances in Neural Information Processing Systems36, 15890–15902 (2023)

  7. [13]

    In: European conference on computer vision

    Gui, G., Gao, B.B., Liu, J., Wang, C., Wu, Y.: Few-shot anomaly-driven generation for anomaly classification and segmentation. In: European conference on computer vision. pp. 210–226. Springer (2024)

  8. [15]

    In: NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications (2021)

    Ho, J., Salimans, T.: Classifier-free diffusion guidance. In: NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications (2021)

  9. [16]

    arXiv preprint arXiv:2106.09685 (2021)

    Hu, E.J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., Chen, W.: Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685 (2021)

  10. [17]

    Hu, T., Zhang, J., Yi, R., Du, Y., Chen, X., Liu, L., Wang, Y., Wang, C.: Anomaly- diffusion:Few-shotanomalyimagegenerationwithdiffusionmodel.In:Proceedings of the AAAI Conference on Artificial Intelligence. pp. 8526–8534 (2024)

  11. [18]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Jiang, J., Zhang, Y., Feng, K., Wu, X., Li, W., Pei, R., Li, F., Zuo, W.: Mcˆ 2: Multi-concept guidance for customized multi-concept generation. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 2802–2812 (2025)

  12. [20]

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

    Jin, Y., Peng, J., He, Q., Hu, T., Wu, J., Chen, H., Wang, H., Zhu, W., Chi, M., Liu, J., et al.: Dual-interrelated diffusion model for few-shot anomaly image generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 30420–30429 (2025)

  13. [21]

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

    Li, C.L., Sohn, K., Yoon, J., Pfister, T.: Cutpaste: Self-supervised learning for anomaly detection and localization. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9664–9674 (2021)

  14. [22]

    In: The Eleventh International Conference on Learning Rep- resentations (2023)

    Lipman, Y., Chen, R.T., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. In: The Eleventh International Conference on Learning Rep- resentations (2023)

  15. [23]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Liu, C., Shah, V., Cui, A., Lazebnik, S.: Unziplora: Separating content and style from a single image. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 16776–16785 (2025)

  16. [24]

    In: The Eleventh International Conference on Learning Representations (2023)

    Liu, X., Gong, C., et al.: Flow straight and fast: Learning to generate and transfer data with rectified flow. In: The Eleventh International Conference on Learning Representations (2023)

  17. [25]

    In: International Conference on Machine Learning

    Liu, Z., Feng, R., Zhu, K., Zhang, Y., Zheng, K., Liu, Y., Zhao, D., Zhou, J., Cao, Y.: Cones: Concept neurons in diffusion models for customized generation. In: International Conference on Machine Learning. pp. 21548–21566 (2023)

  18. [26]

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

    Ojha, U., Li, Y., Lu, J., Efros, A.A., Lee, Y.J., Shechtman, E., Zhang, R.: Few- shot image generation via cross-domain correspondence. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10743– 10752 (2021)

  19. [27]

    In: 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Ouyang, Z., Li, Z., Hou, Q.: K-lora: Unlocking training-free fusion of any subject and style loras. In: 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 13041–13050 (2025) DeCo 17

  20. [28]

    In: Medical Image Computing and Computer-Assisted Intervention

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention. pp. 234–241. Springer (2015)

  21. [29]

    Advances in Neural Information Processing Systems29(2016)

    Salimans, T., Goodfellow, I., Zaremba, W., Cheung, V., Radford, A., Chen, X.: Improved techniques for training gans. Advances in Neural Information Processing Systems29(2016)

  22. [30]

    In: European Conference on Com- puter Vision

    Schlüter, H.M., Tan, J., Hou, B., Kainz, B.: Natural synthetic anomalies for self- supervised anomaly detection and localization. In: European Conference on Com- puter Vision. pp. 474–489. Springer (2022)

  23. [31]

    In: European Conference on Computer Vision

    Shah, V., Ruiz, N., Cole, F., Lu, E., Lazebnik, S., Li, Y., Jampani, V.: Ziplora: Any subject in any style by effectively merging loras. In: European Conference on Computer Vision. pp. 422–438. Springer (2024)

  24. [32]

    In: Proceedings of the AAAI Conference on Artificial Intelligence (2026)

    So, Y., Kang, S.: Anostyler: Text-driven localized anomaly generation via lightweight style transfer. In: Proceedings of the AAAI Conference on Artificial Intelligence (2026)

  25. [33]

    In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference

    Sun, H., Cao, Y., Dong, H., Fink, O.: Unseen visual anomaly generation. In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference. pp. 25508– 25517 (2025)

  26. [34]

    In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition

    Tumanyan, N., Geyer, M., Barmatz, S., Dekel, T.: Plug-and-play diffusion features for text-driven image-to-image translation. In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition. pp. 1921–1930 (2023)

  27. [35]

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

    Wang, C., Zhu, W., Gao, B.B., Gan, Z., Zhang, J., Gu, Z., Qian, S., Chen, M., Ma, L.: Real-iad: A real-world multi-view dataset for benchmarking versatile industrial anomaly detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 2...

  28. [36]

    In: Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing

    Wu, T., Wang, J., Zhao, Z., Wong, N.: Mixture-of-subspaces in low-rank adap- tation. In: Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. pp. 7880–7899 (2024)

  29. [37]

    In: Proceedings of the IEEE/CVF International Conference on Com- puter Vision

    Yang, J., Ma, Y., Di, D., Cui, J., Li, H., Chen, W., Xie, Y., Yang, X., Zuo, W.: Qr-lora: Efficient and disentangled fine-tuning via qr decomposition for customized generation. In: Proceedings of the IEEE/CVF International Conference on Com- puter Vision. pp. 17587–17597 (2025)

  30. [38]

    IEEE Transactions on Image Pro- cessing34, 8145–8158 (2025)

    Yang, Y., Wang, W., Peng, L., Song, C., Chen, Y., Li, H., Yang, X., Lu, Q., Cai, D., He, X., et al.: Lora-composer: Leveraging low-rank adaptation for multi-concept customization in training-free diffusion models. IEEE Transactions on Image Pro- cessing34, 8145–8158 (2025)

  31. [39]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Zavrtanik, V., Kristan, M., Skočaj, D.: Draem – a discriminatively trained re- construction embedding for surface anomaly detection. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 8330–8339 (2021)

  32. [40]

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

    Zhang, X., Xu, M., Zhou, X.: Realnet: A feature selection network with realis- tic synthetic anomaly for anomaly detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 16699–16708 (2024)

  33. [41]

    a <product> with an <anomaly>

    Zou, Y., Jeong, J., Pemula, L., Zhang, D., Dabeer, O.: Spot-the-difference self- supervised pre-training for anomaly detection and segmentation. In: European Conference on Computer Vision. pp. 392–408 (2022) 18 S. Zeng et al. Appendix A Additional Implementation Details Archit...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.