arxiv: 2604.09023 · v1 · submitted 2026-04-10 · 💻 cs.CV

Recognition: no theorem link

CAD 100K: A Comprehensive Multi-Task Dataset for Car Related Visual Anomaly Detection

Jiahua Pang , Ying Li , Dongpu Cao , Jingcai Luo , Yanuo Zheng , Bao Yunfan , Yujie Lei , Rui Yuan

show 5 more authors

Yuxi Tian Guojin Yuan Hongchang Chen Zhi Zheng Yongchun Liu

Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:54 UTC · model grok-4.3

classification 💻 cs.CV

keywords visual anomaly detectionmulti-task learningcar anomaly datasetbenchmark datasetmanufacturing quality assessmentfew-shot learningvehicle inspection

0 comments

The pith

The CAD dataset is the first large-scale benchmark for multi-task visual anomaly detection across car domains.

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

The paper presents the CAD dataset to fill the gap in unified benchmarks for multi-task visual anomaly detection in cars. Current methods stay limited to single tasks because no comprehensive multi-task resource existed. The dataset supplies over 100K images spanning seven vehicle domains and three tasks, with synthesis augmentation to handle few-shot anomaly cases. A multi-task baseline shows that joint training promotes interaction and knowledge transfer between tasks, while also highlighting conflicts that arise. This standardized platform is intended to drive progress in applying anomaly detection to car manufacturing quality assessment.

Core claim

The CAD dataset is the first car-related anomaly dataset specialized for multi-task learning. It contains over 100K images crossing seven vehicle domains and three tasks, combining synthesis data augmentation for few-shot anomaly images. A multi-task baseline and empirical studies show that multi-task learning promotes task interaction and knowledge transfer while exposing challenging conflicts between tasks.

What carries the argument

The CAD dataset, a multi-task benchmark for car visual anomaly detection with cross-domain coverage and synthesis augmentation for rare cases.

If this is right

Multi-task learning can share features across car anomaly tasks to improve overall detection.
Conflicts between tasks require new methods to balance competing objectives in joint training.
The dataset provides a common testbed for comparing future multi-task anomaly models.
Synthesis augmentation enables handling of rare anomaly types that appear in few-shot settings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Manufacturers could integrate the benchmark to train inspection systems that catch defects across different car parts in one model.
The multi-task setup may generalize to anomaly detection in other vehicle or machinery domains where multiple inspection criteria apply simultaneously.
If task conflicts prove severe, hybrid approaches combining MTL with task-specific heads could become necessary for practical deployment.

Load-bearing premise

The multi-task baseline and empirical studies sufficiently demonstrate that task interaction and knowledge transfer occur without detailed quantitative results or controls.

What would settle it

A controlled experiment on the CAD dataset where single-task models consistently outperform the multi-task baseline on all three tasks would show that the claimed benefits of interaction do not hold.

Figures

Figures reproduced from arXiv: 2604.09023 by Bao Yunfan, Dongpu Cao, Guojin Yuan, Hongchang Chen, Jiahua Pang, Jingcai Luo, Rui Yuan, Yanuo Zheng, Ying Li, Yongchun Liu, Yujie Lei, Yuxi Tian, Zhi Zheng.

**Figure 1.** Figure 1: Overview of the CAD 100K Dataset. The CAD 100K dataset contains over 100K images crossing 7 domains and 3 tasks, providing a comprehensive view for car-related anomaly detection. It features a multi-task architecture and synthesis data augmentation methods, solves cross-domain problems, supports both supervised and unsupervised learning paradigms, and is compatible with few-shot learning frameworks. conver… view at source ↗

**Figure 2.** Figure 2: Tasks and structure of the CAD 100K Dataset. CAD 100K adopts a hierarchical domain–system–part anomaly structure to ensure semantic consistency and extensibility across different tasks. new anomalous. The General domain encompasses crosssystem anomalies (e.g. bolt loosening) and ensures future extensibility. Systems and Parts (23 systems, 78 parts). Each domain is decomposed into functional subsystems (e.… view at source ↗

**Figure 3.** Figure 3: The data composition across domains, tasks, and sources in CAD 100K. All images are in RGB format with various resolutions from 100 × 84 to 4096 × 3072, covering diverse lighting and viewpoints. CAD 100K Cls Task Det Task Seg Task Task Samapler Fixed Task Weight Adaptive Task Weight Shared Backbone Cls Det Seg Shared MTL Encoder [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Figure 4: Structure of the proposed Baseline. both CNN- and ViT-based encoders. Specifically, we implement a shared backbone across all domains, ensuring unified feature extraction for classification, detection and segmentation tasks. The choice between ConvNeXt [18] and DINOv3 [25] is determined by deployment constraints and model scaling rather than domain separation—both serve as a common visual encoder for th… view at source ↗

read the original abstract

Multi-task visual anomaly detection is critical for car-related manufacturing quality assessment. However, existing methods remain task-specific, hindered by the absence of a unified benchmark for multi-task evaluation. To fill in this gap, We present the CAD Dataset, a large-scale and comprehensive benchmark designed for car-related multi-task visual anomaly detection. The dataset contains over 100 images crossing 7 vehicle domains and 3 tasks, providing models a comprehensive view for car-related anomaly detection. It is the first car-related anomaly dataset specialized for multi-task learning(MTL), while combining synthesis data augmentation for few-shot anomaly images. We implement a multi-task baseline and conduct extensive empirical studies. Results show MTL promotes task interaction and knowledge transfer, while also exposing challenging conflicts between tasks. The CAD dataset serves as a standardized platform to drive future advances in car-related multi-task visual anomaly detection.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper introduces a new multi-task dataset for car anomaly detection that fills a clear gap, but the title CAD 100K clashes directly with the abstract's claim of only over 100 images.

read the letter

The CAD 100K paper introduces a new dataset for multi-task visual anomaly detection focused on cars, filling a gap for unified benchmarks in this area. But the title and large-scale claims don't line up with the abstract's count of over 100 images, which is the key thing to watch. They cover 7 vehicle domains and 3 tasks, using synthesis augmentation to deal with rare anomaly images. The multi-task baseline they built shows that joint training promotes task interaction and knowledge transfer, but it also brings out conflicts between tasks. This setup gives a standardized platform that prior task-specific approaches lacked. The paper does a good job highlighting why a shared dataset matters for car-related manufacturing quality assessment. Adding synthetic data for few-shot cases is a sensible practical step. The main soft spot is that scale inconsistency. If the full manuscript has around 100,000 images with proper statistics, splits, and annotation details, then the benchmark claim holds up better. As described, though, it looks like the numbers need fixing to support calling it comprehensive. The empirical studies are described without specific metrics or error bars, so the strength of the MTL promotion is not fully clear from the high-level summary. The work shows clear thinking in identifying the literature gap and proposing a solution with MTL. No load-bearing flaws beyond the description mismatch. This is for researchers in computer vision who deal with anomaly detection in industrial or automotive settings. A reader interested in multi-task learning or creating benchmarks would find value in the dataset structure and the baseline results. It deserves a serious referee to sort out the scale details and add quantitative evidence, rather than being desk rejected. Recommendation: Engage with it in review, as the idea is sound even if the presentation has issues.

Referee Report

2 major / 3 minor

Summary. The manuscript introduces the CAD 100K dataset as a large-scale benchmark for car-related multi-task visual anomaly detection. It spans 7 vehicle domains and 3 tasks, incorporates synthetic data augmentation for few-shot anomalies, implements an MTL baseline, and reports empirical results indicating that MTL promotes task interaction and knowledge transfer while exposing inter-task conflicts. The work positions the dataset as the first MTL-specialized car anomaly benchmark to enable standardized evaluation beyond task-specific approaches.

Significance. A well-documented dataset at the claimed scale with reproducible splits, annotation protocols, and quantitative MTL baselines demonstrating measurable knowledge transfer would constitute a useful contribution to industrial anomaly detection in computer vision. It could standardize multi-task evaluation in automotive quality assessment and highlight practical trade-offs in joint training.

major comments (2)

Abstract: The text states the dataset 'contains over 100 images' while the title is 'CAD 100K' and the introduction describes it as 'large-scale and comprehensive'. This numerical inconsistency is load-bearing for the central claim that CAD fills the gap as a large-scale MTL benchmark. The dataset section must state the exact total image count, per-domain and per-task breakdowns, train/validation/test splits, and annotation process to substantiate the scale.
Experiments section: The abstract claims 'extensive empirical studies' show MTL benefits and task conflicts, yet no specific metrics, baselines, ablation controls, or statistical significance are referenced. If the full manuscript lacks tables reporting per-task performance deltas, error bars, or single-task vs. MTL comparisons, the knowledge-transfer claim cannot be evaluated.

minor comments (3)

Abstract: The sentence 'while combining synthesis data augmentation for few-shot anomaly images' is grammatically unclear; rephrase for precision.
Abstract: The three tasks should be named explicitly (e.g., defect classification, localization, severity estimation) at first mention rather than left implicit.
Ensure all figures and tables include clear captions, axis labels, and legends so that MTL vs. single-task comparisons are immediately interpretable without reference to the text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the two major comments point by point below and will revise the manuscript to resolve the identified issues.

read point-by-point responses

Referee: Abstract: The text states the dataset 'contains over 100 images' while the title is 'CAD 100K' and the introduction describes it as 'large-scale and comprehensive'. This numerical inconsistency is load-bearing for the central claim that CAD fills the gap as a large-scale MTL benchmark. The dataset section must state the exact total image count, per-domain and per-task breakdowns, train/validation/test splits, and annotation process to substantiate the scale.

Authors: We acknowledge the inconsistency and apologize for the typographical error in the abstract. The dataset is titled CAD 100K because it contains 100,000 images across 7 vehicle domains and 3 tasks. The phrase 'over 100 images' should have read 'over 100K images'. In the revised version we will correct the abstract and expand the dataset section with the precise total count, per-domain and per-task breakdowns, train/validation/test splits, and a complete description of the annotation protocol to fully substantiate the claimed scale. revision: yes
Referee: Experiments section: The abstract claims 'extensive empirical studies' show MTL benefits and task conflicts, yet no specific metrics, baselines, ablation controls, or statistical significance are referenced. If the full manuscript lacks tables reporting per-task performance deltas, error bars, or single-task vs. MTL comparisons, the knowledge-transfer claim cannot be evaluated.

Authors: The experiments section already reports results from the implemented multi-task baseline and includes comparisons illustrating MTL benefits and inter-task conflicts. To strengthen the presentation and directly address the concern, we will add explicit tables showing per-task performance deltas, single-task versus MTL comparisons, error bars, and any available statistical significance tests in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity; dataset paper with no derivation chain or self-referential reductions

full rationale

The paper introduces a new dataset (CAD) and implements a multi-task baseline for car-related visual anomaly detection. No equations, predictions, fitted parameters, or first-principles derivations appear in the provided text. Claims such as being the 'first' specialized MTL dataset are descriptive assertions supported by literature positioning rather than any self-definitional loop, fitted-input-as-prediction, or self-citation load-bearing step. The title/abstract scale mismatch (CAD 100K vs. 'over 100 images') is a factual inconsistency, not a circular reduction of any claimed result to its own inputs. The contribution remains self-contained as an empirical benchmark release.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the novelty of the dataset and the observed benefits of multi-task learning in the empirical studies. Without the full text, the ledger is limited to assumptions stated at the abstract level.

axioms (1)

domain assumption Multi-task learning promotes task interaction and knowledge transfer while exposing conflicts between tasks in car anomaly detection.
Invoked directly in the description of the baseline results in the abstract.

pith-pipeline@v0.9.0 · 5490 in / 1222 out tokens · 38759 ms · 2026-05-10T17:54:58.540912+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

35 extracted references · 4 canonical work pages · 1 internal anchor

[1]

Supervised anomaly detection for complex indus- trial images

Aimira Baitieva, David Hurych, Victor Besnier, and Olivier Bernard. Supervised anomaly detection for complex indus- trial images. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17754– 17762, 2024. 1

2024
[2]

Mvtec ad–a comprehensive real-world dataset for unsupervised anomaly detection

Paul Bergmann, Michael Fauser, David Sattlegger, and Carsten Steger. Mvtec ad–a comprehensive real-world dataset for unsupervised anomaly detection. InCVPR, pages 9592–9600, 2019. 1

2019
[3]

Vebd-hel: A noval approach to vehicle exterior body damage parts classification in intelli- gent transportation systems.Alexandria Engineering Jour- nal, 108:961–975, 2024

Michael Abebe Berwo, Yong Fang, Asad Khan, Adnan Man- zor, and Jabar Mahmood. Vebd-hel: A noval approach to vehicle exterior body damage parts classification in intelli- gent transportation systems.Alexandria Engineering Jour- nal, 108:961–975, 2024. 4

2024
[4]

Itran: A novel transformer-based approach for industrial anomaly detection and localization.Engineering Applications of Artificial Intelligence, 125:106677, 2023

Xiangyu Cai, Ruliang Xiao, Zhixia Zeng, Ping Gong, and Youcong Ni. Itran: A novel transformer-based approach for industrial anomaly detection and localization.Engineering Applications of Artificial Intelligence, 125:106677, 2023. 3

2023
[5]

A sur- vey on anomaly detection with few-shot learning

Junyang Chen, Changbo Wang, Yifan Hong, Rui Mi, Liang- Jie Zhang, Yirui Wu, Huan Wang, and Yue Zhou. A sur- vey on anomaly detection with few-shot learning. InInter- national Conference on Cognitive Computing, pages 34–50. Springer, 2024. 3

2024
[6]

Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks

Zhao Chen, Vijay Badrinarayanan, Chen-Yu Lee, and An- drew Rabinovich. Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks. InIn- ternational conference on machine learning, pages 794–803. PMLR, 2018. 3

2018
[7]

Seas: few-shot industrial anomaly image gen- eration with separation and sharing fine-tuning

Zhewei Dai, Shilei Zeng, Haotian Liu, Xurui Li, Feng Xue, and Yu Zhou. Seas: few-shot industrial anomaly image gen- eration with separation and sharing fine-tuning. InProceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 23135–23144, 2025. 5

2025
[8]

Anomaly detection via re- verse distillation from one-class embedding

Hanqiu Deng and Xingyu Li. Anomaly detection via re- verse distillation from one-class embedding. InCVPR, pages 9737–9746, 2022. 1

2022
[9]

Fastrecon: Few-shot indus- trial anomaly detection via fast feature reconstruction

Zheng Fang, Xiaoyang Wang, Haocheng Li, Jiejie Liu, Qi- ugui Hu, and Jimin Xiao. Fastrecon: Few-shot indus- trial anomaly detection via fast feature reconstruction. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 17481–17490, 2023. 3

2023
[10]

Coba: convergence balancer for mul- titask finetuning of large language models.arXiv preprint arXiv:2410.06741, 2024

Zi Gong, Hang Yu, Cong Liao, Bingchang Liu, Chaoyu Chen, and Jianguo Li. Coba: convergence balancer for mul- titask finetuning of large language models.arXiv preprint arXiv:2410.06741, 2024. 6

work page arXiv 2024
[11]

One model is all you need: multi-task learning enables simultaneous histology image segmentation and classifica- tion.Medical Image Analysis, 83:102685, 2023

Simon Graham, Quoc Dang Vu, Mostafa Jahanifar, Shan E Ahmed Raza, Fayyaz Minhas, David Snead, and Nasir Ra- jpoot. One model is all you need: multi-task learning enables simultaneous histology image segmentation and classifica- tion.Medical Image Analysis, 83:102685, 2023. 1

2023
[12]

Vehicle damage detection us- ing artificial intelligence: A systematic literature review.Wi- ley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 15(2):e70027, 2025

Md Jahid Hasan, Cong Kha Nguyen, Yee Ling Boo, Hamed Jahani, and Kok-Leong Ong. Vehicle damage detection us- ing artificial intelligence: A systematic literature review.Wi- ley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 15(2):e70027, 2025. 1

2025
[13]

Real-world data is dirty: Data cleansing and the merge/purge problem

Mauricio A Hern ´andez and Salvatore J Stolfo. Real-world data is dirty: Data cleansing and the merge/purge problem. Data mining and knowledge discovery, 2(1):9–37, 1998. 3

1998
[14]

Ning Huyan, Xiangrong Zhang, Dou Quan, Jocelyn Chanus- sot, and Licheng Jiao. Aud-net: A unified deep detector for multiple hyperspectral image anomaly detection via relation and few-shot learning.IEEE Transactions on Neural Net- works and Learning Systems, 35(5):6835–6849, 2022. 3

2022
[15]

Vehide dataset: New dataset for automatic vehicle damage detection in car insurance

Nhan T Huynh, Nguyen ND Tran, Anh T Huynh, Van-Dung Hoang, and Hien D Nguyen. Vehide dataset: New dataset for automatic vehicle damage detection in car insurance. In2023 15th International Conference on Knowledge and Systems Engineering (KSE), pages 1–6. IEEE, 2023. 1

2023
[16]

Co- training transformer for remote sensing image classification, segmentation, and detection.IEEE Transactions on Geo- science and Remote Sensing, 62:1–18, 2024

Qingyun Li, Yushi Chen, Xin He, and Lingbo Huang. Co- training transformer for remote sensing image classification, segmentation, and detection.IEEE Transactions on Geo- science and Remote Sensing, 62:1–18, 2024. 3, 5

2024
[17]

Chuen-Horng Lin, Chia-Ching Yu, and Huan-Yu Chen. Augmentation dataset of a two-dimensional neural network model for use in the car parts segmentation and car classifi- cation of three dimensions.The Journal of Supercomputing, 78(17):18915–18958, 2022. 5, 7

2022
[18]

A convnet for the 2020s

Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feicht- enhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. InProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 11976–11986,
[19]

Is sampled data sufficient for anomaly detec- tion? InProceedings of the 6th ACM SIGCOMM conference on Internet measurement, pages 165–176, 2006

Jianning Mai, Chen-Nee Chuah, Ashwin Sridharan, Tao Ye, and Hui Zang. Is sampled data sufficient for anomaly detec- tion? InProceedings of the 6th ACM SIGCOMM conference on Internet measurement, pages 165–176, 2006. 3

2006
[20]

Ve- hicle damage severity estimation for insurance operations using in-the-wild mobile images.IEEE Access, 11:78644– 78655, 2023

Dimitrios Mallios, Li Xiaofei, Niall McLaughlin, Jesus Mar- tinez Del Rincon, Clare Galbraith, and Rory Garland. Ve- hicle damage severity estimation for insurance operations using in-the-wild mobile images.IEEE Access, 11:78644– 78655, 2023. 1, 4

2023
[21]

Machine learning techniques for anomaly detection: an overview.In- ternational Journal of Computer Applications, 79(2), 2013

Salima Omar, Asri Ngadi, and Hamid H Jebur. Machine learning techniques for anomaly detection: an overview.In- ternational Journal of Computer Applications, 79(2), 2013. 3

2013
[22]

Learning and transferring mid-level image representations using convolutional neural networks

Maxime Oquab, Leon Bottou, Ivan Laptev, and Josef Sivic. Learning and transferring mid-level image representations using convolutional neural networks. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 1717–1724, 2014. 3

2014
[23]

Car damage detection based on multi-view fusion and align- ment: Dataset and method.IEEE Transactions on Intelligent Transportation Systems, 2025

Jinbo Peng, Shoubin Dong, Hua Yuan, and Xiaorou Zheng. Car damage detection based on multi-view fusion and align- ment: Dataset and method.IEEE Transactions on Intelligent Transportation Systems, 2025. 1, 2

2025
[24]

Anomaly detection in time series: a comprehensive evalu- ation.Proceedings of the VLDB Endowment, 15(9):1779– 1797, 2022

Sebastian Schmidl, Phillip Wenig, and Thorsten Papenbrock. Anomaly detection in time series: a comprehensive evalu- ation.Proceedings of the VLDB Endowment, 15(9):1779– 1797, 2022. 1

2022
[25]

DINOv3

Oriane Sim ´eoni, Huy V V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Micha ¨el Ramamonjisoa, et al. Dinov3.arXiv preprint arXiv:2508.10104, 2025. 6, 7, 8

work page internal anchor Pith review Pith/arXiv arXiv 2025
[26]

Object- stitch: Object compositing with diffusion model

Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, and Daniel Aliaga. Object- stitch: Object compositing with diffusion model. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18310–18319, 2023. 5

2023
[27]

Real-time car part instance segmentation: the comparison of the state-of-the-art

Wittawin Susutti, Siwarat Laoprom, Thanaphit Sutthipanyo, Kanok V ongsaroj, Pirun Dilokpatpongsa, and Warin Wat- tanapornprom. Real-time car part instance segmentation: the comparison of the state-of-the-art. In2024 28th Interna- tional Computer Science and Engineering Conference (IC- SEC), pages 1–6. IEEE, 2024. 2, 5, 7

2024
[28]

Real-iad: A real-world multi-view dataset for benchmarking versatile industrial anomaly detec- tion

Chengjie Wang, Wenbing Zhu, Bin-Bin Gao, Zhenye Gan, Jiangning Zhang, Zhihao Gu, Shuguang Qian, Mingang Chen, and Lizhuang Ma. Real-iad: A real-world multi-view dataset for benchmarking versatile industrial anomaly detec- tion. InCVPR, pages 22883–22892, 2024. 1

2024
[29]

Cardd: A new dataset for vision-based car damage detection.IEEE Transactions on Intelligent Transportation Systems, 24(7): 7202–7214, 2023

Xinkuang Wang, Wenjing Li, and Zhongcheng Wu. Cardd: A new dataset for vision-based car damage detection.IEEE Transactions on Intelligent Transportation Systems, 24(7): 7202–7214, 2023. 1, 2, 5, 7

2023
[30]

Few-shot online anomaly detection and segmentation.Knowledge-Based Sys- tems, 300:112168, 2024

Shenxing Wei, Xing Wei, Zhiheng Ma, Songlin Dong, Shaochen Zhang, and Yihong Gong. Few-shot online anomaly detection and segmentation.Knowledge-Based Sys- tems, 300:112168, 2024. 3

2024
[31]

A review on evolutionary multitask optimization: Trends and challenges.IEEE Transactions on Evolutionary Computation, 26(5):941–960, 2021

Tingyang Wei, Shibin Wang, Jinghui Zhong, Dong Liu, and Jun Zhang. A review on evolutionary multitask optimization: Trends and challenges.IEEE Transactions on Evolutionary Computation, 26(5):941–960, 2021. 3

2021
[32]

A comprehensive survey of deep transfer learn- ing for anomaly detection in industrial time series: Methods, applications, and directions.IEEE Access, 12:3768–3789,

Peng Yan, Ahmed Abdulkadir, Paul-Philipp Luley, Matthias Rosenthal, Gerrit A Schatte, Benjamin F Grewe, and Thilo Stadelmann. A comprehensive survey of deep transfer learn- ing for anomaly detection in industrial time series: Methods, applications, and directions.IEEE Access, 12:3768–3789,
[33]

Cabs: Conflict-aware and balanced sparsification for enhancing model merging.arXiv preprint arXiv:2503.01874, 2025

Zongzhen Yang, Binhang Qi, Hailong Sun, Wenrui Long, Ruobing Zhao, and Xiang Gao. Cabs: Conflict-aware and balanced sparsification for enhancing model merging.arXiv preprint arXiv:2503.01874, 2025. 3

work page arXiv 2025
[34]

Achievement-based training progress balancing for multi-task learning

Hayoung Yun and Hanjoo Cho. Achievement-based training progress balancing for multi-task learning. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 16935–16944, 2023. 3

2023
[35]

Pku-goodsad: A supermarket goods dataset for unsupervised anomaly detection and segmentation.IEEE Robotics and Au- tomation Letters, 2024

Jian Zhang, Runwei Ding, Miaoju Ban, and Linhui Dai. Pku-goodsad: A supermarket goods dataset for unsupervised anomaly detection and segmentation.IEEE Robotics and Au- tomation Letters, 2024. 1 A. Ethical Considerations and Data Privacy In accordance with the ethics guidelines, we have imple- mented comprehensive measures to protect privacy and mit- igate...

work page arXiv 2024