Pith. sign in

REVIEW 1 major objections 2 minor 74 references

DoorDet: Semi-Automated Multi-Class Door Detection Dataset via Object Detection and Large Language Models

T0 review · 1 major / 2 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper's body claims TriReWeight, a triplet-connection-based re-weighting method, can be attached to any generative data augmentation method, never downgrade its performance, and improve average accuracy by 7.9 percentage points on natu

desk verdict The abstract promises a door-detection dataset, but the actual text is a different paper on re-weighting generated images—so the announced contribution is unverdictable, and the body is a plausible but under-evidenced separate preprint. read the letter →

arxiv 2508.07714 v1 pith:SMIUN6O5 submitted 2025-08-11 cs.CV cs.AIcs.ET

Licheng Zhang , Bach Le , Naveed Akhtar , Tuan Ngo This is my paper
classification cs.CVcs.AIcs.ET
keywords generativedataaugmentationsamplere-weightingtripletlossnoisygeneratedimagessmall-scaledatasetsimageclassificationbi-leveloptimizationconsistencyregularization
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

This submission is internally mismatched: its abstract describes a DoorDet dataset pipeline for door detection in floor plans, but the full text is a different paper, TriReWeight, about re-weighting generated images for small-scale dataset expansion. Judged by its body, the paper is trying to establish that a triplet-connection-based sample re-weighting method can be attached to any generative data augmentation method, never downgrade the augmentation, and typically improve it — on average +7.9% on six natural image datasets and +3.4% on three medical datasets with the GIF baseline. A sympathetic reader would care because generative augmentation is a practical fix for small-data domains like medical imaging, and TriReWeight promises a drop-in wrapper that makes that fix reliably safer. The mismatch itself is flagged here as a submission-level issue, not part of the scientific claim.

What carries the argument

The central object is the triplet-connection supervision scheme: each generated image is supervised by a triplet loss with its source image as anchor and a randomly sampled image from another class as negative, plus a consistency-regularization loss treating the generated image as unlabeled. The weights are learned online via a bi-level optimization alternating between updating the classifier and updating the weight function, following the approximation of Ren et al. [44]. The theorem that never downgrades holds because setting all weights to zero recovers training on originals, and setting all weights to one recovers the unweighted augmentation baseline.

What would settle it

Measure, on a dataset where TriReWeight is claimed to help (e.g., OrganSMNIST or PathMNIST), the fraction of generated images that actually satisfy the pairwise-connection condition — that is, the fraction for which the distance from the generated image to its own source is smaller than to a random other-class image. If that fraction is far below 92% for any dataset/augmentation combination, the triplet objective encodes wrong structure and the reported gains should not transfer; a reader could also rerun Table 2 with the pairwise loss replaced by a symmetric contrastive loss to test whether t

Watch

Extended reading notes

Core claim

The full text (TriReWeight) claims that the three types of supervision available for re-weighting generated images — strong connection, pairwise connection, and no connection — have different effects: strong connection supervision with incorrect labels increases the generalization risk upper bound by $\gamma A$ (Theorem 4.1), while pairwise and no-connection supervisions are provably safe. Based on this, TriReWeight optimizes a bi-level objective where the inner level combines classification loss with a weighted sum of triplet loss (pairwise connection) and consistency regularization (no connection), and the outer level learns per-sample weights to minimize clean-data risk. The paper proves

Load-bearing premise

The assumption that each generated image is closer to its own source image than to images of any other class (pairwise connection with low noise) is the load-bearing premise; the paper validates it only on the Pets dataset ('over 92%') and then assumes it across all nine datasets and all augmentation methods.

Editorial extensions

If this is right

  • Generative data augmentation for small-scale datasets can be made safer with a drop-in wrapper that requires no change to the augmentation method or inference architecture.
  • With TriReWeight, 5x expansion can outperform 20x expansion without it on Pets, Caltech 101, and DTD, implying at least a 4x data-efficiency gain.
  • The method's guarantees extend to other diffusion backbones and model architectures (e.g., MobileNet-v2), not just GIF + ResNet-50.
  • The theoretical results suggest that down-weighting noisy generated images rather than trusting their labels is the right strategy, which may guide future augmentation designs.

Reading between the lines

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

  • If the pairwise-connection assumption is validated broadly, the triplet-loss re-weighting scheme could serve as a generic quality filter for any generative dataset, beyond the nine tested sets.
  • The 'never downgrade' guarantee is stated for training empirical risk; a stronger claim about test-time safety would require additional assumptions on the distribution shift between generated and real data.
  • The consistency-regularization term treats generated images as unlabeled, which suggests a natural extension to semi-supervised learning settings where only a fraction of the expanded dataset is trusted.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

1 major / 2 minor

Summary. The submission, as titled and abstracted, proposes DoorDet, a semi-automated pipeline for constructing a multi-class door detection dataset from floor plans, combining an object detector, an LLM, and human review. However, the full text is an entirely different manuscript: "Enhancing Small-Scale Dataset Expansion with Triplet-Connection-based Sample Re-Weighting" by Ting Xiang et al., identified as arXiv:2508.07723v1 and formatted as an ACM MM 2025 paper. The body contains no material on door detection, floor plans, or the proposed dataset construction pipeline. The announced contribution is therefore completely absent from the submitted text, making the paper unverdictable as submitted.

Significance. If the DoorDet dataset and pipeline were actually presented, the contribution could be useful for floor plan analysis and for demonstrating LLM-assisted dataset labeling. However, none of that content is present. The accidental inclusion of the TriReWeight paper does not provide evidence for the DoorDet claims. Because the manuscript's stated central claim has no supporting text, its significance cannot be assessed. The submission cannot be accepted in this form.

major comments (1)
  1. [Title/Abstract vs. Full Text] The central claim of the paper is the construction of the DoorDet door-detection dataset. The full text is a different paper, as shown by the page header 'arXiv:2508.07723v1 [cs.CV]' and the title/authors on the first page. A search of the body yields no occurrences of 'door', 'floor plan', 'LLM', or 'human-in-the-loop' beyond the abstract. Thus the announced contribution—the dataset, the semi-automated pipeline, and any evaluation of it—is absent. This is a load-bearing internal inconsistency that cannot be fixed by correction of individual sections; the manuscript would need to be replaced. I recommend rejection.
minor comments (2)
  1. [Page header] The running page header displays 'arXiv:2508.07723v1 11 Aug 2025', while the submission is arXiv:2508.07714; the identifiers should match if this is the intended paper.
  2. [Formatting] The full text is formatted as a camera-ready ACM MM paper with its own DOI and reference format, which conflicts with the arXiv submission style implied by the abstract; the authors should ensure the submitted file corresponds to the announced title and content.

Circularity Check

0 steps flagged · score 1.0 of 10

The body's TriReWeight derivation is not circular; the announced DoorDet claim is absent, and the 'never downgrade' proof is invalid rather than circular.

full rationale

The submitted manuscript is internally mismatched: the abstract announces a DoorDet dataset, but the full text is a different paper (TriReWeight, arXiv:2508.07723v1) by different authors. This is a submission-integrity problem, not a circularity: there is no derivation chain for DoorDet to analyze. For the TriReWeight body that is actually present, I find no significant circularity. The theoretical analysis of supervision types (Sec. 4.1) is imported from external results (Lei et al., Guo et al.) and the paper's own Theorem 4.1 is a standard bound; these are independent sources, not self-citations. The central empirical claim (7.9% and 3.4% accuracy gains) is a measured outcome, not a fitted prediction. The 'never downgrade' guarantee (Theorem 4.3) is not actually obtained by construction: setting w=1 in Eq. (6) yields an objective with triplet and consistency losses, which does not equal the θ_merged objective of Eq. (9) that uses cross-entropy on generated images; the proof in A.3 is therefore incorrect, a correctness risk, not a circular reduction. The self-citations ([6], [7], [63]) are minor and not load-bearing. Consequently the circularity score is low, despite the severe mismatch and proof errors.

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

The ledger is dominated by imported theory. Theorems 4.2, 4.4, and 4.5 come from Guo et al. [21]; the pairwise-learning guarantee from Lei et al. [33]; the meta-learning optimizer from Ren et al. [44]. The only in-house theorem, 4.1, is a standard label-noise bound resting on an ad hoc uniform-random-labeling assumption. The method introduces no new entities (the (c+1)-th 'noisy class' in Theorem 4.1 is a proof device, not an entity with falsifiable handles), but the headline results depend on at least four unreported or hand-picked quantities (weight-function parameterization, beta, delta_margin, expansion ratios), and the central premise (pairwise-connection low noise) is validated on one dataset only.

free parameters (5)
  • Balance weight beta (Eq. 6: triplet vs. consistency loss) = not reported
    Hand-chosen term weight controlling the two auxiliary losses; no value or tuning procedure is given.
  • Triplet margin delta (Eq. 7) = not reported
    Triplet loss margin; standard practice sets it by hand, but no value is given.
  • Weight function parameters alpha (w(x, x_i; alpha), Eq. 6) = functional form unspecified
    The learned per-image weights are the core of the method, yet the explicit parameterization is never provided in the text.
  • Expansion ratio m = 5x-20x by dataset (following GIF/DiffuseMix)
    Controls how many generated images per original; set by convention rather than fitted, but affects every reported number.
  • Noisy-image fraction gamma (Theorem 4.1) = abstract quantity
    Introduced ad hoc to derive the label-noise bound; the bound's tightness scales with it and no estimate is given.
assumptions (5)
  • ad hoc to paper Noisy generated images receive labels uniformly at random among the c classes, so p(ỹ=i|y=c+1)=1/c (Theorem 4.1 proof)
    Assumed to derive the label-noise bound; no evidence that generator confusion is uniform across classes.
  • domain assumption Pairwise-connection supervision holds with low noise (over 92% of generated images closer to source than to other classes), Section 4.1
    Validated on Pets only, without a documented protocol; assumed for all datasets and generators.
  • domain assumption Theorems from Guo et al. [21] transfer unchanged to the generative augmentation setting (Theorems 4.2, 4.4, 4.5)
    The safe-semi-supervised-learning guarantees and O(sqrt(d ln n/n)) rate are restated with proofs replaced by citations; no verification of Guo et al.'s assumptions.
  • standard math Pairwise-learning generalization guarantee under low noise holds (Lei et al. [33])
    External theorem imported in Section 4.1; its low-noise assumption is assumed satisfied.
  • standard math Losses are Lipschitz-smooth with L <= 2 and have rho-bounded gradients (Theorem 4.5)
    Assumed to prove monotone convergence; never verified for cross-entropy plus triplet plus consistency on ResNet-50.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DoorDet: Semi-Automated Multi-Class Door Detection Dataset via Object Detection and Large Language Models." pith.science (2026). https://pith.science/paper/SMIUN6O5

@misc{pith2026250807714,
  author       = {Pith},
  title        = {Pith review of: DoorDet: Semi-Automated Multi-Class Door Detection Dataset via Object Detection and Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SMIUN6O5}},
  note         = {Machine review of arXiv:2508.07714}
}
read the original abstract

Accurate detection and classification of diverse door types in floor plans drawings is critical for multiple applications, such as building compliance checking, and indoor scene understanding. Despite their importance, publicly available datasets specifically designed for fine-grained multi-class door detection remain scarce. In this work, we present a semi-automated pipeline that leverages a state-of-the-art object detector and a large language model (LLM) to construct a multi-class door detection dataset with minimal manual effort. Doors are first detected as a unified category using a deep object detection model. Next, an LLM classifies each detected instance based on its visual and contextual features. Finally, a human-in-the-loop stage ensures high-quality labels and bounding boxes. Our method significantly reduces annotation cost while producing a dataset suitable for benchmarking neural models in floor plan analysis. This work demonstrates the potential of combining deep learning and multimodal reasoning for efficient dataset construction in complex real-world domains.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

74 extracted references · 55 canonical work pages

  1. [1]

    Walid Al-Dhabyani, Mohammed Gomaa, Hussien Khaled, and Aly Fahmy. 2020. Dataset of breast ultrasound images. Data in Brief 28 (2020), 104863

    work page 2020

  2. [2]

    Antreas Antoniou, Amos Storkey, and Harrison Edwards. 2017. Data augmenta- tion generative adversarial networks. arXiv preprint arXiv:1711.04340 (2017)

    arXiv 2017

  3. [3]

    Shekoofeh Azizi, Simon Kornblith, Chitwan Saharia, Mohammad Norouzi, and David J. Fleet. 2023. Synthetic Data from Diffusion Models Improves ImageNet Classification. Trans. Mach. Learn. Res. 2023 (2023)

    work page 2023

  4. [4]

    Jonathan F Bard. 2013. Practical bilevel optimization: algorithms and applications . Vol. 30. Springer Science & Business Media

    work page 2013

  5. [5]

    David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, and Colin A Raffel. 2019. Mixmatch: A holistic approach to semi-supervised learning. Advances in neural information processing systems 32 (2019)

    work page 2019

  6. [6]

    Changjian Chen, Jiashu Chen, Weikai Yang, Haoze Wang, Johannes Knittel, Xibin Zhao, Steffen Koch, Thomas Ertl, and Shixia Liu. 2024. Enhancing single- frame supervision for better temporal action localization. IEEE Transactions on Visualization and Computer Graphics 30, 6 (2024), 2903–2915

    work page 2024

  7. [7]

    Changjian Chen, Fei Lv, Yalong Guan, Pengcheng Wang, Shengjie Yu, Yifan Zhang, and Zhuo Tang. 2025. Human-guided image generation for expanding small-scale training image datasets. IEEE Transactions on Visualization and Computer Graphics (2025)

    work page 2025

  8. [8]

    Jun Chen, Hong Chen, Xue Jiang, Bin Gu, Weifu Li, Tieliang Gong, and Feng Zheng. 2023. On the Stability and Generalization of Triplet Learning. In Thirty- Seventh AAAI Conference on Artificial Intelligence, AAAI 2023, Thirty-Fifth Con- ference on Innovative Applications of Artificial Intelligence, IAAI 2023, Thirteenth Symposium on Educational Advances in...

    work page 2023

Show all 74 references
  1. [9]

    Pengguang Chen, Shu Liu, Hengshuang Zhao, Xingquan Wang, and Jiaya Jia

  2. [10]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A Simple Framework for Contrastive Learning of Visual Representations. In Proceedings of the International Conference on Machine Learning . PMLR, Virtual Only (formerly Vienna), 1597–1607

    2020

  3. [11]

    Mircea Cimpoi, Subhransu Maji, Iasonas Kokkinos, Sammy Mohamed, and An- drea Vedaldi. 2014. Describing Textures in the Wild. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . IEEE Computer Society, Columbus, OH, USA, 3606–3613

    2014

  4. [12]

    Ekin D Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasudevan, and Quoc V Le

  5. [13]

    Ekin Dogus Cubuk, Barret Zoph, Jonathon Shlens, and Quoc Le. 2020. RandAug- ment: Practical Automated Data Augmentation with a Reduced Search Space. In Advances in Neural Information Processing Systems 33: Annual Conference on Neu- ral Information Processing Systems 2020, Neur...

    2020

  6. [14]

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition . Ieee, IEEE Computer Society, Miami, Florida, USA, 248–255

    2009

  7. [15]

    Terrance DeVries. 2017. Improved Regularization of Convolutional Neural Net- works with Cutout. arXiv preprint arXiv:1708.04552 (2017)

    2017 arXiv

  8. [16]

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. 2024. Scaling rectified flow transformers for high-resolution image synthesis. In Forty- first international conference on m...

    2024

  9. [17]

    Mark Everingham, Luc Van Gool, Christopher K. I. Williams, John M. Winn, and Andrew Zisserman. 2010. The Pascal Visual Object Classes (VOC) Challenge. Int. J. Comput. Vis. 88, 2 (2010), 303–338. doi:10.1007/S11263-009-0275-4

    2010 doi

  10. [18]

    Li Fei-Fei, Robert Fergus, and Pietro Perona. 2007. Learning generative visual models from few training examples: An incremental Bayesian approach tested on 101 object categories. Comput. Vis. Image Underst. 106, 1, 59–70

    2007

  11. [19]

    Lei Feng, Senlin Shu, Zhouyi Lin, Fengmao Lv, Li Li, and Bo An. 2020. Can Cross Entropy Loss Be Robust to Label Noise?. InProceedings of the h International Joint Conference on Artificial Intelligence. ijcai.org, Yokohama, Japan, 2206–2212

    2020

  12. [20]

    Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patashnik, Amit Haim Bermano, Gal Chechik, and Daniel Cohen-or. 2023. An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion. In The Eleventh International Conference on Learning Representations . Op...

    2023

  13. [21]

    Lan-Zhe Guo, Zhen-Yu Zhang, Yuan Jiang, Yu-Feng Li, and Zhi-Hua Zhou. 2020. Safe deep semi-supervised learning for unseen-class unlabeled data. In Interna- tional conference on machine learning . PMLR, PMLR, Virtual Event, 3897–3906

    2020

  14. [22]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition . IEEE Computer Society, Las Vegas, NV, USA, 770– 778

    2016

  15. [23]

    Ruifei He, Shuyang Sun, Xin Yu, Chuhui Xue, Wenqing Zhang, Philip Torr, Song Bai, and Xiaojuan Qi. 2022. Is synthetic data from generative models ready for image recognition? arXiv preprint arXiv:2210.07574 (2022)

    2022 arXiv

  16. [24]

    Dan Hendrycks, Norman Mu, Ekin D Cubuk, Barret Zoph, Justin Gilmer, and Balaji Lakshminarayanan. 2019. Augmix: A simple data processing method to improve robustness and uncertainty. arXiv preprint arXiv:1912.02781 (2019)

    2019 arXiv

  17. [25]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. Advances in neural information processing systems 33 (2020), 6840–6851

    2020

  18. [26]

    Jonathan Ho and Tim Salimans. 2022. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598 (2022)

    2022 arXiv

  19. [27]

    Chengkai Hou, Jieyu Zhang, and Tianyi Zhou. 2023. When to learn what: Model- adaptive data augmentation curriculum. In Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision . IEEE, Paris, France, 1717–1728

    2023

  20. [28]

    Khawar Islam, Muhammad Zaigham Zaheer, Arif Mahmood, and Karthik Nan- dakumar. 2024. DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, Seattle, WA, USA, 27621–27630

    2024

  21. [29]

    Val- ous, Dyke Ferber, Lina Jansen, Constantino Carlos Reyes-Aldasoro, Inka Zörnig, Dirk Jäger, Hermann Brenner, Jenny Chang-Claude, Michael Hoffmeister, and Niels Halama

    Jakob Nikolas Kather, Johannes Krisam, Pornpimol Charoentong, Tom Luedde, Esther Herpel, Cleo-Aron Weis, Timo Gaiser, Alexander Marx, Nektarios A. Val- ous, Dyke Ferber, Lina Jansen, Constantino Carlos Reyes-Aldasoro, Inka Zörnig, Dirk Jäger, Hermann Brenner, Jenny Chang-Claud...

    2019

  22. [30]

    Jonathan Krause, Jia Deng, Michael Stark, and Li Fei-Fei. 2013. Collecting a large-scale dataset of fine-grained cars. In In Workshop on Fine-Grained Visual Categorization

    2013

  23. [31]

    Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 2013. 3d object repre- sentations for fine-grained categorization. In Proceedings of the IEEE international conference on computer vision workshops . IEEE Computer Society, Sydney, Aus- tralia, 554–561

    2013

  24. [32]

    Alex Krizhevsky and Geoffrey Hinton. 2009. Learning multiple layers of features from tiny images. Technical Report. University of Toronto

    2009

  25. [33]

    Yunwen Lei, Tianbao Yang, Yiming Ying, and Ding-Xuan Zhou. 2023. General- ization Analysis for Contrastive Representation Learning. In Proceedings of the International Conference on Machine Learning . PMLR, Honolulu, Hawaii, USA, 19200–19227

    2023

  26. [34]

    Bohan Li, Xiao Xu, Xinghao Wang, Yutai Hou, Yunlong Feng, Feng Wang, Xuan- liang Zhang, Qingfu Zhu, and Wanxiang Che. 2024. Semantic-Guided Generative Image Augmentation Method with Diffusion Models for Image Classification. In Proceedings of the AAAI Conference on Artificial ...

    2024

  27. [35]

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and gen- eration. In International conference on machine learning . PMLR, PMLR, Baltimore, Maryland, USA, 12888–12900

    2022

  28. [36]

    Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C

    Tsung-Yi Lin, Michael Maire, Serge J. Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C. Lawrence Zitnick. 2014. Microsoft COCO: Common Objects in Context. In Computer Vision - ECCV 2014 - 13th European Conference, Zurich, Switzerland, September 6-12, 2014...

    2014 doi

  29. [37]

    Subhransu Maji, Esa Rahtu, Juho Kannala, Matthew Blaschko, and Andrea Vedaldi

  30. [38]

    Samuel G Müller and Frank Hutter. 2021. Trivialaugment: Tuning-free yet state-of- the-art data augmentation. In Proceedings of the IEEE/CVF international conference �� ���� ������� ������ ����� ������� �������� ���� ����� �� ��� on computer vision. IEEE, Montreal, QC, Canada, 774–782

    2021

  31. [39]

    Maria-Elena Nilsback and Andrew Zisserman. 2008. Automated flower classifica- tion over a large number of classes. In 2008 Sixth Indian conference on computer vision, graphics & image processing . IEEE, IEEE Computer Society, Bhubaneswar, India, 722–729

    2008

  32. [40]

    Omkar M Parkhi, Andrea Vedaldi, Andrew Zisserman, and C. V. Jawahar. 2012. Cats and dogs. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE Computer Society, Providence, RI, USA, 3498–3505

    2012

  33. [41]

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. 2023. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952 (2023)

    2023 arXiv

  34. [42]

    Guo-Jun Qi and Jiebo Luo. 2020. Small data challenges in big data era: A survey of recent progress on unsupervised and semi-supervised methods.IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 4 (2020), 2168–2187

    2020

  35. [43]

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research 21, 140 (2020), 1–67

    2020

  36. [44]

    Mengye Ren, Wenyuan Zeng, Bin Yang, and Raquel Urtasun. 2018. Learning to reweight examples for robust deep learning. In International conference on machine learning. PMLR, Stockholm, Sweden, 4334–4343

    2018

  37. [45]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. 2022. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . IEEE, New Orleans, LA, USA, 10684–10695

    2022

  38. [46]

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. 2022. Photorealistic text-to-image diffusion models with deep language understanding. Advances in neural infor...

    2022

  39. [47]

    Dvir Samuel, Rami Ben-Ari, Simon Raviv, Nir Darshan, and Gal Chechik. 2024. Generating images of rare concepts using pre-trained diffusion models. In Pro- ceedings of the AAAI Conference on Artificial Intelligence , Vol. 38. AAAI Press, Vancouver, Canada, 4695–4703

    2024

  40. [48]

    Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang- Chieh Chen. 2018. Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on computer vision and pattern recognition . Computer Vision Foundation / IEEE Computer Socie...

    2018

  41. [49]

    Teppei Suzuki. 2022. Teachaugment: Data augmentation optimization using teacher knowledge. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. IEEE, New Orleans, LA, USA, 10904–10914

    2022

  42. [50]

    Brandon Trabucco, Kyle Doherty, Max Gurinas, and Ruslan Salakhutdinov. 2024. Effective Data Augmentation With Diffusion Models. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11,

    2024

  43. [51]

    Catherine Wah, Steve Branson, Peter Welinder, Pietro Perona, and Serge Belongie

  44. [52]

    Jun Wang, Bohan Lei, Liya Ding, Xiaoyin Xu, Xianfeng Gu, and Min Zhang. 2024. Autoencoder-based conditional optimal transport generative adversarial network for medical image generation. Visual Informatics 8, 1 (2024), 15–25

    2024

  45. [53]

    Weijia Wu, Yuzhong Zhao, Mike Zheng Shou, Hong Zhou, and Chunhua Shen

  46. [54]

    Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. 2017. Aggregated residual transformations for deep neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition . IEEE Computer Society, Honolulu, HI, USA, 1492–1500

    2017

  47. [55]

    Xuanang Xu, Fugen Zhou, Bo Liu, Dongshan Fu, and Xiangzhi Bai. 2019. Efficient Multiple Organ Localization in CT Image Using 3D Region Proposal Network. IEEE Transactions on Medical Imaging 38, 8 (2019), 1885–1898

    2019

  48. [56]

    Jiancheng Yang, Rui Shi, and Bingbing Ni. 2021. MedMNIST Classification Decathlon: A Lightweight AutoML Benchmark for Medical Image Analysis. In Proceedings of the International Symposium on Biomedical Imaging . IEEE, Nice, France, 191–195

    2021

  49. [57]

    Suorong Yang, Furao Shen, and Jian Zhao. 2024. EntAugment: Entropy-Driven Adaptive Data Augmentation Framework for Image Classification. In European Conference on Computer Vision . Springer, Springer, Milan, Italy, 197–214

    2024

  50. [58]

    Weikai Yang, Yukai Guo, Jing Wu, Zheng Wang, Lan-Zhe Guo, Yu-Feng Li, and Shixia Liu. 2024. Interactive Reweighting for Mitigating Label Quality Issues. IEEE Transactions on Visualization and Computer Graphics30, 3 (2024), 1837–1852

    2024

  51. [59]

    Liang Yuan, Dingkun Yan, Suguru Saito, and Issei Fujishiro. 2024. DiffMat: Latent diffusion models for image-guided material generation. Visual Informatics 8, 1 (2024), 6–14

    2024

  52. [60]

    Sangdoo Yun, Dongyoon Han, Sanghyuk Chun, Seong Joon Oh, Youngjoon Yoo, and Junsuk Choe. 2019. CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features. In 2019 IEEE/CVF International Conference on Computer Vision, ICCV 2019, Seoul, Korea (South), ...

    2019

  53. [61]

    Sergey Zagoruyko and Nikos Komodakis. 2016. Wide Residual Networks. In British Machine Vision Conference 2016. British Machine Vision Association, BMVA Press, York, UK

    2016

  54. [62]

    L Zhang, Z Deng, K Kawaguchi, A Ghorbani, and J Zou. 2021. How Does Mixup Help With Robustness and Generalization? (2021)

    2021

  55. [63]

    Qifeng Zhang, Changjian Chen, Zhizhong Liu, and Zhuo Tang. 2024. I-adapt: Using iou adapter to improve pseudo labels in cross-domain object detection. In ECAI 2024. IOS Press, 57–64

    2024

  56. [64]

    Yifan Zhang, Daquan Zhou, Bryan Hooi, Kai Wang, and Jiashi Feng. 2024. Expand- ing small-scale datasets with guided imagination. Advances in Neural Information Processing Systems 36 (2024)

    2024

  57. [65]

    Hongliang Zhong, Can Wang, Jingbo Zhang, and Jing Liao. 2025. Generative object insertion in gaussian splatting with a multi-view diffusion model. Visual Informatics (2025), 100238

    2025

  58. [66]

    Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. 2020. Random erasing data augmentation. In Proceedings of the AAAI conference on artificial intelligence, Vol. 34. AAAI Press, New York, NY, USA, 13001–13008

    2020

  59. [67]

    IEEE, Korea (South), 6022–6031

  60. [74]

    Haowei Zhu, Ling Yang, Jun-Hai Yong, Wentao Zhang, and Bin Wang. 2024. Distribution-Aware Data Expansion with Diffusion Models. arXiv preprint arXiv:2403.06741 (2024). ��������� ����������� ������� ��������� ���� ������������������������ ������ ������������ �� ���� ������� ���...

    2024 arXiv

  61. [2011]

    The caltech-ucsd birds-200-2011 dataset. (2011)

    2011

  62. [2013]

    arXiv preprint arXiv:1306.5151 abs/1306.5151 (2013)

    Fine-grained visual classification of aircraft. arXiv preprint arXiv:1306.5151 abs/1306.5151 (2013)

    2013 arXiv

  63. [2019]

    In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Autoaugment: Learning augmentation strategies from data. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . Computer Vision Foundation / IEEE, Long Beach, CA, USA, 113–123

  64. [2020]

    arXiv preprint arXiv:2001.04086 (2020)

    GridMask Data Augmentation. arXiv preprint arXiv:2001.04086 (2020)

    2001 arXiv

  65. [2023]

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

    Diffumask: Synthesizing images with pixel-level annotations for semantic segmentation using diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision . IEEE, Paris, France, 1206–1217

  66. [2024]

    OpenReview.net, Vienna, Austria

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.