pith. sign in

arxiv: 2605.31153 · v1 · pith:PBOFWP4Lnew · submitted 2026-05-29 · 💻 cs.CV

BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors

Jonas Ricker , Asja Fischer , Erwin Quiring This is my paper

Pith reviewed 2026-06-28 22:44 UTC · model grok-4.3

classification 💻 cs.CV
keywords AI-generated image detectiontransformation biasbias evaluationimage forensicsdetector robustnessspurious correlationsforensic artifacts
0
0 comments X

The pith

BIAS-ID framework shows many state-of-the-art AI image detectors rely on transformation biases rather than true forensic signals.

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

Detectors that distinguish real images from AI-generated ones often work in controlled tests but fail on real-world data. The paper distinguishes this failure mode from general lack of robustness by focusing on transformation biases, where detectors pick up spurious cues introduced by specific image edits instead of authentic artifacts. It introduces BIAS-ID as a structured way to apply controlled transformations and measure how much each detector depends on them. Testing six detectors on two datasets finds that several leading methods are strongly affected. Understanding these biases matters because it explains why current detectors cannot be trusted outside narrow settings and points to the need for evaluation that accounts for them.

Core claim

The BIAS-ID framework provides a transparent protocol to quantify transformation biases in AI-generated image detectors by testing performance shifts under controlled transformations, separating this from general robustness failures. Validation across six detectors and two datasets shows that several state-of-the-art methods exhibit strong dependence on these biases.

What carries the argument

BIAS-ID, a framework that applies systematic transformations to input images and measures resulting changes in detector performance to isolate bias from other robustness issues.

If this is right

  • Detectors must undergo bias-aware testing before claims of reliability can be accepted.
  • Training data and procedures need adjustment to reduce dependence on transformation-specific cues.
  • Benchmarks should incorporate transformation analysis to avoid overestimating real-world performance.
  • New detector designs should target features that remain stable across transformations.

Where Pith is reading between the lines

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

  • The same bias-measurement approach could be applied to detectors for other media such as video or audio.
  • Developers could use BIAS-ID results to prioritize which transformations to mitigate during training.
  • As generation tools evolve, periodic re-evaluation with BIAS-ID would track whether new biases emerge.

Load-bearing premise

The chosen transformations and datasets capture the real-world variations that cause detectors to fail, and the framework can cleanly separate transformation bias from other generalization problems.

What would settle it

A detector that passes BIAS-ID tests with low bias scores but still collapses on diverse real-world images, or conversely one that shows high bias scores yet maintains performance when transformations are controlled.

Figures

Figures reproduced from arXiv: 2605.31153 by Asja Fischer, Erwin Quiring, Jonas Ricker.

Figure 1
Figure 1. Figure 1: Bias analysis for three detectors with varying degrees of bias w.r.t. JPEG (quality factor [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Bias analysis for JPEG compression. For each detector, we report mean scores, score [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Bias analysis for WebP compression. Resizing Compared to compression, detectors are less biased w.r.t. resizing [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Bias analysis for bilinear resizing. 0 1 Score 1 0 1 Shift -135 -90 -45 0 45 90 135 180 Rotation (Degree) 0 1 AUC (a) UnivFD 0 1 Score 1 0 1 Shift -135 -90 -45 0 45 90 135 180 Rotation (Degree) 0 1 AUC (b) DRCT 0 1 Score 1 0 1 Shift -135 -90 -45 0 45 90 135 180 Rotation (Degree) 0 1 AUC (c) RINE 0 1 Score 1 0 1 Shift -135 -90 -45 0 45 90 135 180 Rotation (Degree) 0 1 AUC (d) AIDE 0 1 Score 1 0 1 Shift -135… view at source ↗
Figure 5
Figure 5. Figure 5: Bias analysis for rotation. Color Finally, we analyze the bias sensitivity w.r.t. image color space in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Bias analysis for color. 6 Discussion Key Takeaways Our evaluation shows that transformation biases in AIGI detectors are a common phenomenon that can affect their performance when tested on real-world data. We observe that simple 8 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Bias analysis for JPEG compression with plots for SynthCLIC. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Bias analysis for WebP compression with plots for SynthCLIC. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Bias analysis for bilinear resizing with plots for SynthCLIC. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Bias analysis for rotation with plots for SynthCLIC. [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Bias analysis for color with plots for SynthCLIC. [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Bias analysis for bicubic resizing with plots for SynthBuster. [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Bias analysis for bicubic resizing with plots for SynthCLIC. [PITH_FULL_IMAGE:figures/full_fig_p015_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Bias analysis for JPEG compression with plots for Synthbuster (original RAISE-1k). [PITH_FULL_IMAGE:figures/full_fig_p016_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Bias analysis for bilinear resizing with plots for Synthbuster (original RAISE-1k). [PITH_FULL_IMAGE:figures/full_fig_p016_15.png] view at source ↗
read the original abstract

Given the surge of harmful AI-generated imagery online, reliably distinguishing authentic images from generated ones has become an urgent research topic. While many proposed detection methods perform well under controlled settings, they often collapse when tested on real-world data. A potential root cause are subtle biases in the detectors' training data. As a result, detectors may rely on spurious correlations instead of learning true forensic artifacts. While a recent line of work has identified the problem, there is not yet an established protocol to evaluate how biased a detector actually is. In this work, we therefore take a step back: First, we discuss what it means for a detector to be biased, and how this differs from a lack of robustness. Second, we propose BIAS-ID, a transparent framework for analyzing and quantifying the presence of transformation biases in AI-generated image detectors. We validate our framework by performing an evaluation of six detectors across two datasets, revealing that several state-of-the-art detection methods are strongly affected by biases. Our results highlight the importance of bias-aware evaluation for developing reliable AI-generated image detectors.

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.

Referee Report

0 major / 2 minor

Summary. The paper distinguishes transformation bias (spurious correlations from training data transformations) from general lack of robustness in AI-generated image detectors. It proposes the BIAS-ID framework as a transparent protocol to quantify such biases via controlled evaluations, then applies it to six detectors across two datasets and concludes that several state-of-the-art methods are strongly affected by biases.

Significance. If the framework successfully isolates transformation bias and the evaluations are reproducible, the work supplies a needed evaluation protocol that could improve the reliability of forensic detectors. The multi-detector, multi-dataset validation provides concrete evidence supporting the central claim and gives credit for addressing a practical gap in the literature.

minor comments (2)
  1. The abstract states that BIAS-ID 'quantifies the presence of transformation biases' but does not name the exact bias metric or the controlled transformation set; adding these definitions in §3 would improve clarity without altering the central claim.
  2. The validation section reports results on six detectors and two datasets; including the precise list of detectors, datasets, and the bias scores in a table would make the 'strongly affected' conclusion easier to verify.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary of our work and the recommendation for minor revision. The provided summary accurately captures the distinction between transformation bias and general robustness, as well as the BIAS-ID framework and its validation on six detectors across two datasets.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper proposes BIAS-ID as an evaluation framework that distinguishes transformation bias from general lack of robustness, then applies it via controlled experiments on six detectors across two datasets. No equations, derivations, fitted parameters, or predictions are described that could reduce to inputs by construction. The abstract and outline reference prior work only in passing without load-bearing self-citations or uniqueness theorems. The central contribution is an empirical protocol whose validity rests on the experimental design rather than self-referential definitions or renamings.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no information on free parameters, axioms, or invented entities can be extracted.

pith-pipeline@v0.9.1-grok · 5720 in / 977 out tokens · 19247 ms · 2026-06-28T22:44:01.103669+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 2 canonical work pages

  1. [1]

    PyTorch 2: Faster machine learning through dynamic python bytecode transformation and graph compi- lation

    Jason Ansel, Edward Yang, Horace He, Natalia Gimelshein, Animesh Jain, Michael V oznesen- sky, Bin Bao, Peter Bell, David Berard, Evgeni Burovski, Geeta Chauhan, Anjali Chourdia, Will Constable, Alban Desmaison, Zachary DeVito, Elias Ellison, Will Feng, Jiong Gong, Michael Gschwind, Brian Hirsh, Sherlock Huang, Kshiteej Kalambarkar, Laurent Kirsch, Michae...

    2024

  2. [2]

    Synthbuster: Towards detection of diffusion model generated images.IEEE Open Journal of Signal Processing (OJSP), 2023

    Quentin Bammey. Synthbuster: Towards detection of diffusion model generated images.IEEE Open Journal of Signal Processing (OJSP), 2023

    2023

  3. [3]

    Recognition in terra incognita

    Sara Beery, Grant Van Horn, and Pietro Perona. Recognition in terra incognita. InEuropean Conference on Computer Vision (ECCV), 2018

    2018

  4. [4]

    Vbench: Comprehensive benchmark suite for video generative models

    George Cazenavette, Avneesh Sud, Thomas Leung, and Ben Usman. FakeInversion: Learning to detect images from unseen text-to-image models by inverting Stable Diffusion. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024.https://doi.org/ 10.1109/CVPR52733.2024.01023

    work page doi:10.1109/cvpr52733.2024.01023 2024

  5. [5]

    What makes fake images detectable? Understanding properties that generalize

    Lucy Chai, David Bau, Ser-Nam Lim, and Phillip Isola. What makes fake images detectable? Understanding properties that generalize. InEuropean Conference on Computer Vision (ECCV), 2020

    2020

  6. [6]

    DRCT: Diffusion reconstruction contrastive training towards universal detection of diffusion generated images

    Baoying Chen, Jishen Zeng, Jianquan Yang, and Rui Yang. DRCT: Diffusion reconstruction contrastive training towards universal detection of diffusion generated images. InInternational Conference on Machine Learning (ICML), 2024

    2024

  7. [7]

    Hyeongjun Choi, Inho Jung, and Simon S. Woo. Combating dataset misalignment for robust AI-generated image detection in the real world. InWorkshop on Security Implications of Deepfakes and Cheapfakes, 2025

    2025

  8. [8]

    Workshop and challenge on learned image compression

    CLIC2020. Workshop and challenge on learned image compression. https://archive.co mpression.cc/, 2020. Accessed May 5, 2026

    2020

  9. [9]

    C2PA.https://c2pa.org/

    Coalition for Content Provenance and Authenticity (C2PA). C2PA.https://c2pa.org/

  10. [10]

    Unveiling the impact of image transformations on deepfake detection: An experimental analysis

    Federico Cocchi, Lorenzo Baraldi, Samuele Poppi, Marcella Cornia, Lorenzo Baraldi, and Rita Cucchiara. Unveiling the impact of image transformations on deepfake detection: An experimental analysis. InInternational Conference on Image Analysis and Processing (ICIAP), 2023

    2023

  11. [11]

    As good as a coin toss: Human detection of AI-generated images, videos, audio, and audiovisual stimuli.arXiv Preprint, 2025

    Di Cooke, Abigail Edwards, Sophia Barkoff, and Kathryn Kelly. As good as a coin toss: Human detection of AI-generated images, videos, audio, and audiovisual stimuli.arXiv Preprint, 2025

    2025

  12. [12]

    Intrigu- ing properties of synthetic images: From generative adversarial networks to diffusion models

    Riccardo Corvi, Davide Cozzolino, Giovanni Poggi, Koki Nagano, and Luisa Verdoliva. Intrigu- ing properties of synthetic images: From generative adversarial networks to diffusion models. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023

    2023

  13. [13]

    RAISE: A raw images dataset for digital image forensics

    Duc-Tien Dang-Nguyen, Cecilia Pasquini, Valentina Conotter, and Giulia Boato. RAISE: A raw images dataset for digital image forensics. InACM International Conference on Multimedia (MM), 2015.https://dl.acm.org/doi/10.1145/2713168.2713194

    work page doi:10.1145/2713168.2713194 2015

  14. [14]

    ImageNet: A large- scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. ImageNet: A large- scale hierarchical image database. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2009

    2009

  15. [15]

    AMMeBa: A large-scale survey and dataset of media-based misinformation in-the-wild.arXiv Preprint, 2024

    Nicholas Dufour, Arkanath Pathak, Pouya Samangouei, Nikki Hariri, Shashi Deshetti, Andrew Dudfield, Christopher Guess, Pablo Hernández Escayola, Bobby Tran, Mevan Babakar, and Christoph Bregler. AMMeBa: A large-scale survey and dataset of media-based misinformation in-the-wild.arXiv Preprint, 2024

    2024

  16. [16]

    A survey on bias in visual datasets.Computer Vision and Image Understanding, 2022

    Simone Fabbrizzi, Symeon Papadopoulos, Eirini Ntoutsi, and Ioannis Kompatsiaris. A survey on bias in visual datasets.Computer Vision and Image Understanding, 2022

    2022

  17. [17]

    Malicious actors almost certainly will leverage synthetic content for cyber and foreign influence operations, 2021

    Federal Bureau of Investigation (FBI). Malicious actors almost certainly will leverage synthetic content for cyber and foreign influence operations, 2021. https://www.ic3.gov/Media/Ne ws/2021/210310-2.pdf, Accessed May 5, 2026. 10

    2021

  18. [18]

    Criminals use generative artificial intelligence to facilitate financial fraud, 2024

    Federal Bureau of Investigation (FBI). Criminals use generative artificial intelligence to facilitate financial fraud, 2024. https://www.ic3.gov/PSA/2024/PSA241203 , Accessed May 5, 2026

    2024

  19. [19]

    The stable signature: Rooting watermarks in latent diffusion models

    Pierre Fernandez, Guillaume Couairon, Hervé Jégou, Matthijs Douze, and Teddy Furon. The stable signature: Rooting watermarks in latent diffusion models. InIEEE/CVF International Conference on Computer Vision (ICCV), 2023

    2023

  20. [20]

    Leveraging frequency analysis for deep fake image recognition

    Joel Frank, Thorsten Eisenhofer, Lea Schönherr, Asja Fischer, Dorothea Kolossa, and Thorsten Holz. Leveraging frequency analysis for deep fake image recognition. InInternational Confer- ence on Machine Learning (ICML), 2020

    2020

  21. [21]

    A representative study on human detection of artificially generated media across countries

    Joel Frank, Franziska Herbert, Jonas Ricker, Lea Schönherr, Thorsten Eisenhofer, Asja Fischer, Markus Dürmuth, and Thorsten Holz. A representative study on human detection of artificially generated media across countries. InIEEE Symposium on Security and Privacy (SP), 2024

    2024

  22. [22]

    Wichmann

    Robert Geirhos, Jörn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Felix A. Wichmann. Shortcut learning in deep neural networks.Nature Machine Intelligence, 2020

    2020

  23. [23]

    Gragnaniello, D

    D. Gragnaniello, D. Cozzolino, F. Marra, G. Poggi, and L. Verdoliva. Are GAN generated images easy to detect? A critical analysis of the state-of-the-art. InIEEE International Conference on Multimedia and Expo (ICME), 2021

    2021

  24. [24]

    Will AI-generated images create a new crisis for fact-checkers? Experts are not so sure.Reuters Institute for the Study of Journalism, 2023

    Gretel Kahn. Will AI-generated images create a new crisis for fact-checkers? Experts are not so sure.Reuters Institute for the Study of Journalism, 2023. https://reutersinstitute .politics.ox.ac.uk/news/will-ai-generated-images-create-new-crisis-fac t-checkers-experts-are-not-so-sure, Accessed May 5, 2026

    2023

  25. [25]

    Fake or JPEG? Revealing common biases in generated image detection datasets

    Patrick Grommelt, Louis Weiss, Franz-Josef Pfreundt, and Janis Keuper. Fake or JPEG? Revealing common biases in generated image detection datasets. InEuropean Conference on Computer Vision (ECCV) Workshops, 2024

    2024

  26. [26]

    A bias-free training paradigm for more general AI-generated image detection

    Fabrizio Guillaro, Giada Zingarini, Ben Usman, Avneesh Sud, Davide Cozzolino, and Luisa Verdoliva. A bias-free training paradigm for more general AI-generated image detection. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

    2025

  27. [27]

    An undetectable watermark for generative image models

    Sam Gunn, Xuandong Zhao, and Dawn Song. An undetectable watermark for generative image models. InInternational Conference on Learning Representations (ICLR), 2025

    2025

  28. [28]

    Reducing the content bias for AI-generated image detection

    Seoyeon Gye, Junwon Ko, Hyounguk Shon, Minchan Kwon, and Junmo Kim. Reducing the content bias for AI-generated image detection. InIEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025

    2025

  29. [29]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2016

    2016

  30. [30]

    Any-resolution AI-generated image detection by spectral learning

    Dimitrios Karageorgiou, Symeon Papadopoulos, Ioannis Kompatsiaris, and Efstratios Gavves. Any-resolution AI-generated image detection by spectral learning. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

    2025

  31. [31]

    Leveraging representations from intermediate encoder-blocks for synthetic image detection

    Christos Koutlis and Symeon Papadopoulos. Leveraging representations from intermediate encoder-blocks for synthetic image detection. InEuropean Conference on Computer Vision (ECCV), 2025

    2025

  32. [32]

    Improving synthetic image detection towards generalization: An image transformation perspective

    Ouxiang Li, Jiayin Cai, Yanbin Hao, Xiaolong Jiang, Yao Hu, and Fuli Feng. Improving synthetic image detection towards generalization: An image transformation perspective. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2025

    2025

  33. [33]

    Landscape more secure than portrait? Zooming into the directionality of digital images with security implications

    Benedikt Lorch and Rainer Böhme. Landscape more secure than portrait? Zooming into the directionality of digital images with security implications. InUSENIX Security Symposium, 2024. 11

    2024

  34. [34]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations (ICLR), 2019

    2019

  35. [35]

    Nightingale and Hany Farid

    Sophie J. Nightingale and Hany Farid. AI-synthesized faces are indistinguishable from real faces and more trustworthy.Proceedings of the National Academy of Sciences (PNAS), 2022

    2022

  36. [36]

    Towards universal fake image detectors that generalize across generative models

    Utkarsh Ojha, Yuheng Li, and Yong Jae Lee. Towards universal fake image detectors that generalize across generative models. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  37. [37]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar- wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. InInterna- tional Conference on Machine Learning (ICML), 2021

    2021

  38. [38]

    Aligned datasets improve detection of latent diffusion-generated images

    Anirudh Sundara Rajan, Utkarsh Ojha, Jedidiah Schloesser, and Yong Jae Lee. Aligned datasets improve detection of latent diffusion-generated images. InInternational Conference on Learning Representations (ICLR), 2025

    2025

  39. [39]

    AI- generated faces in the real world: A large-scale case study of Twitter profile images

    Jonas Ricker, Dennis Assenmacher, Thorsten Holz, Asja Fischer, and Erwin Quiring. AI- generated faces in the real world: A large-scale case study of Twitter profile images. In International Symposium on Research in Attacks, Intrusions and Defenses (RAID), 2024

    2024

  40. [40]

    High- resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High- resolution image synthesis with latent diffusion models. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022

    2022

  41. [41]

    How generative AI is boosting the spread of disinformation and propaganda

    Tate Ryan-Mosley. How generative AI is boosting the spread of disinformation and propaganda. MIT Technology Review, 2023. https://www.technologyreview.com/2023/10/04/10 80801/generative-ai-boosting-disinformation-and-propaganda-freedom-hou se/, Accessed May 5, 2026

    2023

  42. [42]

    On the robustness and generalizability of face synthesis detection methods

    Johan Sabel and Fredrik Johansson. On the robustness and generalizability of face synthesis detection methods. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021

    2021

  43. [43]

    Antonio Torralba and Alexei A. Efros. Unbiased look at dataset bias. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2011

    2011

  44. [44]

    Sheng-Yu Wang, Oliver Wang, Richard Zhang, Andrew Owens, and Alexei A. Efros. CNN- generated images are surprisingly easy to spot... for now. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020

    2020

  45. [45]

    Synthetic image detection with CLIP: Understanding and assessing predictive cues.arXiv Preprint, 2026

    Marco Willi, Melanie Mathys, and Michael Graber. Synthetic image detection with CLIP: Understanding and assessing predictive cues.arXiv Preprint, 2026

    2026

  46. [46]

    On the generalization of GAN image forensics

    Xinsheng Xuan, Bo Peng, Wei Wang, and Jing Dong. On the generalization of GAN image forensics. InChinese Conference on Biometric Recognition (CCBR), 2019

    2019

  47. [47]

    A sanity check for AI-generated image detection

    Shilin Yan, Ouxiang Li, Jiayin Cai, Yanbin Hao, Xiaolong Jiang, Yao Hu, and Weidi Xie. A sanity check for AI-generated image detection. InInternational Conference on Learning Representations (ICLR), 2025

    2025

  48. [48]

    Breaking semantic artifacts for generalized AI-generated image detection

    Chende Zheng, Chenhao Lin, Zhengyu Zhao, Hang Wang, Xu Guo, Shuai Liu, and Chao Shen. Breaking semantic artifacts for generalized AI-generated image detection. InAdvances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  49. [49]

    HiDDeN: Hiding data with deep networks

    Jiren Zhu, Russell Kaplan, Justin Johnson, and Li Fei-Fei. HiDDeN: Hiding data with deep networks. InEuropean Conference on Computer Vision (ECCV), 2018

    2018

  50. [50]

    GenImage: A million-scale benchmark for detecting AI-generated image.Advances in Neural Information Processing Systems (NeurIPS), 2023

    Mingjian Zhu, Hanting Chen, Qiangyu Yan, Xudong Huang, Guanyu Lin, Wei Li, Zhijun Tu, Hailin Hu, Jie Hu, and Yunhe Wang. GenImage: A million-scale benchmark for detecting AI-generated image.Advances in Neural Information Processing Systems (NeurIPS), 2023. 12 A Experimental Details Training DetailsThe ResNet-50 detectors analyzed in Section 4 were initial...

    2023