Pith. sign in

REVIEW 1 major objections 6 minor 2 cited by

A Bias-Free Training Paradigm for More General AI-generated Image Detection

T0 review · 1 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read This paper claims that training on self-conditioned Stable Diffusion fakes, which regenerate real photos so that real and fake differ only in synthesis artifacts, yields a detector that generalizes to unseen generators and stays…

desk verdict Solid, useful training-data recipe for AI-image detection; the empirical case is strong, but the 'bias-free' framing overpromises and missing error bars need attention. read the letter →

arxiv 2412.17671 v2 pith:UUHRTLOD submitted 2024-12-23 cs.CV

Fabrizio Guillaro , Giada Zingarini , Ben Usman , Avneesh Sud , Davide Cozzolino , Luisa Verdoliva This is my paper · ORCID
classification cs.CV
keywords AI-generatedimagedetectiondeepfaketrainingdatabiassemanticalignmentself-conditioningStableDiffusioninpaintingcontentaugmentationdetectorcalibration
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

AI-generated image detectors often fail in the real world because they learn spurious cues from their training data, such as compression, resolution, or subject matter, rather than the subtle artifacts left by synthesis. This paper argues the fix is not a better algorithm but a better training set, and proposes B-Free, a paradigm in which fake images are created by regenerating real photographs through Stable Diffusion's conditioning procedure. Because each fake is a self-conditioned reconstruction of a specific real image, the two differ only in generation artifacts, so the detector cannot shortcut on content or format. Trained end-to-end on these aligned fakes plus inpainting-based content augmentation, a ViT-based detector reaches about 96.4% balanced accuracy averaged over 27 generators, including unseen models like FLUX, Stable Diffusion 3.5, and autoregressive architectures, and remains well-calibrated at a fixed threshold.

What carries the argument

The load-bearing device is the self-conditioned reconstruction: an SD2.1 inpainting model applied with an empty mask to a real photograph, which forces the diffusion process to reproduce the same scene instead of generating a new one from a text prompt. This yields a fake image that is pixel-level aligned with its real counterpart, so the only differences are the low-level artifacts introduced by synthesis. Crucially, unlike plain autoencoder reconstructions, running the full diffusion pass deposits forensic clues at the lowest frequencies, as shown by averaged power-spectrum differences, giving the detector a broader artifact band to exploit. The second component is content augmentation through inpainting, replacing an object with a new one of the same or a different category, with or without restoring the original background, which forces the detector to rely on localized editing artifacts rather than global scene statistics. Together, these mechanisms let a detector trained on only one generative model, SD2.1, separate real from fake for 27 generators.

What would settle it

A concrete test: evaluate B-Free on outputs from a pure pixel-space autoregressive generator that does not use a latent autoencoder or iterative denoising, and where a detector trained on that generator's own fakes is highly accurate. If B-Free's balanced accuracy falls to chance (around 50%) while the control detector performs well, the claim that SD2.1 self-conditioned artifacts span the space of generative artifacts is falsified. A complementary observation is to measure the averaged difference spectrum between real and fake images for a new generator: if that spectrum is indistinguishable from real-versus-real differences at low frequencies, the detector should lose its main cue.

Watch

Extended reading notes

Core claim

B-Free builds a dataset of 51,517 real images from MS-COCO and 309,102 fakes, produced by passing each real image through the Stable Diffusion 2.1 inpainting model with an empty mask, so the diffusion process regenerates the same scene rather than generating a new one from text. The resulting self-conditioned images carry the forensic artifacts of diffusion synthesis while preserving pixel-level semantic content, and averaged power spectra show these artifacts extend to lower frequencies than those from a plain autoencoder reconstruction. Additional content augmentation, which replaces an object with one of the same or a different category and optionally restores the original background, further improves generalization and calibration. Fine-tuned end-to-end on this data, a DINOv2 Vision Transformer with registers, evaluated at 504 by 504 pixels without resizing, achieves an average balanced accuracy of 96.4% and an average AUC of 99.3% across 27 generators, outperforming the second-best state-of-the-art method by 20.7 percentage points in balanced accuracy, and maintains about 92% accuracy on web-scraped viral images months after upload.

Load-bearing premise

The load-bearing premise is that the artifacts present in Stable Diffusion 2.1 self-conditioned reconstructions, produced by empty-mask inpainting, are representative enough of the artifacts of unseen generators, including GANs, FLUX, SD3.5, and autoregressive models, for a detector trained only on those fakes to learn a general real-versus-fake boundary.

Editorial extensions

If this is right

  • A detector trained on semantically aligned, self-conditioned fakes generalizes to generators never seen in training, including GANs, FLUX, Stable Diffusion 3.5, and autoregressive models, sustaining above 90% balanced accuracy on most of them.
  • Content-based augmentation through inpainting substantially improves robustness to JPEG compression, resizing, and blurring, and keeps accuracy high on images that have been reposted online over time, with accuracy around 92% even after many days from the initial post.
  • The B-Free training data improves existing architectures: CLIP- and RINE-based detectors gain large accuracy increases when trained on the B-Free dataset instead of ProGAN or Latent Diffusion data, suggesting that data design can matter more than architectural novelty.
  • Because the method improves calibration metrics (lower NLL and ECE), predictions can be thresholded at 0.5 without per-generator calibration, making the detector directly usable in realistic settings where calibration images are unavailable.
  • The approach reduces the gap between AUC and balanced accuracy, meaning high ranking performance translates into actually usable fixed-threshold decisions rather than merely good ordering.

Reading between the lines

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

  • Editorial inference: The paradigm can be turned into a continuously updating self-labeling pipeline. Any new diffusion-based generator can regenerate an existing corpus of real images to produce training data for the next detector without manual annotation; the paper trains only on SD2.1 fakes, so a natural extension is to pool self-conditioned reconstructions from several diffusion models and mea
  • Editorial inference: The power-spectrum rationale yields a testable predictor. For a new generator, one can compute the average difference between its output spectrum and that of real images, then compare it to the corresponding spectrum for SD2.1 self-conditioned fakes. Generators whose difference spectrum lies far from the training distribution, particularly at low frequencies, should be the har
  • Editorial inference: The paper's success on autoregressive models is attributed to the similarity between their tokenizers and Stable Diffusion's latent embedders. A discriminating stress test is a pure pixel-space autoregressive generator that shares no latent projection or iterative denoising module with Stable Diffusion; if the detector's balanced accuracy collapses on such a model, the general
  • Editorial inference: The content-augmentation recipes (same-category vs. different-category inpainting, background restoration) are coarse but suggest a broader design space. Varying the fraction of locally edited images, or mixing self-conditioned fakes from multiple diffusion steps, could offer a principled way to tune the artifact distribution that the detector must learn.
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 / 6 minor

Summary. The paper proposes B-Free, a training paradigm for AI-generated image detection. The authors construct a dataset of 51,517 real images from COCO and 309,102 synthetic images produced by self-conditioned inpainting with Stable Diffusion 2.1, plus content-augmented variants obtained by inpainting objects with same- or different-category replacements and background restoration. A Vision Transformer based on DINOv2+registers is fine-tuned end-to-end on this dataset. The paper reports balanced accuracy above 90% on average across 27 unseen generators, including recent models such as FLUX and SD3.5, along with improved calibration (lower NLL/ECE) compared with state-of-the-art detectors. Ablations vary the alignment strategy, augmentation type, architecture, and training dataset.

Significance. If the results hold, the paper makes a useful empirical contribution: it demonstrates that semantic alignment of real and fake training images, combined with content-based augmentation, yields detectors that generalize substantially better than standard ProGAN- or LDM-trained detectors. The evaluation is broad (27 generators, multiple independent datasets, ablations controlling architecture and training set), and the authors provide code and data, which supports reproducibility. The main weakness is that the interpretative claim of a 'bias-free' paradigm is stronger than the evidence directly supports: the representativeness of SD2.1 self-conditioned artifacts for unseen generators is inferred only from end-task accuracy, not from a direct analysis of the learned decision cues.

major comments (1)
  1. [Sec. 4.1 and Sec. 6] The central claim that B-Free is 'bias-free' and detects 'subtle artifacts introduced by AI generation' rests on the premise that self-conditioned SD2.1 reconstructions contain artifacts representative of all unseen generators. The paper provides only indirect evidence for this premise: high end-to-end accuracy on 27 generators. Figure 5 shows power spectra only for SD2.1 real/reconstructed/self-conditioned images; it does not compare with spectra or feature distributions of unseen generators such as StyleGAN, FLUX, or SD3.5. The strong balanced accuracy on GANs (e.g., StyleGAN 88.5% in Table 10) is not explained by the diffusion-specific rationale, and Sec. 6 concedes that a completely different synthesis process would likely cause failure. The authors should either (i) provide a direct artifact-overlap analysis (e.g., spectral signatures, feature-space distances, or per-generator error analysis) to support the representativeness assumption, or (ii) revise the abstract and Sec. 4 to state the scope more precisely (e.g., 'diffusion-like generators') instead of claiming a bias-free paradigm. As written, the abstract's causal statement 'allowing any differences to stem solely from the subtle artifacts introduced by AI generation' is not validated.
minor comments (6)
  1. [Sec. 3.1] The construction of 'GenImage unbiased' is underspecified; please state the exact JPEG quality factor (or distribution of qualities) used for the fake images, how it was matched to the real class, and release the modified subset for reproducibility.
  2. [Tables 2, 5, 7-10] All reported numbers are point estimates from single runs with no confidence intervals or repeated seeds; the 'bold with a margin of 1%' criterion is not interpretable without variance estimates, particularly for the small WildRF subsets and for ablation variants in Table 2 that differ by only 1-2%.
  3. [Sec. 1 and Supplementary Material] There are typos ('discrminate' in Sec. 1, 'adpots' and 'deatils' in Sec. B of the supplement, 'is it very likely' in Sec. 6); these should be corrected.
  4. [Figure 5] Please add axis labels and specify whether the power spectra are azimuthally averaged; without this detail the qualitative claim about low-frequency artifacts is hard to reproduce.
  5. [Eq. (1)] With M=15 bins and small test sets, ECE estimates can be biased; please report the average bin counts or use a debiased ECE estimator.
  6. [Table 6] For a fair comparison, indicate the inference resolution used for each baseline and whether the authors' multi-crop averaging is also applied to the baselines.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper's generalization claim is an empirical evaluation result, not a reduction to its own training inputs.

full rationale

The paper's central claim — that fine-tuning a DINOv2+reg detector end-to-end on SD2.1 self-conditioned reconstructions plus inpainting augmentation yields above-90% balanced accuracy on unseen generators — is established by direct external evaluation on Synthbuster, GenImage (unbiased), FakeInversion, FakeBench, WildRF, and SynthWildX (Tables 5, 7, 9, 10). The training fakes are fixed at dataset construction, and the evaluation generators are not used to fit any threshold, calibration constant, or parameter. No equation in the paper defines the predicted quantity in terms of the fitted data; balanced accuracy, ECE, and NLL are standard metrics applied to held-out generators. The only self-citations (e.g., [7], [10], [11]) provide contextual background on semantic alignment and diffusion-based reconstruction; they do not carry the generalization claim, which is tested against external benchmarks and independent methods with publicly available code. The Sec. 6 limitation — 'If new generators will be deployed in the future that have a completely different synthesis process, then it is very likely that this approach will fail' — explicitly conditions the result on artifact representativeness; a conditional premise is a scope statement, not a circular definition. The ablation in Table 2 compares alignment strategies empirically, and the power-spectrum argument in Sec. 4.1 is an observational rationale, not a fitted input relabeled as a prediction. No self-definitional, fitted-input-as-prediction, self-citation-load-bearing, uniqueness-imported, ansatz-smuggled, or renaming pattern is present.

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

This is an empirical machine learning paper, not a derivation. There are no fitted constants in the mathematical sense. The central claim rests on domain assumptions about generation artifacts and evaluation fairness. Standard training hyperparameters are reported in the supplementary but were not fitted to test benchmarks, so they are not counted as free parameters. No new physical or architectural entities are introduced.

assumptions (5)
  • domain assumption Self-conditioned SD2.1 reconstructions of real images are semantically aligned so that remaining differences are due to generation artifacts.
    Core premise of B-Free, stated in Sec. 1 and 4 and used to justify the training set design.
  • domain assumption Artifacts in SD2.1 self-conditioned fakes generalize to unseen generators, including GANs and autoregressive models.
    Needed for the generalization claim; tested empirically but not proven. Sec. 6 acknowledges fragility for very different synthesis processes.
  • domain assumption No-resizing, large-crop training preserves low-level forensic traces needed for detection.
    Invoked in Sec. 4, based on prior work [18].
  • domain assumption The evaluation datasets are sufficiently free of content and format bias to measure true generalization.
    The protocol constructs 'GenImage unbiased' by compressing fakes, and relies on Synthbuster raw format and web datasets; the 'bias-free' claim depends on this protocol being fair.
  • domain assumption Stable Diffusion 2.1 inpainting with an empty mask regenerates the input image rather than copying it, so fakes carry generation artifacts.
    Tooling assumption stated in Sec. 4: 'Using an empty mask, we induce the diffusion steps to regenerate the input'.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Bias-Free Training Paradigm for More General AI-generated Image Detection." pith.science (2026). https://pith.science/paper/UUHRTLOD

@misc{pith2026241217671,
  author       = {Pith},
  title        = {Pith review of: A Bias-Free Training Paradigm for More General AI-generated Image Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UUHRTLOD}},
  note         = {Machine review of arXiv:2412.17671}
}
read the original abstract

Successful forensic detectors can produce excellent results in supervised learning benchmarks but struggle to transfer to real-world applications. We believe this limitation is largely due to inadequate training data quality. While most research focuses on developing new algorithms, less attention is given to training data selection, despite evidence that performance can be strongly impacted by spurious correlations such as content, format, or resolution. A well-designed forensic detector should detect generator specific artifacts rather than reflect data biases. To this end, we propose B-Free, a bias-free training paradigm, where fake images are generated from real ones using the conditioning procedure of stable diffusion models. This ensures semantic alignment between real and fake images, allowing any differences to stem solely from the subtle artifacts introduced by AI generation. Through content-based augmentation, we show significant improvements in both generalization and robustness over state-of-the-art detectors and more calibrated results across 27 different generative models, including recent releases, like FLUX and Stable Diffusion 3.5. Our findings emphasize the importance of a careful dataset design, highlighting the need for further research on this topic. Code and data are publicly available at https://grip-unina.github.io/B-Free/.

Figures

Figures reproduced from arXiv: 2412.17671 by the authors.

Figure 1
Figure 1. We introduce a new training paradigm for AI-generated image detection. To avoid possible biases, we generate synthetic images [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Forensic detectors can exhibit opposite behaviors depending on their training dataset. The four plots show the prediction [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Overview of existing (a, b, c) and proposed (d) strategies for building an aligned training dataset. Some methods try to match synthetic images to the corresponding real images by using class-based generation (a) or text-to-image generation with real images’ descriptions (b). In (c) real images are fed to an autoencoder to generate a reconstructed fake with the same content. Differently from (c), in our approach a s… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Content augmentation process. Starting with a real image, we use its generated variants (first row) and their locally manipulated versions (last row), created by replacing the original background. When inpainting with a different category, we use a bounding box instead…
Figure 5
Figure 5. Figure 5: Power spectra computed by averaging (2000 images) [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Robustness analysis in terms of balanced Accuracy car [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Average performance in term of AUC and bAcc on four datasets: Synthbuster, GenImage, FakeInversion, SynthWildX. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Results of SoTA detectors on real and fake images that [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Examples of content augmented images from our train [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: SoTA performance evaluated in terms of AUC and balanced Accuracy on Midjourney, SDXL and DALL·E generators from [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Robustness analysis in terms of balanced Accuracy [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. V-FIND: Revealing the Intrinsic Forgery Knowledge Encoded in Video Forgery Detectors

    cs.CV 2026-08 conditional novelty 6.0 of 10

    A frozen video forgery detector can detect AI-generated videos using only 211 selected anchor neurons and a linear classifier.

  2. Continuously Evolving Deepfake Detection: An Architecture and Public-Benchmark Evaluation of a Dynamic Detection System

    cs.CV 2026-07 conditional novelty 6.0 of 10

    A continuously refreshed, incentive-driven deepfake detector beats static detectors on in-the-wild benchmarks and improves on post-export AI-generated media.

Reference graph

Works this paper leans on

58 extracted references · 48 canonical work pages · cited by 2 Pith papers

  1. [38]

    On the Effectiveness of Dataset Alignment for Fake Image Detection

    Anirudh Sundara Rajan, Utkarsh Ojha, Jedidiah Schloesser, and Yong Jae Lee. On the Effectiveness of Dataset Alignment for Fake Image Detection. arXiv preprint arXiv:2410.11835, 2024. 2, 3

    arXiv 2024

  2. [1]

    Parents and Children: Distinguishing Multimodal Deep- Fakes from Natural Images

    Roberto Amoroso, Davide Morelli, Marcella Cornia, Lorenzo Baraldi, Alberto Del Bimbo, and Rita Cucchiara. Parents and Children: Distinguishing Multimodal Deep- Fakes from Natural Images. ACM Trans. Multimedia Com- put. Commun. Appl., 2024. 3

    work page 2024

  3. [2]

    Synthbuster: Towards detection of diffu- sion model generated images

    Quentin Bammey. Synthbuster: Towards detection of diffu- sion model generated images. IEEE Open Journal of Signal Processing, 2023. 2, 3, 4, 11

    work page 2023

  4. [3]

    Contrasting Deepfakes Diffusion via Contrastive Learning and Global-Local Similarities

    Lorenzo Baraldi, Federico Cocchi, Marcella Cornia, Lorenzo Baraldi, Alessandro Nicolosi, and Rita Cucchiara. Contrasting Deepfakes Diffusion via Contrastive Learning and Global-Local Similarities. In ECCV, 2024. 2, 3, 4, 7, 10, 11

    work page 2024

  5. [4]

    Identi- fying and Mitigating the Security Risks of Generative AI

    Clark Barrett, Brad Boyd, Elie Bursztein, Nicholas Car- lini, Brad Chen, Jihye Choi, Amrita Roy Chowdhury, Mihai Christodorescu, Anupam Datta, Soheil Feizi, et al. Identi- fying and Mitigating the Security Risks of Generative AI. Foundations and Trends® in Privacy and Security , 6(1):1– 52, 2023. 1

    work page 2023

  6. [5]

    A possible pit- fall in the experimental analysis of tampering detection algo- rithms

    Giuseppe Cattaneo and Gianluca Roscigno. A possible pit- fall in the experimental analysis of tampering detection algo- rithms. In International Conference on Network-Based In- formation Systems, 2014. 2

    work page 2014

  7. [6]

    Real-Time Deepfake Detection in the Real-World

    Bar Cavia, Eliahu Horwitz, Tal Reiss, and Yedid Hoshen. Real-Time Deepfake Detection in the Real-World. arXiv preprint arXiv:2406.09398, 2024. 4, 7, 10

    arXiv 2024

  8. [7]

    FakeInversion: Learning to Detect Images from Un- seen Text-to-Image Models by Inverting Stable Diffusion

    George Cazenavette, Avneesh Sud, Thomas Leung, and Ben Usman. FakeInversion: Learning to Detect Images from Un- seen Text-to-Image Models by Inverting Stable Diffusion. In CVPR, pages 10759–10769, 2024. 2, 3, 4, 7, 11, 12

    work page 2024

Show all 58 references
  1. [8]

    AntifakePrompt: Prompt-Tuned Vision-Language Models are Fake Image Detectors

    You-Ming Chang, Chen Yeh, Wei-Chen Chiu, and Ning Yu. AntifakePrompt: Prompt-Tuned Vision-Language Models are Fake Image Detectors. arXiv preprint arXiv:2310.17419,

    arXiv

  2. [9]

    Intriguing properties of syn- thetic images: from generative adversarial networks to dif- fusion models

    Riccardo Corvi, Davide Cozzolino, Giovanni Poggi, Koki Nagano, and Luisa Verdoliva. Intriguing properties of syn- thetic images: from generative adversarial networks to dif- fusion models. In CVPR Workshops, pages 973–982, 2023. 3

    2023

  3. [10]

    On the de- tection of synthetic images generated by diffusion models

    Riccardo Corvi, Davide Cozzolino, Giada Zingarini, Gio- vanni Poggi, Koki Nagano, and Luisa Verdoliva. On the de- tection of synthetic images generated by diffusion models. In ICASSP, pages 1–5, 2023. 3, 4, 6, 7, 9

    2023

  4. [11]

    Raising the Bar of AI-generated Image Detection with CLIP

    Davide Cozzolino, Giovanni Poggi, Riccardo Corvi, Matthias Nießner, and Luisa Verdoliva. Raising the Bar of AI-generated Image Detection with CLIP. In CVPR Work- shops, pages 4356–4366, 2024. 2, 3, 4, 6, 7

    2024

  5. [12]

    RAISE: a raw images dataset for dig- ital image forensics

    Duc-Tien Dang-Nguyen, Cecilia Pasquini, Valentina Conot- ter, and Giulia Boato. RAISE: a raw images dataset for dig- ital image forensics. In ACM MMSys, page 219–224. Asso- ciation for Computing Machinery, 2015. 2

    2015

  6. [13]

    Vision Transformers Need Registers

    Timoth ´ee Darcet, Maxime Oquab, Julien Mairal, and Piotr Bojanowski. Vision Transformers Need Registers. In ICLR,

  7. [14]

    CASIA Image Tam- pering Detection Evaluation Database

    Jing Dong, Wei Wang, and Tieniu Tan. CASIA Image Tam- pering Detection Evaluation Database. In IEEE ChinaSIP,

  8. [15]

    Watch your up-convolution: CNN based Generative Deep Neural Networks are failing to reproduce spectral distributions

    Ricard Durall, Margret Keuper, and Janis Keuper. Watch your up-convolution: CNN based Generative Deep Neural Networks are failing to reproduce spectral distributions. In CVPR, pages 7890–7899, 2020. 3

    2020

  9. [16]

    Fourier spectrum discrepancies in deep network generated images

    Tarik Dzanic, Karan Shah, and Freddie Witherden. Fourier spectrum discrepancies in deep network generated images. In NeurIPS, pages 3022–3032, 2020. 3

    2020

  10. [17]

    Art and the science of generative AI

    Ziv Epstein, Aaron Hertzmann, et al. Art and the science of generative AI. Science, 380(6650):1110–1111, 2023. 1

    2023

  11. [18]

    Are GAN generated images easy to detect? A critical analysis of the state-of-the- art

    Diego Gragnaniello, Davide Cozzolino, Francesco Marra, Giovanni Poggi, and Luisa Verdoliva. Are GAN generated images easy to detect? A critical analysis of the state-of-the- art. In ICME, pages 1–6, 2021. 2

    2021

  12. [19]

    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. In ECCV Workshops,

  13. [20]

    FingerprintNet: Synthesized Fin- gerprints for Generated Image Detection

    Yonghyun Jeong, Doyeon Kim, Youngmin Ro, Pyounggeon Kim, and Jongwon Choi. FingerprintNet: Synthesized Fin- gerprints for Generated Image Detection. In ECCV, pages 76–94, 2022. 3

    2022

  14. [21]

    Evolution of Detection Performance throughout the Online Lifespan of Synthetic Images

    Dimitrios Karageorgiou, Quentin Bammey, Valentin Por- cellini, Bertrand Goupil, Denis Teyssou, and Symeon Pa- padopoulos. Evolution of Detection Performance throughout the Online Lifespan of Synthetic Images. In ECCV Work- shops, 2024. 7

    2024

  15. [22]

    Leveraging Representations from Intermediate Encoder-blocks for Syn- thetic Image Detection

    Christos Koutlis and Symeon Papadopoulos. Leveraging Representations from Intermediate Encoder-blocks for Syn- thetic Image Detection. In ECCV, pages 394–411, 2024. 2, 3, 6, 7, 10

    2024

  16. [23]

    Faster Than Lies: Real-time Deepfake Detection using Binary Neural Networks

    Romeo Lanzino, Federico Fontana, Anxhelo Diko, Marco Raoul Marini, and Luigi Cinque. Faster Than Lies: Real-time Deepfake Detection using Binary Neural Networks. In CVPR Workshops, pages 3771–3780, 2024. 3, 7, 10

    2024

  17. [24]

    Autoregressive Image Generation without Vec- tor Quantization

    Tianhong Li, Yonglong Tian, He Li, Mingyang Deng, and Kaiming He. Autoregressive Image Generation without Vec- tor Quantization. In NeurIPS, pages 56424–56445, 2024. 12

    2024

  18. [25]

    MaskSim: Detection of Syn- thetic Images by Masked Spectrum Similarity Analysis

    Yanhao Li, Quentin Bammey, Marina Gardella, Tina Nikoukhah, Jean-Michel Morel, Miguel Colom, and Rafael Grompone V on Gioi. MaskSim: Detection of Syn- thetic Images by Masked Spectrum Similarity Analysis. In CVPR, pages 3855–3865, 2024. 11

    2024

  19. [26]

    FakeBench: Prob- ing Explainable Fake Image Detection via Large Multimodal Models

    Yixuan Li, Xuelin Liu, Xiaoyang Wang, Bu Sung Lee, Shiqi Wang, Anderson Rocha, and Weisi Lin. FakeBench: Prob- ing Explainable Fake Image Detection via Large Multimodal Models. arXiv preprint arXiv:2404.13306, 2024. 11

    2024 arXiv

  20. [27]

    Detecting multimedia generated by large AI models: A survey

    Li Lin, Neeraj Gupta, Yue Zhang, Hainan Ren, Chun-Hao Liu, Feng Ding, Xin Wang, Xin Li, Luisa Verdoliva, and Shu Hu. Detecting multimedia generated by large AI models: A survey. arXiv preprint arXiv:2204.06125, 2024. 1

    2024 arXiv

  21. [28]

    Lawrence Zitnick

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C. Lawrence Zitnick. Microsoft COCO: Common objects in context. In ECCV, pages 740–755, 2014. 5

    2014

  22. [29]

    Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection

    Huan Liu, Zichang Tan, Chuangchuang Tan, Yunchao Wei, Jingdong Wang, and Yao Zhao. Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection. In CVPR, pages 10770–10780, 2024. 2, 3, 7, 10

    2024

  23. [30]

    A Decade’s Battle on Dataset Bias: Are We There Yet? In ICLR, 2025

    Zhuang Liu and Kaiming He. A Decade’s Battle on Dataset Bias: Are We There Yet? In ICLR, 2025. 2

    2025

  24. [31]

    When Synthetic Traces Hide Real Content: Analysis of Stable Dif- fusion Image Laundering

    Sara Mandelli, Paolo Bestagini, and Stefano Tubaro. When Synthetic Traces Hide Real Content: Analysis of Stable Dif- fusion Image Laundering. In WIFS, pages 1–6, 2024. 3

    2024

  25. [32]

    Do GANs Leave Artificial Fingerprints? In MIPR, pages 506–511, 2019

    Francesco Marra, Diego Gragnaniello, Luisa Verdoliva, and Giovanni Poggi. Do GANs Leave Artificial Fingerprints? In MIPR, pages 506–511, 2019. 2

    2019

  26. [33]

    Towards uni- versal fake image detectors that generalize across generative models

    Utkarsh Ojha, Yuheng Li, and Yong Jae Lee. Towards uni- versal fake image detectors that generalize across generative models. In CVPR, pages 24480–24489, 2023. 1, 2, 3, 4, 6, 7, 9

    2023

  27. [34]

    DINOv2: Learning Robust Visual Features without Supervi- sion

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. DINOv2: Learning Robust Visual Features without Supervi- sion. Transactions on Machine Learning Research Journal,

  28. [35]

    Obtaining Well Calibrated Probabilities Using Bayesian Binning

    Mahdi Pakdaman Naeini, Gregory Cooper, and Milos Hauskrecht. Obtaining Well Calibrated Probabilities Using Bayesian Binning. AAAI, 29(1), 2015. 4

    2015

  29. [36]

    Evaluating Predictive Uncertainty Challenge

    Joaquin Qui ˜nonero-Candela, Carl Edward Rasmussen, Fabian Sinz, Olivier Bousquet, and Bernhard Sch ¨olkopf. Evaluating Predictive Uncertainty Challenge. In Machine Learning Challenges Workshop, pages 1–27, 2006. 4

    2006

  30. [37]

    Learning Transferable Visual Models From Natural Language Super- vision

    Alec Radford, JongWook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning Transferable Visual Models From Natural Language Super- vision. In ICML, pages 8748–8763, 2021. 1, 6

    2021

  31. [39]

    Stable Diffu- sion

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. Stable Diffu- sion. https : / / github . com / Stability - AI / stablediffusion, 2022. 5

    2022

  32. [40]

    DE- FAKE: Detection and Attribution of Fake Images Gener- ated by Text-to-Image Generation Models

    Zeyang Sha, Zheng Li, Ning Yu, and Yang Zhang. DE- FAKE: Detection and Attribution of Fake Images Gener- ated by Text-to-Image Generation Models. InACM SIGSAC, pages 3418–3432, 2023. 3, 7, 10

    2023

  33. [41]

    Learning on Gradients: Generalized Arti- facts Representation for GAN-Generated Images Detection

    Chuangchuang Tan, Yao Zhao, Shikui Wei, Guanghua Gu, and Yunchao Wei. Learning on Gradients: Generalized Arti- facts Representation for GAN-Generated Images Detection. In CVPR, pages 12105–12114, 2023. 3, 7, 9

    2023

  34. [42]

    Rethinking the Up-Sampling Operations in CNN-based Generative Network for Generalizable Deepfake Detection

    Chuangchuang Tan, Huan Liu, Yao Zhao, Shikui Wei, Guanghua Gu, Ping Liu, and Yunchao Wei. Rethinking the Up-Sampling Operations in CNN-based Generative Network for Generalizable Deepfake Detection. In CVPR, 2024. 7, 10

    2024

  35. [43]

    C2P- CLIP: Injecting Category Common Prompt in CLIP to En- hance Generalization in Deepfake Detection

    Chuangchuang Tan, Renshuai Tao, Huan Liu, Guanghua Gu, Baoyuan Wu, Yao Zhao, and Yunchao Wei. C2P- CLIP: Injecting Category Common Prompt in CLIP to En- hance Generalization in Deepfake Detection. arXiv preprint arXiv:2408.09647, 2024. 3, 7, 10

    2024 arXiv

  36. [44]

    Synthetic Image Verification in the Era of Generative AI: What Works and What Isn’t There Yet

    Diangarti Tariang, Riccardo Corvi, Davide Cozzolino, Gio- vanni Poggi, Koki Nagano, and Luisa Verdoliva. Synthetic Image Verification in the Era of Generative AI: What Works and What Isn’t There Yet. IEEE Security & Privacy , 22: 37–49, 2024. 1, 4

    2024

  37. [45]

    Antonio Torralba and Alexei A. Efros. Unbiased look at dataset bias. In CVPR, 2011. 2

    2011

  38. [46]

    CNN-generated images are sur- prisingly easy to spot

    Sheng-Yu Wang, Oliver Wang, Richard Zhang, Andrew Owens, and Alexei A Efros. CNN-generated images are sur- prisingly easy to spot... for now. InCVPR, pages 8695–8704,

  39. [47]

    DIRE for diffusion-generated image detection

    Zhendong Wang, Jianmin Bao, Wengang Zhou, Weilun Wang, Hezhen Hu, Hong Chen, and Houqiang Li. DIRE for diffusion-generated image detection. In ICCV, pages 22445– 22455, 2023. 3, 7, 10

    2023

  40. [48]

    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. In ICLR, 2025. 7, 10 14

    2025

  41. [49]

    Diffusion Probabilistic Model Made Slim

    Xingyi Yang, Daquan Zhou, Jiashi Feng, and Xinchao Wang. Diffusion Probabilistic Model Made Slim. In CVPR, pages 22552–22562, 2023. 3

    2023

  42. [50]

    Attributing Fake Images to GANs: Learning and Analyzing GAN Finger- prints

    Ning Yu, Larry S Davis, and Mario Fritz. Attributing Fake Images to GANs: Learning and Analyzing GAN Finger- prints. In ICCV, pages 7556–7566, 2019. 2

    2019

  43. [51]

    Randomized Autoregressive Visual Generation

    Qihang Yu, Ju He, Xueqing Deng, Xiaohui Shen, and Liang- Chieh Chen. Randomized Autoregressive Visual Generation. arXiv preprint arXiv:2411.00776, 2024. 12

    2024 arXiv

  44. [52]

    CutMix: Regu- larization strategy to train strong classifiers with localizable features

    Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, and Youngjoon Yoo. CutMix: Regu- larization strategy to train strong classifiers with localizable features. In ICCV, pages 6023–6032, 2019. 6

    2019

  45. [53]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. In ICCV, pages 11975–11986, 2023. 6

    2023

  46. [54]

    Multimodal Image Synthesis and Editing: The Generative AI Era

    Fangneng Zhan, Yingchen Yu, Rongliang Wu, Jiahui Zhang, Shijian Lu, Lingjie Liu, Adam Kortylewski, Christian Theobalt, and Eric Xing. Multimodal Image Synthesis and Editing: The Generative AI Era. IEEE TPAMI, 45(12): 15098–15119, 2021. 1

    2021

  47. [55]

    Dauphin, and David Lopez-Paz

    Hongyi Zhang, Moustapha Cisse, Yann N. Dauphin, and David Lopez-Paz. mixup: Beyond Empirical Risk Mini- mization. In ICLR, 2018. 6

    2018

  48. [56]

    Detecting and Simulating Artifacts in GAN Fake Images

    Xu Zhang, Svebor Karaman, and Shih-Fu Chang. Detecting and Simulating Artifacts in GAN Fake Images. In WIFS,

  49. [57]

    GenImage: A Million-Scale Benchmark for Detecting AI-Generated Image

    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. NeurIPS, 36:77771–77782,

  50. [2020]

    3, 4, 5, 6, 7, 9, 10

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.