Pith. sign in

REVIEW 3 major objections 6 minor 42 references

GlobalForge claims that AI-generated-image detection becomes robust to real-world degradation when detectors are steered from fragile local artifacts to degradation-stable global structural coherence, and reports a 5.89% average balanced-ac

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 01:20 UTC pith:JJWEWXKW

load-bearing objection GlobalForge is a credible robustness-first detector with a new benchmark; the headline SOTA gap is a system-level comparison, but the internal same-backbone ablation mostly addresses the backbone confound, so the fix is reporting, not a redesign. the 3 major comments →

arxiv 2607.14684 v1 pith:JJWEWXKW submitted 2026-07-16 cs.CV

GlobalForge: Towards Robust AI-Generated Image Detection

Manni Cui , Ruiqi Liu , Dianyuan Zou , Ziheng Qin , Jingrui Xu , ZiAn Wang , Jianglan Wei , Han Zhou
show 3 more authors
Yu Liu Yan Wang Shu Wu
This is my paper
classification cs.CV
keywords AI-generated image detectiondeepfake detectiondegradation robustnessshortcut learningglobal structural reasoningLocal Information Bottleneckcontrastive learningbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that AI-generated-image detectors fail under real-world degradation because they rely on local generator artifacts that JPEG, blur, and resize destroy. To fix that, GlobalForge trains detectors to ignore local details and reason over long-range structural coherence: a Local Information Bottleneck suppresses high-frequency components, a Global Structural Reasoning module masks nearby tokens so each token must gather evidence from distant regions, and a contrastive loss keeps features stable under degradation. The paper backs the claim with a new benchmark, RealDeg-Bench, that chains one to five degradation operators, and reports that GlobalForge improves average balanced accuracy on 8 in-the-wild benchmark groups by 5.89% over the previous state of the art, reaching 85.81% on RealDeg-Bench. A sympathetic reader would care because it proposes a mechanism, not just data augmentation, for why detectors break and how to build them to survive social-media transmission.

Core claim

The paper's central claim is that AI-generated-image detectors fail under real-world degradation because they overfit to generator-specific local artifacts, and that actively suppressing local information while forcing long-range structural reasoning produces a cue that survives JPEG, blur, resize, and multi-step chains. In the authors' experiments, this design—trained with a degradation-aware contrastive loss—raises average balanced accuracy on eight in-the-wild benchmark groups to 85.93%, which they report as 5.89 points above the strongest prior detector, and reaches 85.81% on their new RealDeg-Bench. Ablations show each module (LIB, GSR, and the contrastive loss) contributes, and that in

What carries the argument

The framework's load-bearing mechanism is a pair of modules inserted after a frozen pretrained vision transformer. LIB (Local Information Bottleneck) forms a convex combination of deep features with a Gaussian-blurred copy, suppressing high-frequency local components in feature space rather than in pixel space. GSR (Global Structural Reasoning) masks self-attention within a Chebyshev distance of 3, so each token can only aggregate evidence from distant regions; with masking probability 1.0 during training, the model cannot fall back on adjacent-token correlations. A degradation-aware contrastive structural loss (L_DCS) pulls clean and compound-degraded views together in feature space, stabil

Load-bearing premise

The claim that the 5.89% gain comes from the proposed modules, rather than from the stronger newer backbone used in the best GlobalForge variant, is not fully isolated by an explicit backbone-controlled baseline; the only controlled comparison with the older backbone shows a smaller 3.19% gain over the same prior detector.

What would settle it

Train the strongest prior detector with the same newer backbone, fine-tuning recipe, training data, and schedule as GlobalForge and evaluate on the same eight in-the-wild groups; if it reaches or exceeds 85.93% average balanced accuracy, the proposed modules add little beyond a backbone upgrade.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the claim holds, detectors can be made robust without restoration or augmentation-only defenses; the cue itself is stable under degradation.
  • Accuracy on clean images need not be sacrificed: the authors report the best average degraded accuracy across six standard benchmarks and near-best clean accuracy.
  • The failure mode should become graceful rather than cliff-like: performance decays smoothly from one to five chained degradations.
  • The mechanism should also transfer across generators, because global structure is less tied to a specific generator's local signature.
  • RealDeg-Bench supplies a fixed 13-condition protocol (clean, seven single degradations, five compound chains) so future detectors can be compared on the same degradation axis.

Where Pith is reading between the lines

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

  • Editorial inference: the same local-suppression/global-reasoning recipe could be transferred to other forensic tasks, such as video deepfake detection, where compression chains tend to be deeper; the paper's appendix lists video extension as future work.
  • Editorial inference: if global structural coherence is the real cue, targeted attacks that distort long-range statistics while leaving local patches intact should be more effective against GlobalForge than local-patch attacks; the paper does not test this.
  • Editorial inference: the benchmark's with-replacement chain sampling makes repeated re-compression a first-class citizen; a natural community norm would be to report accuracy as a function of chain depth, not only endpoint accuracy.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper argues that existing AI-generated image detectors fail under real-world degradations because they rely on fragile local artifacts, and proposes GlobalForge, which suppresses local information via a Local Information Bottleneck (LIB), forces long-range evidence gathering via Global Structural Reasoning (GSR), and aligns clean/degraded representations with a degradation-aware contrastive loss (DCS). The authors introduce RealDeg-Bench, a compound-degradation benchmark, and report that GlobalForge improves average BAcc by 5.89% over DDA on 8 in-the-wild benchmark groups and reaches 85.81% average BAcc on RealDeg-Bench. Extensive experiments, ablations, and sensitivity analyses are included.

Significance. If the reported gains hold up, the work is a useful step in a practically important direction: robustness of AI-generated image detectors under realistic propagation. RealDeg-Bench is a sensible controlled benchmark that addresses a real gap, and the explicit release of code and benchmark details supports reproducibility. The paper also makes a conceptually interesting design choice — actively suppressing local shortcuts rather than defending them. The ablations in Table 3 and the alternative analysis in Table 4 provide support for the mechanism. The main weaknesses are attribution of the headline number to the proposed modules versus the backbone and training configuration, and the absence of variance estimates. The paper does not contain formal proofs, but its claims are empirical and falsifiable.

major comments (3)
  1. [§6.3, Table 2, Appendix D] The headline '+5.89% over DDA' uses Ours-d3 (DINOv3-L) against DDA's official released weights, which do not use DINOv3. Table 3 does provide a plain fine-tuned no-module baseline (80.21 in-the-wild) that is essentially tied with DDA (80.04), so I do not think the full gain is purely a backbone effect. However, the gap between Ours-d3 (85.93) and the DINOv2-L variant Ours-d2 (83.23) is 2.70 points, and Appendix D shows those variants also differ in patch size, epochs, learning rate, batch organization, and GPU count. Thus the incremental contribution of LIB/GSR/DCS relative to DDA's pipeline is not cleanly isolated. Please add a DINOv3-L-based DDA run or otherwise control for the backbone, and explicitly separate the 'method + backbone' gain from the 'modules only' gain. At minimum, the 5.89% claim should be qualified as GlobalForge-with-DINOv3-L versus DDA-with-DINOv2-L.
  2. [§6.2, Tables 1–3] All performance numbers are point estimates from a single run, with no error bars, confidence intervals, or multi-seed variation. Several load-bearing comparisons rest on differences of 1–2 BAcc points, e.g., Table 3's w/o-DCS versus Full (83.94 vs 85.93 in-the-wild; 82.61 vs 85.81 on RealDeg-Bench), and some Table 1 rows where Ours-d2 and Ours-d3 are close. Without repeated seeds or a paired statistical test, the component-wise necessity and some benchmark-level gaps are not established. Please add multiple seeds for the main variants and ablations, or report confidence intervals based on subset bootstrapping.
  3. [§3 and §4.2] The paper's central conceptual claim is that robustness comes from 'global structural coherence' rather than local artifacts. The architecture enforces this by masking local attention and suppressing local features, and the alternative analysis in Table 4 shows that a global-attention variant performs worse. However, there is no direct measurement that the learned decisions actually rely on long-range structure — for example, a perturbation study that compares sensitivity to distant versus adjacent regions, or a statistic over token-to-token attention distances. As written, 'global structural cue' is an interpretation of why the architecture succeeds, not an independently verified property. I recommend either adding such an experiment or softening the 'root cause' phrasing in the abstract and introduction.
minor comments (6)
  1. [Abstract, §1] The 5.89% headline should state explicitly that it compares the DINOv3-L variant against DDA's official weights; otherwise readers may infer a controlled module-only comparison.
  2. [Table 2] The table lists 17 subset scores across 8 groups, but the group structure is not fully legible in the printed header. A subtable or explicit group-to-subset mapping would improve reproducibility.
  3. [Table 4, Appendix D] The notation 'RealDeg-1Step' is used in the caption but formally defined only in Appendix D. Define it in the caption or use 'RealDeg-Bench compound N=1' for clarity.
  4. [Eq. (6)] The denominator sums over j without specifying whether the index range includes the positive pair in both terms. State explicitly that j runs over the full batch including the positive; otherwise the loss expression is ambiguous.
  5. [Appendix D, Table 7] The training differences between ours-d2 and ours-d3 (epochs 10/8 vs 8/6, LR 4e-4 vs 5e-5, batch/GPU 128/16 vs 128/4, patch sizes 14 vs 16) should be highlighted in the main text, since readers will otherwise interpret the Ours-d2/Ours-d3 gap as purely a backbone effect.
  6. [Figure 5] The x-axis for JPEG quality factor is drawn with high quality on the left and low quality on the right, which is opposite to the horizontal direction used in most other panels. A consistent monotonic direction would avoid misreading.

Circularity Check

1 steps flagged

No equation-level circularity; minor mechanism-claim tautology; headline comparison has a backbone confound that is a correctness risk, not circularity.

specific steps
  1. other [Section 4.2 (GSR definition) and Appendix E ('Stability of spatial focus under degradation')]
    "This design forces each token’s representation update to come exclusively from distant regions, so the model can only discriminate by integrating cross-region structural information. ... We attribute this to the GSR design (Section 4.2), which forces every token to aggregate evidence from distant patches outside its local window."

    The explanatory claim that GlobalForge's attention is degradation-stable because of GSR is a restatement of GSR's construction: GSR is defined as local-window masking, so showing that attention is non-local is implied by the mask itself. This makes the mechanistic narrative partially self-validating rather than an independent diagnostic test. However, this is not a derivation-level circle: the headline BAcc improvements are not forced by the equations, and the ablation in Table 3 (baseline 80.21 vs full 85.93 in-the-wild) is an empirical result. The main quantitative caveat is a backbone confound in the Ours-d3 vs DDA comparison, which is a correctness risk, not circularity.

full rationale

The paper's training objective (Eqs. 5-6) is not defined in terms of the evaluation benchmarks, and no parameter is fitted to RealDeg-Bench or the in-the-wild groups. The LIB/GSR modules are architectural interventions whose effects are measured empirically, and the losses do not reduce to the reported metrics by construction. The only self-referential aspect is the paper's mechanistic interpretation: GSR explicitly masks local attention and LDCS explicitly aligns clean/degraded views, so observing that GlobalForge relies on global, degradation-stable features is partly implied by the construction. This weakens the diagnostic narrative but does not invalidate the empirical robustness gains, which are supported by same-backbone ablations. The largest threat to the headline '5.89% over previous SOTA' is that Ours-d3 uses DINOv3-L while DDA is evaluated with official weights, and the same-backbone Ours-d2 variant gains only 3.19%—a comparison confound, not a circularity. Overall, no significant derivation-level circularity is present.

Axiom & Free-Parameter Ledger

7 free parameters · 4 axioms · 0 invented entities

The paper is empirical and does not introduce physical entities. It relies on several hand-chosen hyperparameters (masking window, loss weight, augmentation strengths) and on the domain assumption that global structure is more robust than local artifacts. The DCS augmentation ranges deliberately mimic the test-time degradations, which is a form of training-test overlap worth noting.

free parameters (7)
  • LIB fusion gate β (learned scalar) = learned; α=σ(β)
    Controls the degree of feature smoothing; trained jointly, effectively fitted to the data.
  • GSR masking window size w_gsr = 3 (default)
    Chosen by hand; sensitivity analyzed in Figure 6.
  • GSR masking probability p_mask = 1.0
    Chosen by hand; found to be optimal in sensitivity analysis.
  • DCS loss weight λ_dcs = 0.01
    Chosen by hand; sensitivity analyzed in Figure 6.
  • InfoNCE temperature κ = 0.07
    Standard value chosen by hand.
  • DCS online augmentation ranges (JPEG Q 20-80, blur σ 0.5-1.5, color jitter 0.4/hue 0.06) = Q∈(20,80), σ∈(0.5,1.5), jitter 0.4
    Hand-chosen to approximate test degradations; overlap with RealDeg-Bench operators may inflate robustness gains.
  • LIB Gaussian kernel σ=k/3 = 1.0 (k=3)
    Chosen to control the smoothing receptive field.
axioms (4)
  • domain assumption Global structural cues are more stable under degradation and generator shift than local artifacts.
    Central hypothesis, supported by experiments but not proven; if false, the method's mechanism is invalid.
  • domain assumption The seven degradation operators and random compound chains in RealDeg-Bench approximate real-world propagation sufficiently to diagnose robustness.
    Benchmark validity; real-world channels are more diverse, as acknowledged in Limitations.
  • domain assumption The DDA-aligned training set provides a fair common training ground for comparing GlobalForge with the six retrained baselines.
    Training-set composition is not fully described; comparison fairness relies on this.
  • domain assumption DINOv2/DINOv3 pretrained features are suitable backbones for AIGI detection.
    Backbone choice is a design decision; its effect on the headline gain is not isolated.

pith-pipeline@v1.3.0-alltime-deepseek · 235 in / 14788 out tokens · 177514 ms · 2026-08-02T01:20:53.726597+00:00 · methodology

0 comments
read the original abstract

AI-generated image (AIGI) detectors achieve strong accuracy on clean benchmarks, but their performance drops sharply after images are propagated through real-world channels. We trace this fragility to what these detectors actually learn: they overfit to local artifacts left by generators in small spatial neighborhoods, which are easily destroyed by common propagation degradations such as JPEG compression and blur. Instead, we shift the discriminative cue from fragile local artifacts to more robust global structure. Building on this, we propose GlobalForge, a framework with two complementary modules. The Local Information Bottleneck (LIB) suppresses local components to block shortcut learning, while the Global Structural Reasoning (GSR) module forces every token to gather evidence from distant regions. Both modules are trained jointly under a contrastive structural loss based on degradation that keeps the resulting features stable under degradation. To support fine-grained robustness evaluation, we further introduce RealDeg-Bench, covering 7 common degradation operators and multi-step compound chains. GlobalForge improves average BAcc on 8 in-the-wild benchmark groups by $\mathbf{5.89\%}$ over the previous state-of-the-art, and is clearly ahead of representative baselines on RealDeg-Bench under both single and compound degradations. Code is available at https://anonymous.4open.science/r/GlobalForge-BE0F/.

Figures

Figures reproduced from arXiv: 2607.14684 by Dianyuan Zou, Han Zhou, Jianglan Wei, Jingrui Xu, Manni Cui, Ruiqi Liu, Shu Wu, Yan Wang, Yu Liu, Zian Wang, Ziheng Qin.

Figure 1
Figure 1. Figure 1: Under clean conditions, existing detectors and GlobalForge both correctly identify AI [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Attention collapse under image degradation. Grad-CAM visualizations for DDA, UnivFD, and NPR under different degra￾dations. For each method, the top row shows the inputs and the bottom row shows the corre￾sponding attention maps; within each method, the left column is the clean image and the right column is its degraded counterpart. The three methods are evaluated under resizing, JPEG compression, and blur… view at source ↗
Figure 4
Figure 4. Figure 4: Overall architecture of GlobalForge. trained on them struggle on unseen generators. They behave less like useful cues and more like shortcuts [28, 29] that crowd out more reliable evidence. Common practice in the field already points in this direction. Existing AIGI detectors almost uni￾versally adopt center crop rather than resize at inference time [8, 9, 15], because resize causes a substantial drop in d… view at source ↗
Figure 5
Figure 5. Figure 5: Fine-grained strength performance curves on single degradations on [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Sensitivity analysis of our method on RealDeg-Bench under the single degradation setting, [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Example samples from RealDeg-Bench. Rows alternate between clean images from our dataset (odd rows) and their corresponding degraded versions (even rows), illustrating the diversity of generators, content, and degradation types covered by the benchmark. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative success and failure cases of GlobalForge. The figure uses a 2×2 block layout (correct vs. wrong × real vs. fake). Each block contains five columns; the input image is on top and the Grad-CAM heatmap (annotated with the predicted fake probability p) is on the bottom. The four blocks are: correctly classified real images (top-left, p ≤ 0.12), correctly classified fake images (top-right, p≥0.90), … view at source ↗
Figure 9
Figure 9. Figure 9: Attention stability under degradation: GlobalForge versus six representative baselines. Each pair of consecutive rows is one image, shown clean (top) and degraded (bottom). Columns from left to right are: the input image, GlobalForge (ours), and six competing detectors (DDA, AIDE, NPR, B-Free, UnivFD, C2P-CLIP). For each method, the two heatmaps in a pair use the same normalization range, so a visible chan… view at source ↗

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

42 extracted references · 9 linked inside Pith

  1. [1]

    The creation and detection of deepfakes: A survey.ACM computing surveys (CSUR), 54(1):1–41, 2021

    Yisroel Mirsky and Wenke Lee. The creation and detection of deepfakes: A survey.ACM computing surveys (CSUR), 54(1):1–41, 2021

    2021

  2. [2]

    Deepfakes and beyond: A survey of face manipulation and fake detection.Information fusion, 64:131–148, 2020

    Ruben Tolosana, Ruben Vera-Rodriguez, Julian Fierrez, Aythami Morales, and Javier Ortega- Garcia. Deepfakes and beyond: A survey of face manipulation and fake detection.Information fusion, 64:131–148, 2020

    2020

  3. [3]

    On the detection of synthetic images generated by diffusion models

    Riccardo Corvi, Davide Cozzolino, Giada Zingarini, Giovanni Poggi, Koki Nagano, and Luisa Verdoliva. On the detection of synthetic images generated by diffusion models. InICASSP 2023- 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE, 2023

    2023

  4. [4]

    Raising the Bar of AI-generated Image Detection with CLIP

    Davide Cozzolino, Giovanni Poggi, Riccardo Corvi, Matthias Niesner, and Luisa Verdoliva. Raising the Bar of AI-generated Image Detection with CLIP . In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 4356–4366, Los Alami- tos, CA, USA, June 2024. IEEE Computer Society. doi: 10.1109/CVPRW63382.2024.00439. URLhttp...

    arXiv 2024

  5. [5]

    Anisha Pal, Julia Kruk, Mansi Phute, Manognya Bhattaram, Diyi Yang, Duen Horng Chau, and Judy Hoffman. Semi-truths: A large-scale dataset of ai-augmented images for evaluating robustness of ai-generated image detectors.Advances in Neural Information Processing Systems, 37:118025–118051, 2024

    2024

  6. [6]

    A sanity check for ai-generated image detection.arXiv preprint arXiv:2406.19435, 2024

    Shilin Yan, Ouxiang Li, Jiayin Cai, Yanbin Hao, Xiaolong Jiang, Yao Hu, and Weidi Xie. A sanity check for ai-generated image detection.arXiv preprint arXiv:2406.19435, 2024

    Pith/arXiv arXiv 2024

  7. [7]

    Df-udetector: An effective method towards robust deepfake detection via feature restoration.Neural Networks, 160:216–226, 2023

    Jianpeng Ke and Lina Wang. Df-udetector: An effective method towards robust deepfake detection via feature restoration.Neural Networks, 160:216–226, 2023

    2023

  8. [8]

    Cnn- generated images are surprisingly easy to spot

    Sheng-Yu Wang, Oliver Wang, Richard Zhang, Andrew Owens, and Alexei A Efros. Cnn- generated images are surprisingly easy to spot... for now. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8695–8704, 2020

    2020

  9. [9]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 24480–24489, 2023

    2023

  10. [10]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10770–10780, 2024

    2024

  11. [11]

    C2p-clip: Injecting category common prompt in clip to enhance 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 enhance generalization in deepfake detection. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 7184–7192, 2025

    2025

  12. [12]

    Dual data alignment makes ai-generated image detector easier generalizable.arXiv preprint arXiv:2505.14359, 2025

    Ruoxin Chen, Junwei Xi, Zhiyuan Yan, Ke-Yue Zhang, Shuang Wu, Jingyi Xie, Xu Chen, Lei Xu, Isabel Guan, Taiping Yao, et al. Dual data alignment makes ai-generated image detector easier generalizable.arXiv preprint arXiv:2505.14359, 2025

    arXiv 2025

  13. [13]

    Scaling up ai-generated image detection via generator-aware prototypes.arXiv preprint arXiv:2512.12982, 2025

    Ziheng Qin, Yuheng Ji, Renshuai Tao, Yuxuan Tian, Yuyang Liu, Yipu Wang, and Xiaolong Zheng. Scaling up ai-generated image detection via generator-aware prototypes.arXiv preprint arXiv:2512.12982, 2025

    Pith/arXiv arXiv 2025

  14. [14]

    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, pages 3247–3258. PMLR, 2020. 10

    2020

  15. [15]

    Rethink- ing the up-sampling operations in cnn-based generative network for generalizable deepfake detection

    Chuangchuang Tan, Yao Zhao, Shikui Wei, Guanghua Gu, Ping Liu, and Yunchao Wei. Rethink- ing the up-sampling operations in cnn-based generative network for generalizable deepfake detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 28130–28139, 2024

    2024

  16. [16]

    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 Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 1, pages 2405–2414, 2025

    2025

  17. [17]

    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. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 22445–22455, 2023

    2023

  18. [18]

    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. InForty-first International Conference on Machine Learning, 2024

    2024

  19. [19]

    Generalizable synthetic image detection via language-guided contrastive learning.IEEE Transactions on Artificial Intelligence, 2025

    Haiwei Wu, Jiantao Zhou, and Shile Zhang. Generalizable synthetic image detection via language-guided contrastive learning.IEEE Transactions on Artificial Intelligence, 2025

    2025

  20. [20]

    Antifakeprompt: Prompt-tuned vision-language models are fake image detectors.arXiv preprint arXiv:2310.17419, 2023

    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, 2023

    Pith/arXiv arXiv 2023

  21. [21]

    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 Proceedings of the Computer Vision and Pattern Recognition Conference, pages 18685–18694, 2025

    2025

  22. [22]

    Texturecrop: Enhanc- ing synthetic image detection through texture-based cropping

    Despina Konstantinidou, Christos Koutlis, and Symeon Papadopoulos. Texturecrop: Enhanc- ing synthetic image detection through texture-based cropping. InProceedings of the Winter Conference on Applications of Computer Vision, pages 1459–1468, 2025

    2025

  23. [23]

    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. InProceedings of the Com- puter Vision and Pattern Recognition Conference, pages 18706–18717, 2025

    2025

  24. [24]

    Genimage: A million-scale benchmark for detecting ai-generated image.Advances in neural information processing systems, 36:77771–77782, 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, 36:77771–77782, 2023

    2023

  25. [25]

    Real-time deepfake detection in the real-world.arXiv preprint arXiv:2406.09398, 2024

    Bar Cavia, Eliahu Horwitz, Tal Reiss, and Yedid Hoshen. Real-time deepfake detection in the real-world.arXiv preprint arXiv:2406.09398, 2024

    Pith/arXiv arXiv 2024

  26. [26]

    Beyond artifacts: Real-centric envelope modeling for reliable ai-generated image detection.arXiv preprint arXiv:2512.20937, 2025

    Ruiqi Liu, Yi Han, Zhengbo Zhang, Liwei Yao, Zhiyuan Yan, Jialiang Shen, ZhiJin Chen, Boyi Sun, Lubin Weng, Jing Dong, et al. Beyond artifacts: Real-centric envelope modeling for reliable ai-generated image detection.arXiv preprint arXiv:2512.20937, 2025

    arXiv 2025

  27. [27]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 973–982, 2023

    2023

  28. [28]

    Imagenet-trained cnns are biased towards texture; increasing shape bias improves accuracy and robustness

    Robert Geirhos, Patricia Rubisch, Claudio Michaelis, Matthias Bethge, Felix A Wichmann, and Wieland Brendel. Imagenet-trained cnns are biased towards texture; increasing shape bias improves accuracy and robustness. InInternational conference on learning representations, 2018

    2018

  29. [29]

    Shortcut learning in deep neural networks.Nature Machine Intelligence, 2(11):665–673, 2020

    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, 2(11):665–673, 2020. 11

    2020

  30. [30]

    Representation learning with contrastive predictive coding.arXiv preprint arXiv:1807.03748, 2018

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding.arXiv preprint arXiv:1807.03748, 2018

    Pith/arXiv arXiv 2018

  31. [31]

    Easier painting than thinking: Can text-to-image models set the stage, but not direct the play?arXiv preprint arXiv:2509.03516, 2025

    Ouxiang Li, Yuan Wang, Xinting Hu, Huijuan Huang, Rui Chen, Jiarong Ou, Xin Tao, Pengfei Wan, Xiaojuan Qi, and Fuli Feng. Easier painting than thinking: Can text-to-image models set the stage, but not direct the play?arXiv preprint arXiv:2509.03516, 2025

    arXiv 2025

  32. [32]

    Is artificial intelligence generated image detection a solved problem?arXiv preprint arXiv:2505.12335, 2025

    Ziqiang Li, Jiazhen Yan, Ziwen He, Kai Zeng, Weiwei Jiang, Lizhi Xiong, and Zhangjie Fu. Is artificial intelligence generated image detection a solved problem?arXiv preprint arXiv:2505.12335, 2025

    arXiv 2025

  33. [33]

    Co-spy: Combining semantic and pixel features to detect synthetic images by ai

    Siyuan Cheng, Lingjuan Lyu, Zhenting Wang, Xiangyu Zhang, and Vikash Sehwag. Co-spy: Combining semantic and pixel features to detect synthetic images by ai. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 13455–13465, 2025

    2025

  34. [34]

    Bridging the gap between ideal and real-world evaluation: Benchmarking ai-generated image detection in challenging scenarios

    Chunxiao Li, Xiaoxiao Wang, Meiling Li, Boming Miao, Peng Sun, Yunjian Zhang, Xiangyang Ji, and Yao Zhu. Bridging the gap between ideal and real-world evaluation: Benchmarking ai-generated image detection in challenging scenarios. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 20379–20389, 2025

    2025

  35. [35]

    Effort: Efficient orthogonal modeling for generalizable ai-generated image detection.arXiv preprint arXiv:2411.15633, 2(6):7, 2024

    Zhiyuan Yan, Jiangming Wang, Zhendong Wang, Peng Jin, Ke-Yue Zhang, Shen Chen, Taiping Yao, Shouhong Ding, Baoyuan Wu, and Li Yuan. Effort: Efficient orthogonal modeling for generalizable ai-generated image detection.arXiv preprint arXiv:2411.15633, 2(6):7, 2024

    Pith/arXiv arXiv 2024

  36. [36]

    Aligned datasets improve detection of latent diffusion-generated images.arXiv preprint arXiv:2410.11835, 2024

    Anirudh Sundara Rajan, Utkarsh Ojha, Jedidiah Schloesser, and Yong Jae Lee. Aligned datasets improve detection of latent diffusion-generated images.arXiv preprint arXiv:2410.11835, 2024

    Pith/arXiv arXiv 2024

  37. [37]

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herve Jegou, Julien Mairal, Patrick L...

    2024

  38. [38]

    Oriane Siméoni, Huy V V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Michaël Ramamonjisoa, et al. Dinov3. arXiv preprint arXiv:2508.10104, 2025

    Pith/arXiv arXiv 2025

  39. [39]

    LoRA: Low-rank adaptation of large language models

    Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations, 2022. URL https://openreview. net/forum?id=nZeVKeeFYf9

    2022

  40. [40]

    Patchcraft: Exploring texture patch for efficient ai-generated image detection.arXiv preprint arXiv:2311.12397, 2023

    Nan Zhong, Yiran Xu, Sheng Li, Zhenxing Qian, and Xinpeng Zhang. Patchcraft: Exploring texture patch for efficient ai-generated image detection.arXiv preprint arXiv:2311.12397, 2023

    Pith/arXiv arXiv 2023

  41. [41]

    Synthbuster: Towards detection of diffusion model generated images.IEEE Open Journal of Signal Processing, 5:1–9, 2024

    Quentin Bammey. Synthbuster: Towards detection of diffusion model generated images.IEEE Open Journal of Signal Processing, 5:1–9, 2024. doi: 10.1109/OJSP.2023.3337714

    arXiv 2024

  42. [42]

    pretrained vision backbone + lightweight structural modules + joint training objectives

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. InEuropean conference on computer vision, pages 740–755. Springer, 2014. 12 A RealDeg-Bench: Construction Details Section 5 of the main text describes the design goal and degradation operato...

    arXiv 2014