Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Dual-branch detector spots AI images by fusing semantics with noise

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 · glm-5.2

2026-07-08 08:28 UTC pith:VNQ54A5X

load-bearing objection Solid empirical SOTA for synthetic image detection, but ablation can't separate architecture gains from training data gains the 3 major comments →

arxiv 2607.06354 v1 pith:VNQ54A5X submitted 2026-07-07 cs.CV cs.CR

Generalized Synthetic Image Detection with Enhanced RGB-Noise Representation Learning

Zhen Li , Gang Cao , Tian Zhang , Lifang Yu , Shaowei Weng This is my paper
classification cs.CV cs.CR
keywords synthetic image detectionFiLM modulationcontrastive learningBayar convolutionimage forensicsAI-generated contentcross-model generalizationhard sample mining
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.

RNSIDNet detects AI-generated images by running two parallel analyses—one on the image's visual content (RGB semantics via a frozen CLIP encoder) and one on its high-frequency noise residuals (via Bayar convolution)—then fusing them through a mechanism called FiLM, where the RGB features dynamically scale, shift, and gate the noise features. The core claim is that this conditional, dynamic fusion is superior to naive concatenation or addition because it lets semantic context guide which noise artifacts matter, preventing the two modalities from interfering. A second key component is a Hard Sample-aware Contrastive Learning (HSCL) strategy that, unlike standard contrastive losses, explicitly up-weights the top 10% most confusable cross-class samples, forcing the feature space to widen its decision boundary precisely where real and synthetic images are hardest to distinguish. The authors also construct a training dataset (AMSID) of ~235,000 pixel-aligned real/synthetic image pairs from diverse generators, arguing that content alignment forces the network to learn generator-intrinsic artifacts rather than dataset-specific semantic biases.

Core claim

The central mechanism is the FiLM-based dynamic modulation: RGB features extracted by a CLIP encoder are passed through an MLP that produces scaling, shift, and gating parameters, which are then applied to the noise-feature vector. This means the noise branch is not simply concatenated with the RGB branch but is adaptively recalibrated by it. Combined with the HSCL strategy that concentrates optimization pressure on the hardest negative samples, the architecture produces a feature space where real, GAN-generated, and diffusion-generated images form distinct clusters. The authors demonstrate this across eight benchmark datasets covering multiple generator families, achieving an average AUC of

What carries the argument

FiLM modulation (Feature-wise Linear Modulation): RGB features generate affine parameters (scale γ, shift β, gate g) that dynamically transform noise features; HSCL (Hard Sample-aware Contrastive Learning): a contrastive loss variant that selects the top-K most similar cross-class samples and applies an extra penalty weight α to their exponentiated similarities; Bayar convolution: a constrained kernel with fixed center weight -1 and normalized non-center weights summing to 1, guaranteeing a zero-sum high-pass filter that suppresses low-frequency content and amplifies generative artifacts

Load-bearing premise

The paper assumes that the FiLM modulation and HSCL strategy are the primary causes of the performance gains, but the ablation study removes modules without isolating the effect of the new AMSID training dataset composition, which could independently account for much of the improvement.

What would settle it

Train RNSIDNet's architecture (FiLM + HSCL + Bayar conv) on a single-generator unaligned dataset (e.g., ProGAN-only without pixel alignment) and compare against a simple concatenation-fusion baseline trained on the AMSID dataset; if the latter generalizes better, the dataset composition—not the architectural innovations—is the primary driver.

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

If this is right

  • If dynamic cross-modal modulation (FiLM) is the key driver, then any forensic task combining heterogeneous feature types—e.g., audio-visual deepfake detection or text-image consistency checking—could benefit from replacing concatenation with conditional modulation.
  • The HSCL strategy's success suggests that hard-sample mining is more impactful than uniform contrastive objectives for boundary-critical classification tasks where the decision margin in ambiguous regions matters more than global separation.
  • The pixel-aligned training data design (AMSID) implies that the field may be overestimating model capability when training on unaligned data: models may be learning content shortcuts rather than generator fingerprints, and aligned data is a necessary control.
  • RNSIDNet's robustness to JPEG compression but vulnerability to Gaussian blur indicates that diffusion and GAN artifacts are partially encoded in mid-frequency bands that survive compression but are destroyed by low-pass filtering, narrowing the search space for universal forensic traces.

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 / 5 minor

Summary. The paper proposes RNSIDNet, a dual-branch architecture for synthetic image detection. The RGB branch uses a frozen CLIP-ViT encoder with a Balanced Attention Module (BAM), while the noise branch uses Bayar convolutions. These branches are fused via a Feature-wise Linear Modulation (FiLM) module, where RGB features condition noise features. The model is trained with a joint loss combining Binary Cross-Entropy and a proposed Hard Sample-aware Contrastive Learning (HSCL) strategy that re-weights hard negative samples. The authors also introduce AMSID, a pixel-aligned multi-source training dataset. Evaluations are conducted across eight public benchmarks, demonstrating strong cross-model generalization and robustness against degradations like JPEG compression and blur.

Significance. The paper addresses a timely and important problem in image forensics. The combination of FiLM modulation for RGB-noise fusion and hard-sample-aware contrastive learning is a reasonable architectural contribution. The release of the AMSID dataset and the promise of publicly available code are notable strengths that enhance reproducibility. The comprehensive evaluation across eight diverse benchmarks, including robustness analysis and computational complexity comparisons, provides substantial empirical backing for the claims.

major comments (3)
  1. Section 4.3, Table 8: The ablation study is conducted exclusively on the Synthbuster dataset. While it demonstrates the contributions of BAM, FiLM, and HSCL on this single benchmark, it does not establish whether these module-level gains hold across the diverse distribution of the other seven test sets. Evaluating the ablations on at least one or two additional benchmarks (e.g., AIGCDetectionBenchmark, DDA-COCO) would significantly strengthen the claim that these architectural components drive the overall generalization improvement.
  2. Section 4.1.1, Table 1, and Section 4.6, Table 10: The central claim attributes the SOTA performance to the combination of FiLM, HSCL, and the dual-branch design. However, the ablation study (Table 8) removes modules but does not isolate the effect of the new AMSID training data composition from the architectural contributions. Table 10 compares ProGAN-only vs. multi-source training, confirming data source matters, but no experiment trains a simple baseline architecture (e.g., CLIP + noise branch with naive concatenation + BCE, no FiLM, no HSCL) on the same AMSID dataset. Without this control, the attribution of the ~13% average ACC gap over baselines (Table 3) to the proposed modules rather than to the new aligned multi-source training data remains unverified. A control experiment isolating the data contribution from the architecture contribution is needed to support the central claim.
  3. Section 4.1.3: The model is trained for only 1 epoch using the Adam optimizer. This is an unusual protocol and is not justified. While the CLIP backbone is frozen, 1 epoch may be insufficient for the trainable components (BAM, noise branch, FiLM, HSCL) to fully converge, raising questions about result stability. The authors should either justify this choice empirically (e.g., showing convergence curves) or report results across multiple random seeds to demonstrate that the reported performance is stable and not an artifact of a single short training run.
minor comments (5)
  1. Section 3.1.4, Eq. (7): The MLP generates a scaling coefficient γ, a shift coefficient β, and a gating vector g. However, in Section 3.1.2, Eq. (4), γ is already used to denote a learnable gating parameter for the BAM module. Reusing γ for different quantities in closely related sections is confusing and should be clarified with distinct notation.
  2. Section 4.2: The text states 'RNSIDNet maintains robust discrimination (80.94% ACC, 98.00% AUC)' on the DDA-COCO dataset. However, Table 7 reports the average ACC as 85.18% and average AUC as 98.22%. The text should be corrected to match the table, or the discrepancy should be clarified.
  3. Table 3: The formatting of the best and second-best results is inconsistent. For instance, on GenImage, RINE's ACC (95.15) is higher than RNSIDNet's (91.70), but RNSIDNet's value is bolded. The bolding convention should be applied consistently and accurately across all columns.
  4. Section 3.1.5, Eq. (11): The cosine similarity formula is written as s_{i,j} = (z_i · z_j^T) / τ. Since z_i and z_j are already L2-normalized vectors in R^d, the dot product z_i · z_j^T is a scalar. The notation is slightly awkward; writing z_i^T z_j would be more standard.
  5. Section 4.1.1: The data augmentation pipeline mentions 'randomly alternate between OpenCV and PIL encoders' for JPEG compression. It would be helpful to briefly note why both encoders are used (e.g., to simulate different compression artifacts) for readers unfamiliar with this detail.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful and constructive review. The comments are well-taken, and we address each below.

read point-by-point responses

  1. Referee: Section 4.3, Table 8: The ablation study is conducted exclusively on the Synthbuster dataset. Evaluating the ablations on at least one or two additional benchmarks would significantly strengthen the claim that these architectural components drive the overall generalization improvement.

    Authors: We agree that ablation on a single benchmark does not fully establish that the module-level gains generalize across diverse distributions. We will extend the ablation study to include AIGCDetectionBenchmark and DDA-COCO as additional evaluation sets. These two datasets are complementary to Synthbuster: AIGCDetectionBenchmark covers 16 generators spanning both DMs and GANs, while DDA-COCO contains pixel-aligned fakes that are among the most challenging in our test suite. Reporting the leave-one-out and replacement experiments on these additional benchmarks will allow readers to verify that the contributions of BAM, FiLM, and HSCL are not artifacts of a single test distribution. revision: yes

  2. Referee: The ablation study removes modules but does not isolate the effect of the new AMSID training data composition from the architectural contributions. No experiment trains a simple baseline architecture (e.g., CLIP + noise branch with naive concatenation + BCE, no FiLM, no HSCL) on the same AMSID dataset. A control experiment isolating the data contribution from the architecture contribution is needed.

    Authors: This is a fair and important point. Our current experiments do not fully disentangle the contribution of the AMSID training data from that of the proposed architectural components. We acknowledge this gap and will add a control experiment in which a stripped-down baseline—CLIP RGB branch + Bayar-conv noise branch with naive concatenation fusion and BCE-only loss, no BAM, no FiLM, no HSCL—is trained on the same AMSID dataset under identical settings. This will be evaluated on the same eight benchmarks. We expect that the aligned multi-source data will improve the baseline's generalization relative to ProGAN-only training (as Table 10 already suggests), but that the full RNSIDNet architecture will still show a meaningful margin, particularly on the harder benchmarks (Synthbuster, DDA-COCO, Chameleon). Including this control will allow us to make a properly qualified claim about the relative contributions of data and architecture. revision: yes

  3. Referee: The model is trained for only 1 epoch using the Adam optimizer. This is an unusual protocol and is not justified. 1 epoch may be insufficient for the trainable components to fully converge. The authors should either justify this choice empirically or report results across multiple random seeds.

    Authors: We agree that the 1-epoch protocol requires justification beyond what is currently in the manuscript. The rationale is as follows: the CLIP ViT-L/14 backbone (427M parameters) is frozen, leaving only 13.58M trainable parameters (BAM, noise branch, FiLM MLP, classification head). With approximately 200,000 training images at batch size 64, one epoch corresponds to roughly 3,125 iterations, which is comparable in iteration count to multi-epoch training on smaller datasets used by prior work (e.g., ProGAN-only training with ~100K images for 5–10 epochs). Additionally, the paired alternating sampling and aggressive data augmentation (JPEG, blur, geometric transforms) provide strong per-iteration signal diversity. However, we recognize this argument is currently implicit and should be made explicit in the paper. We will add (1) a convergence curve showing training loss and validation ACC on a held-out subset over iterations, demonstrating that the model reaches a stable plateau within one epoch, and (2) results across three random seeds on at least two benchmarks (Synthbuster and AIGCDetectionBenchmark) to confirm that the reported performance is not an artifact of a single short run. If the convergence curves reveal that the model has not fully converged, we will adjust the training protocol accordingly and re-report all results. revision: yes

Circularity Check

0 steps flagged

No circularity found: the derivation chain is self-contained, uses externally cited techniques, and evaluates on independent benchmarks.

full rationale

The paper's derivation chain is straightforward and non-circular. Each architectural component (FiLM from Perez et al. [32], Bayar convolution from [4], BAM from [31], frozen CLIP from [34]) is externally cited and independently verifiable. The HSCL loss (Eqs. 11–13) is derived from standard supervised contrastive learning [20] with a self-contained modification for hard negative re-weighting; no step reduces to its own inputs by construction. The AMSID training dataset is self-constructed but is used only for training, while all performance claims are evaluated on eight external public benchmark datasets (Table 2) that are distinct from the training data. The ablation study (Table 8) removes or replaces modules and measures performance deltas on an external test set—no prediction is forced by a fitted parameter. The self-citations ([2], [16], [26], [27], [51]) appear only in related work as examples of forensic methods and are not load-bearing for any derivation step. The concern that the ablation does not fully isolate training-data contributions from architectural contributions is a valid experimental-design critique, but it is a correctness risk, not circularity: no claim in the paper is equivalent to its inputs by definition or by a self-citation chain.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 2 invented entities

The axiom ledger reveals several empirically set hyperparameters (lambda, tau, alpha, K) that are fitted to the task. The core architectural and methodological assumptions (CLIP utility, Bayar convolution, aligned data) are standard domain assumptions. The 1-epoch training protocol is an ad-hoc choice that lacks justification.

free parameters (6)
  • lambda (contrastive loss weight) = 0.8
    Empirically set to 0.8 in Section 4.1.3 to balance classification and contrastive losses.
  • tau (temperature) = 0.07
    Contrastive temperature set to 0.07 in Section 4.1.3.
  • alpha (hard negative penalty) = 2.0
    Penalty weight for top 10% hard negatives, set to 2.0 in Section 4.1.3.
  • K (hard negative ratio) = 10%
    Top 10% of negative samples within each batch are mined as hard negatives.
  • gamma (gating parameter) = learnable
    Learnable gating parameter introduced in Eq. (4) to blend original and intermediate features.
  • d (feature dimension) = 512
    Unified dimensional space set to 512 in Section 4.1.3.
axioms (4)
  • domain assumption CLIP ViT-L/14 pre-trained features provide robust generalizable semantic representations for image forensics.
    The RGB branch relies entirely on a frozen CLIP-ViT encoder (Section 3.1.2).
  • domain assumption Bayar constrained convolution captures high-frequency generative artifacts better than standard CNNs or SRM filters.
    The noise branch is built on this assumption (Section 3.1.3).
  • domain assumption Pixel-level aligned training data (AMSID) eliminates content-driven distribution biases.
    The AMSID dataset is constructed based on this premise (Section 3.2).
  • ad hoc to paper Standard backpropagation with Adam optimizer for 1 epoch is sufficient to converge the model.
    The 1-epoch training protocol in Section 4.1.3 is unusual and not justified.
invented entities (2)
  • AMSID (Aligned Multi-source Synthetic Image Dataset) independent evidence
    purpose: Training dataset to force the model to capture semantic-independent generation traces.
    The dataset is described in detail and will be publicly released, allowing independent verification of its effect on training.
  • HSCL (Hard Sample-aware Contrastive Learning) independent evidence
    purpose: Loss function to dynamically re-weight hard negative samples and reshape the feature space.
    The loss is defined mathematically (Eqs. 12-13) and its effect is tested in ablation studies (Table 8).

pith-pipeline@v1.1.0-glm · 24467 in / 2629 out tokens · 484349 ms · 2026-07-08T08:28:05.287965+00:00 · methodology

0 comments
read the original abstract

The rapid advancement of large-scale generative models has accelerated the spread of highly deceptive AI-generated images, making generalized synthetic image detection a critical imperative. Existing forensic networks often struggle with cross-model generalization and realworld degradations due to their reliance on single-domain representations and conventional binary classification optimization. To overcome these limitations, we propose RNSIDNet, a novel forensic framework that achieves robust detection through enhanced RGB-Noise representation learning. Specifically, our method employs a dual-branch architecture where global RGB semantics, extracted by an attention-refined CLIP backbone, dynamically modulate highfrequency noise artifacts captured by Bayar convolutions via a Feature-wise Linear Modulation (FiLM) module. To further enhance the learned representations, we design a Hard Sample-aware Contrastive Learning (HSCL) strategy. By explicitly penalizing challenging training samples, HSCL reshapes the latent feature space to maximize the discriminative margin between pristine and synthetic domains. Extensive experiments across eight public benchmark datasets verify that our model achieves state-of-the-art performance, delivering superior generalization ability, robustness, and computational efficiency. Code and dataset will be publicly available on https://github.com/multimediaFor/RNSIDNet.

Figures

Figures reproduced from arXiv: 2607.06354 by Gang Cao, Lifang Yu, Shaowei Weng, Tian Zhang, Zhen Li.

Figure 1
Figure 1. Figure 1: Overview of the proposed RNSIDNet framework. The input image is processed through two parallel branches: an RGB branch that extracts multi-scale representations using a frozen CLIP-ViT encoder and a Balanced Attention Module (BAM), and a noise branch that captures high-frequency residuals via Bayar convolution followed by Global Average Pooling (GAP). The heterogeneous features are subsequently integrated … view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of local details between synthetic and pristine images. From left to right: original image, zoomed-in local patch, SD XL [33] generated image, and Real-ESRGAN [45] generated image. 3.2. Construction of Aligned Multi-source Synthetic Image Dataset (AMSID) Existing deepfake detection datasets often suffer from limited content diversity and outdated generative models, which cause models to overfit … view at source ↗
Figure 3
Figure 3. Figure 3: Performance evaluation of different models under JPEG compression and Gaussian blur on Synthbuster [3] and AIGCDetectionBenchmark [52] datasets. Beyond data efficiency, [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance comparison of the RNSIDNet model trained with different sample set sizes on the Synthbuster [3] dataset [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Feature distribution visualization of real and synthetic images via t-SNE 5. Conclusion In this work, we propose RNSIDNet, an innovative dual-branch framework for robust and generalizable synthetic image detection. By synergizing global RGB semantics extracted from a visual foundation model with high-frequency artifacts captured by Bayar-constrained convolutions, our method effectively constructs a compreh… 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

53 extracted references · 53 canonical work pages · 5 internal anchors

  1. [1]

    Ntire 2017 challenge on single image super-resolution: Dataset and study, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp

    Agustsson, E., Timofte, R., 2017. Ntire 2017 challenge on single image super-resolution: Dataset and study, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp. 126–135

    work page 2017

  2. [2]

    Ai-generatedvideodetectionviaspatial-temporalanomalylearning,in:ChineseConferenceonPattern Recognition and Computer Vision (PRCV), pp

    Bai,J.,Lin,M.,Cao,G.,Lou,Z.,2024. Ai-generatedvideodetectionviaspatial-temporalanomalylearning,in:ChineseConferenceonPattern Recognition and Computer Vision (PRCV), pp. 460–470

    work page 2024

  3. [3]

    Synthbuster: Towards detection of diffusion model generated images

    Bammey, Q., 2023. Synthbuster: Towards detection of diffusion model generated images. IEEE Open Journal of Signal Processing 5, 1–9

    work page 2023

  4. [4]

    Bayar, B., Stamm, M.C., 2016. A deep learning approach to universal image manipulation detection using a new convolutional layer, in: Proceedings of the 4th ACM Workshop on Information Hiding and Multimedia Security, pp. 5–10

    work page 2016

  5. [5]

    Real-Time Deepfake Detection in the Real-World

    Cavia, B., Horwitz, E., Reiss, T., Hoshen, Y., 2024. Real-time deepfake detection in the real-world. arXiv preprint arXiv:2406.09398

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  6. [6]

    Mmp-2k:Abenchmarkmulti-labeledmacrophotographyimagequalityassessmentdatabase

    Chang,J.,Li,Z.,Lou,J.,Qiu,Z.,Lin,H.,2025. Mmp-2k:Abenchmarkmulti-labeledmacrophotographyimagequalityassessmentdatabase. arXiv preprint arXiv:2505.19065

    work page arXiv 2025

  7. [7]

    Chen, B., Zeng, J., Yang, J., Yang, R., 2024. DRCT: Diffusion reconstruction contrastive training towards universal detection of diffusion generated images, in: Proceedings of the International Conference on Machine Learning, pp. 7621–7639

    work page 2024

  8. [8]

    Dual dataalignment makes ai-generated image detector easier generalizable, in: Advances in Neural Information Processing Systems, pp

    Chen, R.,Xi, J.,Yan, Z., Zhang,K.Y., Wu, S.,Xie, J., Chen,X., Xu,L., Guan, I.,Yao, T., Ding,S., . Dual dataalignment makes ai-generated image detector easier generalizable, in: Advances in Neural Information Processing Systems, pp. 106475–106500

    work page

  9. [9]

    Raising the bar of AI-generated image detection with CLIP, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Cozzolino, D., Poggi, G., Corvi, R., Nießner, M., Verdoliva, L., 2024. Raising the bar of AI-generated image detection with CLIP, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4356–4366

    work page 2024

  10. [10]

    Raise: A raw images dataset for digital image forensics, in: Proceedings of the ACM Multimedia Systems Conference, pp

    Dang-Nguyen, D.T., Pasquini, C., Conotter, V., Boato, G., 2015. Raise: A raw images dataset for digital image forensics, in: Proceedings of the ACM Multimedia Systems Conference, pp. 219–224. Zhen Li et al.:Preprint submitted to ElsevierPage 15 of 17 Generalized Synthetic Image Detection with Enhanced RGB-Noise Representation Learning

    work page 2015

  11. [11]

    3247–3258

    Frank,J.,Eisenhofer,T.,Schönherr,L.,Fischer,A.,Kolossa,D.,Holz,T.,2020.Leveragingfrequencyanalysisfordeepfakeimagerecognition, in: Proceedings of the International Conference on Machine Learning, pp. 3247–3258

    work page 2020

  12. [12]

    Rich models for steganalysis of digital images

    Fridrich, J., Kodovsky, J., 2012. Rich models for steganalysis of digital images. IEEE Transactions on information Forensics and Security 7, 868–882

    work page 2012

  13. [13]

    Generative adversarial nets

    Goodfellow, I.J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y., 2014. Generative adversarial nets. Advances in neural information processing systems 27

    work page 2014

  14. [14]

    20606–20615

    Guillaro,F.,Cozzolino,D.,Sud,A.,Dufour,N.,Verdoliva,L.,2023.Trufor:Leveragingall-roundcluesfortrustworthyimageforgerydetection and localization, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 20606–20615

    work page 2023

  15. [15]

    18685–18694

    Guillaro,F.,Zingarini,G.,Usman,B.,Sud,A.,Cozzolino,D.,Verdoliva,L.,2025.Abias-freetrainingparadigmformoregeneralAI-generated image detection, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 18685–18694

    work page 2025

  16. [16]

    4895–4899

    Guo,K.,Zhu,H.,Cao,G.,2024.Effectiveimagetamperinglocalizationviaenhancedtransformerandco-attentionfusion,in:Icassp2024-2024 ieee international conference on acoustics, speech and signal processing, pp. 4895–4899

    work page 2024

  17. [17]

    Denoising diffusion probabilistic models, in: Advances in Neural Information Processing Systems, pp

    Ho, J., Jain, A., Abbeel, P., 2020. Denoising diffusion probabilistic models, in: Advances in Neural Information Processing Systems, pp. 6840–6851

    work page 2020

  18. [18]

    Fusing global and local features for generalized ai-synthesized image detection, in: 2022 IEEE International Conference on Image Processing, pp

    Ju, Y., Jia, S., Ke, L., Xue, H., Nagano, K., Lyu, S., 2022. Fusing global and local features for generalized ai-synthesized image detection, in: 2022 IEEE International Conference on Image Processing, pp. 3465–3469

    work page 2022

  19. [19]

    Elucidating the design space of diffusion-based generative models, in: Advances in Neural Information Processing Systems, pp

    Karras, T., Aittala, M., Aila, T., Laine, S., 2022. Elucidating the design space of diffusion-based generative models, in: Advances in Neural Information Processing Systems, pp. 26565–26577

    work page 2022

  20. [20]

    Supervised contrastive learning, in: Advances in Neural Information Processing Systems, pp

    Khosla, P., Teterwak, P., Wang, C., Sarna, A., Tian, Y., Isola, P., Maschinot, A., Liu, C., Krishnan, D., 2020. Supervised contrastive learning, in: Advances in Neural Information Processing Systems, pp. 18661–18673

    work page 2020

  21. [21]

    Leveraging representations from intermediate encoder-blocks for synthetic image detection, in: Proceedings of the European Conference on Computer Vision, pp

    Koutlis, C., Papadopoulos, S., 2024. Leveraging representations from intermediate encoder-blocks for synthetic image detection, in: Proceedings of the European Conference on Computer Vision, pp. 394–411

    work page 2024

  22. [22]

    Enhanced deep residual networks for single image super-resolution, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp

    Lim, B., Son, S., Kim, H., Nah, S., Mu Lee, K., 2017. Enhanced deep residual networks for single image super-resolution, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp. 136–144

    work page 2017

  23. [23]

    Microsoft COCO: Common objects in context, in: Proceedings of the European Conference on Computer Vision, pp

    Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., Zitnick, C.L., 2014. Microsoft COCO: Common objects in context, in: Proceedings of the European Conference on Computer Vision, pp. 740–755

    work page 2014

  24. [24]

    Detecting generated images by real images, in: European conference on computer vision, pp

    Liu, B., Yang, F., Bi, X., Xiao, B., Li, W., Gao, X., 2022. Detecting generated images by real images, in: European conference on computer vision, pp. 95–110

    work page 2022

  25. [25]

    Forgery-aware adaptive transformer for generalizable synthetic image detection, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Liu, H., Tan, Z., Tan, C., Wei, Y., Wang, J., Zhao, Y., 2024. Forgery-aware adaptive transformer for generalizable synthetic image detection, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 10770–10780

    work page 2024

  26. [26]

    Exploring multi-view pixel contrast for general and robust image forgery localization

    Lou, Z., Cao, G., Guo, K., Yu, L., Weng, S., 2025a. Exploring multi-view pixel contrast for general and robust image forgery localization. IEEE Transactions on Information Forensics and Security

    work page

  27. [27]

    Trustedvideoinpaintinglocalizationviadeepattentivenoiselearning

    Lou,Z.,Cao,G.,Lin,M.,Yu,L.,Weng,S.,2025b. Trustedvideoinpaintinglocalizationviadeepattentivenoiselearning. IEEETransactions on Dependable and Secure Computing

    work page

  28. [28]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps, in: Advances in Neural Information Processing Systems, pp

    Lu, C., Zhou, Y., Bao, F., Chen, J., Li, C., Zhu, J., 2022. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps, in: Advances in Neural Information Processing Systems, pp. 5775–5787

    work page 2022

  29. [29]

    Towards universal fake image detectors that generalize across generative models, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Ojha, U., Li, Y., Lee, Y.J., 2023. Towards universal fake image detectors that generalize across generative models, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 24480–24489

    work page 2023

  30. [30]

    Representation Learning with Contrastive Predictive Coding

    Oord, A.v.d., Li, Y., Vinyals, O., 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  31. [31]

    BAM: Bottleneck Attention Module

    Park, J., Woo, S., Lee, J.Y., Kweon, I.S., 2018. Bam: Bottleneck attention module. arXiv preprint arXiv:1807.06514

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  32. [32]

    Film:Visualreasoningwithageneralconditioninglayer,in:Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Perez,E.,Strub,F.,DeVries,H.,Dumoulin,V.,Courville,A.,2018. Film:Visualreasoningwithageneralconditioninglayer,in:Proceedings of the AAAI Conference on Artificial Intelligence, pp. 3942–3951

    work page 2018

  33. [33]

    SDXL: Improving latent diffusion models for high-resolution image synthesis, in: Proceedings of the International Conference on Learning Representations, pp

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., Müller, J., Penna, J., Rombach, R., 2024. SDXL: Improving latent diffusion models for high-resolution image synthesis, in: Proceedings of the International Conference on Learning Representations, pp. 1862–1874

    work page 2024

  34. [34]

    Learning transferable visual models from natural language supervision, in: Proceedings of the International Conference on Machine Learning, pp

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al., 2021. Learning transferable visual models from natural language supervision, in: Proceedings of the International Conference on Machine Learning, pp. 8748–8763

    work page 2021

  35. [35]

    Aligneddatasetsimprovedetectionoflatentdiffusion-generatedimages,in:Proceedings of the International Conference on Learning Representations

    Rajan,A.S.,Ojha,U.,Schloesser,J.,Lee,Y.J.,2025. Aligneddatasetsimprovedetectionoflatentdiffusion-generatedimages,in:Proceedings of the International Conference on Learning Representations

    work page 2025

  36. [36]

    Contrastive learning with hard negative samples, in: Proceedings of the International Conference on Learning Representations

    Robinson, J.D., Chuang, C., Sra, S., Jegelka, S., 2021. Contrastive learning with hard negative samples, in: Proceedings of the International Conference on Learning Representations

    work page 2021

  37. [37]

    High-resolution image synthesis with latent diffusion models, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B., 2022. High-resolution image synthesis with latent diffusion models, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 10684–10695

    work page 2022

  38. [38]

    Sha, Z., Li, Z., Yu, N., Zhang, Y., 2023. De-fake: Detection and attribution of fake images generated by text-to-image generation models, in: Proceedings of the 2023 ACM SIGSAC conference on computer and communications security, pp. 3418–3432

    work page 2023

  39. [39]

    Deep Image Fingerprint: Towards Low Budget Synthetic Image Detection and Model Lineage Analysis

    Sinitsa, S., Fried, O., 2023. Deep image fingerprint: Accurate and low budget synthetic image detector. arXiv preprint arXiv:2303.10762 1

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  40. [40]

    7184–7192

    Tan,C.,Tao,R.,Liu,H.,Gu,G.,Wu,B.,Zhao,Y.,Wei,Y.,2025.C2p-clip:Injectingcategorycommonpromptincliptoenhancegeneralization in deepfake detection, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 7184–7192

    work page 2025

  41. [41]

    Tan, C., Zhao, Y., Wei, S., Gu, G., Liu, P., Wei, Y., 2024. Rethinking the up-sampling operations in cnn-based generative network for generalizabledeepfakedetection,in:ProceedingsoftheIEEE/CVFconferenceoncomputervisionandpatternrecognition,pp.28130–28139

    work page 2024

  42. [42]

    Tan, C., Zhao, Y., Wei, S., Gu, G., Wei, Y., 2023. Learning on gradients: Generalized artifacts representation for GAN-generated images detection, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 12105–12114. Zhen Li et al.:Preprint submitted to ElsevierPage 16 of 17 Generalized Synthetic Image Detection with Enha...

    work page 2023

  43. [43]

    Attention is all you need, in: Advances in Neural Information Processing Systems

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, Ł., Polosukhin, I., 2017. Attention is all you need, in: Advances in Neural Information Processing Systems

    work page 2017

  44. [44]

    8695–8704

    Wang,S.Y.,Wang,O.,Zhang,R.,Owens,A.,Efros,A.A.,2020.CNN-generatedimagesaresurprisinglyeasytospot...fornow,in:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 8695–8704

    work page 2020

  45. [45]

    Real-ESRGAN: Training real-world blind super-resolution with pure synthetic data, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Wang, X., Xie, L., Dong, C., Shan, Y., 2021. Real-ESRGAN: Training real-world blind super-resolution with pure synthetic data, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 1905–1914

    work page 2021

  46. [46]

    DIRE for diffusion-generated image detection, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Wang, Z., Bao, J., Zhou, W., Wang, W., Hu, H., Chen, H., Li, H., 2023. DIRE for diffusion-generated image detection, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 22445–22455

    work page 2023

  47. [47]

    A sanity check for AI-generated image detection, in: Proceedings of the International Conference on Learning Representations

    Yan, S., Li, O., Cai, J., Hao, Y., Jiang, X., Hu, Y., Xie, W., 2025. A sanity check for AI-generated image detection, in: Proceedings of the International Conference on Learning Representations

    work page 2025

  48. [48]

    Bdd100k:Adiversedrivingdatasetforheterogeneous multitask learning, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Yu,F.,Chen,H.,Wang,X.,Xian,W.,Chen,Y.,Liu,F.,Madhavan,V.,Darrell,T.,2020. Bdd100k:Adiversedrivingdatasetforheterogeneous multitask learning, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 2636–2645

    work page 2020

  49. [49]

    Adding conditional control to text-to-image diffusion models, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Zhang, L., Rao, A., Agrawala, M., 2023. Adding conditional control to text-to-image diffusion models, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 3836–3847

    work page 2023

  50. [50]

    Makingconvolutionalnetworksshift-invariantagain,in:ProceedingsoftheInternationalConferenceonMachineLearning, pp

    Zhang,R.,2019. Makingconvolutionalnetworksshift-invariantagain,in:ProceedingsoftheInternationalConferenceonMachineLearning, pp. 7324–7334

    work page 2019

  51. [51]

    Hybrid transformer-cnn for real image denoising

    Zhao, M., Cao, G., Huang, X., Yang, L., 2022. Hybrid transformer-cnn for real image denoising. IEEE Signal Processing Letters 29, 1252– 1256

    work page 2022

  52. [52]

    PatchCraft: Exploring Texture Patch for Efficient AI-generated Image Detection

    Zhong, N., Xu, Y., Li, S., Qian, Z., Zhang, X., 2023. Patchcraft: Exploring texture patch for efficient AI-generated image detection. arXiv preprint arXiv:2311.12397

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  53. [53]

    Genimage: A million-scale benchmark for detecting AI-generated image, in: Advances in Neural Information Processing Systems, pp

    Zhu, M., Chen, H., Yan, Q., Huang, X., Lin, G., Li, W., Tu, Z., Hu, H., Hu, J., Wang, Y., 2023. Genimage: A million-scale benchmark for detecting AI-generated image, in: Advances in Neural Information Processing Systems, pp. 77771–77782. Zhen Li et al.:Preprint submitted to ElsevierPage 17 of 17

    work page 2023