Pith. sign in

REVIEW 5 major objections 4 minor 73 references

PE-Mamba: Bidirectional Selective Layer Aggregation for AI-Generated Image Detection

T0 review · 5 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read PE-Mamba claims that treating a vision transformer's layer-wise classification tokens as an ordered sequence—scanned forward and backward by a selective state-space model—generalizes to unseen AI-generated image generators better than…

desk verdict Solid benchmark results, but the paper's core causal claim is undercut by its own ablations. read the letter →

arxiv 2608.07999 v1 pith:OTO4ZFXM submitted 2026-08-08 cs.CV

Kutub Uddin , Nusrat Tasnim , Khalid Malik This is my paper · ORCID
classification cs.CV
keywords AI-generatedimagedetectiondeepfakeselectivestatespacemodelcross-layerfeatureaggregationvisiontransformerLoRAzero-shotgeneralizationforensicrepresentation
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

The paper sets out to establish that cross-layer aggregation in vision-transformer detectors should respect the order of the layers. Instead of combining intermediate CLS tokens with a learned weighted average, PE-Mamba scans them forward and backward with a selective state-space model, letting the detector accumulate forensic evidence from texture-level to semantic-level features and then reinterpret low-level cues in light of high-level context. If the claim holds, a detector trained on a single GAN generator can generalize to diffusion models and other unseen generators, outperforming 18 prior detectors while training just 1.3% of parameters. The reported results—96.6% mACC and 99.5% mAP on UniversalFakeDetect, and 95.3% mACC and 98.1% mAP on AIGCDetect—support the idea that directional aggregation, not just the backbone, drives zero-shot generalization.

What carries the argument

The load-bearing mechanism is the bidirectional selective SSM aggregator (BSA): a Mamba-style selective state space model—a content-adaptive recurrent filter—that scans the sequence of projected layer-wise CLS tokens in natural and reversed order. The scan is content-adaptive because the input, output, and step-size matrices depend on each token, and it has a hard causal inductive bias over the layer hierarchy, unlike a learnable weighted average. A softmax-weighted aggregator (SWA) computes a learned global summary of all layer tokens as a second path, and a sigmoid-gated blend (SGA) with a learnable scalar gate fuses the two outputs before a binary detection head. The whole aggregation sits on a frozen PE-Core vision transformer adapted only through LoRA on the QKV projections, so the new modules and adaptation account for 1.3% of total parameters.

What would settle it

Hold out a validation split of unseen generators, tune the hyperparameters on that split, and retrain the weighted-average baseline and PE-Mamba with identical budget, augmentation, and seeds; if the AIGCDetect gap of 20.5 mACC does not reproduce under these conditions, the claim that the bidirectional scan causes the generalization gain is falsified.

Watch

Extended reading notes

Core claim

PE-Mamba claims that the way a vision transformer's internal layers are combined is a primary driver of AI-generated-image detection generalization. The paper treats the sequence of layer-wise CLS tokens as an ordered signal—shallow layers carrying texture and frequency cues, deep layers carrying semantic content—and scans it in both directions with a selective state space model. The forward scan accumulates shallow-to-deep forensic evidence, the backward scan reinterprets low-level cues in light of deep context, and mean-pooling of all hidden states integrates evidence across every layer. A parallel softmax-weighted global summary and a learnable sigmoid gate fuse the two paths. On ProGAN-trained zero-shot protocols, this configuration reports 96.6% mACC and 99.5% mAP on UniversalFakeDetect and 95.3% mACC and 98.1% mAP on AIGCDetect, outperforming 18 detectors and the weighted-average baseline that uses the same cross-layer aggregation paradigm.

Load-bearing premise

The central claim depends on the comparison being clean: all 18 baselines were trained under the same ProGAN-only zero-shot protocol, and the final hyperparameters were chosen without using the reported test results.

Editorial extensions

If this is right

  • A detector trained only on ProGAN can generalize across GAN families and diffusion generators, reducing the need for per-generator retraining.
  • The directional scan is the cause of the gain, not the backbone alone: in ablation, BSA alone reaches 95.3% mACC versus 90.6% with no aggregation, and the full model beats the weighted-average baseline by 20.5 mACC on AIGCDetect.
  • Parameter-efficient forensic adaptation is sufficient: 0.13% LoRA parameters and 1.3% total trainable parameters achieve the reported results.
  • The detector degrades gracefully under JPEG compression, Gaussian blur, and additive noise, keeping at least 88.4% mACC under the harshest combined perturbation.
  • BSA and SWA capture complementary signals, since combining them reaches 96.1 mACC versus 95.3 and 95.1 mACC for each alone, and the learned gate adds another 0.5 mACC.

Reading between the lines

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

  • If layer ordering is the reason, the same bidirectional scan over sequentially ordered internal representations could help other forensics tasks, such as video deepfake detection, where frames add a second ordered dimension that the paper names as future work.
  • A cleaner causal test would freeze the backbone and vary only the aggregator across random seeds; the paper's ablation approximates this, but per-generator variance would make the 20.5-point gap over the weighted-average baseline more airtight.
  • The weak spots the paper reports—Midjourney and ADM—suggest a natural extension: mixing a small number of non-ProGAN generators into training should lift those two cases, which the paper's own future-work section proposes.
  • Because LoRA rank and dropout barely change results across the tested ranges, the method may be insensitive to adaptation hyperparameters, which would ease deployment.
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

5 major / 4 minor

Summary. PE-Mamba is a ProGAN-trained AIGI detector built on a frozen PE-Core ViT with LoRA adaptation. It extracts per-layer CLS tokens, projects them to a common dimension, aggregates them through a bidirectional selective SSM (BSA) and a softmax-weighted sum (SWA), and fuses the two with a learned sigmoid gate. The paper reports 96.6% mACC / 99.5% mAP on UniversalFakeDetect and 95.3% mACC / 98.1% mAP on AIGCDetect, claiming state-of-the-art results over 18 detectors and attributing the gain to replacing weighted-average cross-layer aggregation with the SSM.

Significance. If the causal attribution were established, the idea of treating cross-layer ViT features as an ordered sequence scanned by a selective SSM would be a novel and useful contribution to AIGI detection. The paper's strengths are its standard ProGAN-only zero-shot protocol, broad baseline coverage, extensive ablations over SSM/LoRA/projection/backbone choices, and careful parameter accounting. The weakness is that the current evidence does not pin the gains on the BSA mechanism: the same-backbone BSA-versus-SWA gap is only 0.2 points on UniversalFakeDetect, no AIGCDetect component ablation is shown, and the learned gate value is not reported. With additional controlled experiments, the result could become a solid empirical contribution.

major comments (5)
  1. [§4.7.3, Table 8] The central mechanistic claim, that the bidirectional selective SSM is what produces the large generalization gains, is not supported by the same-backbone ablation. With the identical PE-Core backbone and training protocol, BSA-only achieves 95.3% mACC and SWA-only achieves 95.1% mACC on UniversalFakeDetect, a 0.2-point difference, whereas Table 9 shows that changing the backbone alone moves mACC from 91.7 (DINOv2-L14-448) to 96.6 (PE-Core-G14-448). Since RINE uses a different backbone, the +5.3 and +20.5 point gains over RINE reported in Sections 4.3-4.4 cannot be attributed to the SSM aggregator without a same-backbone weighted-average or RINE-style baseline; please add that comparison.
  2. [§3.5, Eq. (6)] The deployed model does not replace weighted averaging: Eq. (6) blends the BSA output z_m with the SWA output z_alpha through a learned sigmoid gate gamma, and the trained value of gamma is never reported. If delta(gamma) is near zero at convergence, the BSA branch contributes almost nothing at inference. Please report the learned gate value and add a sensitivity analysis of performance with respect to gamma.
  3. [§4.7.3 / §4.4] All aggregation ablations in Table 8 are performed only on UniversalFakeDetect. The largest claimed advantage over RINE occurs on AIGCDetect (+20.5 mACC), where no component ablation is reported. Please run the BSA/SWA/SGA ablations on AIGCDetect under the same ProGAN-trained protocol; without this, the claim that the BSA drives cross-generator generalization is unverified.
  4. [§4.7 / §4.1] The hyperparameters (d_state, d_expand, LoRA rank, alpha, dropout, projection dimension, number of projection layers, and training steps) are selected by ablations evaluated on the same UniversalFakeDetect benchmark that is then reported as the headline result, and all numbers are single runs without repeated seeds or error bars. This makes differences such as the 96.1-to-96.6 mACC gain from adding SGA potentially within run-to-run noise. Please add at least three seeds with mean and standard deviation for the main results and key ablations, or use an independent validation split for model selection.
  5. [§4.2, Tables 1-3] The comparison protocol is not fully specified: only the starred methods in Tables 1-2 are described as retrained on ProGAN, and no per-baseline information is given about which numbers were re-computed, input resolutions, augmentation, or training steps. Table 3 states that all detectors use the same ProGAN-trained weights, but this is only clearly true for the starred methods. Please provide a precise per-baseline protocol so that the claim of outperforming 18 detectors is verifiable under a single evaluation pipeline.
minor comments (4)
  1. [§3.2] The text says forward hooks are registered on 'all normalization modules', while the Abstract and Section 1 say 'second normalization layers'; please make the extraction point consistent.
  2. [§1, references] Reference [52] is cited twice in the first paragraph of the Introduction; one duplicate should be removed.
  3. [§4.5, Table 4] Table 4 reports robustness only for PE-Mamba, without comparisons to any prior detector under the same perturbations, so the conclusion that the method is 'inherently more robust' is not directly supported; please add competitor robustness results or temper the claim.
  4. [§4.6, Figure 2] The Grad-CAM discussion is qualitative; adding a quantitative faithfulness or localization metric would strengthen the interpretability claim, though this is not required for the main result.

Circularity Check

1 steps flagged · score 2.0 of 10

No definitional circularity; the core cross-generator results are independent, but UniversalFakeDetect is used both to select hyperparameters and to report the headline numbers, making the UFD result partly selected rather than predicted.

  1. fitted input called prediction [Section 4.7 (Ablation Study) and Tables 1-2]
    "All experiments follow the standard ProGAN-trained protocol on UniversalFakeDetect [68], with a single variable changed at a time ... Based on these results, we adopt r=8, α=8, and p=0.1 as the final LoRA configuration ... We therefore adopt 2,000 steps as our default."

    Section 4.7 selects the final configuration (LoRA rank/alpha/dropout, SSM state size and expansion, projection dimension and number of projection layers, and 2,000 training steps) by maximizing UniversalFakeDetect mACC, and the same UniversalFakeDetect mACC (96.6%) and mAP (99.5%) are then reported in Tables 1-2 as the method's headline result. The reported UFD numbers are therefore the optimization target of the ablation sweep, not an independent prediction from a pre-specified model; the UFD 'prediction' is forced by the selection procedure. AIGCDetect is genuinely held out, which limits the overall circularity.

full rationale

The paper's central contribution is empirical rather than derivational: PE-Mamba is trained on ProGAN and evaluated zero-shot on UniversalFakeDetect and AIGCDetect, so there is no equation-level circularity in the main results. The AIGCDetect benchmark is never used for ablations and provides an independent test of cross-generator generalization. The main circularity-adjacent flaw is that Section 4.7 selects all hyperparameters—including training steps, projection dimension, LoRA rank/alpha/dropout, SSM state size, and expansion—using UniversalFakeDetect mACC, and the same UFD mACC/mAP are then reported as the headline numbers in Tables 1-2. That is benchmark-selection on the test set rather than a definitional equivalence, and it does not affect the held-out AIGCDetect claims. Self-citations (e.g., refs. 51-54, 66) appear in related work and robustness framing but are not load-bearing for the main comparison. The mechanistic claim that the SSM 'replaces' weighted-average aggregation is internally weakened by the final model's learned gate combining BSA with SWA (Eq. 6) and by Table 8 showing BSA-only (95.3) and SWA-only (95.1) nearly tied, but this is an attribution/validity concern, not circularity.

Assumptions & free parameters 9 free parameters · 4 assumptions · 0 invented entities

The empirical claims rest on standard deep-learning assumptions (pretrained features transfer, zero-order-hold discretization is stable, LoRA preserves representations) plus a batch of hyperparameters that are selected by ablation against the same benchmark used for the final headline numbers. No new physical entities are introduced; the only new objects are neural network modules whose behavior is entirely defined by the training data.

free parameters (9)
  • SSM state dimension d_state = 16
    Selected by ablation on UniversalFakeDetect (Table 6); 16 yields the best mACC/mAP.
  • SSM expansion factor d_expand = 2
    Selected by ablation on UniversalFakeDetect (Table 7).
  • Number of SSM blocks/layers = 1
    Selected by ablation on UniversalFakeDetect (Table 5); deeper stacking degrades performance.
  • LoRA rank r = 8
    Selected by ablation on UniversalFakeDetect (Figure 3a).
  • LoRA scaling factor alpha = 8
    Selected by ablation on UniversalFakeDetect (Figure 3b).
  • LoRA dropout probability p = 0.1
    Selected by ablation on UniversalFakeDetect (Figure 3c).
  • Projection dimension D' = 1024
    Selected by ablation on UniversalFakeDetect (Figure 4b).
  • Number of projection layers n = 2
    Selected by ablation on UniversalFakeDetect (Figure 4a).
  • Training steps = 2000
    Selected by early stopping on UniversalFakeDetect (Figure 4c).
assumptions (4)
  • domain assumption Layer-wise CLS tokens from a frozen vision transformer form a semantically ordered progression from shallow texture cues to deep semantic representations.
    Section 3.2 and Section 1; the design of BSA depends on this ordering being meaningful and useful for forensics.
  • domain assumption Training only on ProGAN and evaluating zero-shot on UniversalFakeDetect and AIGCDetect is a valid measure of generalization.
    Section 4.2; this protocol is standard in the field and is the basis of all benchmarking, but it assumes the test generators are truly unseen and the labels are correct.
  • standard math The selective SSM update (zero-order hold discretization, Equations 1-2) is numerically stable and appropriate for the layer-aggregation task.
    Borrowed from Mamba [19]; the paper does not re-derive stability but relies on it.
  • domain assumption PE-Core pretrained features are sufficiently rich that LoRA adaptation on QKV projections alone can recover forensic cues.
    Sections 3.1-3.2 and the backbone ablation in Table 9; if PE-Core features were unsuitable, the backbone choice would not support the claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PE-Mamba: Bidirectional Selective Layer Aggregation for AI-Generated Image Detection." pith.science (2026). https://pith.science/paper/OTO4ZFXM

@misc{pith2026260807999,
  author       = {Pith},
  title        = {Pith review of: PE-Mamba: Bidirectional Selective Layer Aggregation for AI-Generated Image Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OTO4ZFXM}},
  note         = {Machine review of arXiv:2608.07999}
}
read the original abstract

AI-generated image (AIGI) detection has become increasingly challenging due to the rapid advancement of generative models and the diminishing gap between synthetic and authentic content. Existing vision transformer-based detectors commonly rely on weighted-sum strategies to aggregate intermediate representations across transformer layers, often overlooking the inherently ordered semantic progression of hierarchical features from shallow texture cues to deep semantic representations. In this work, we propose \textbf{PE-Mamba}, a novel framework built upon a pre-trained PE-Core vision transformer with lightweight LoRA adaptation that introduces three complementary components for cross-layer feature aggregation and fusion. First, a bidirectional selective aggregator (BSA) processes layer-wise classification tokens through forward and backward selective scans, where the forward scan progressively accumulates shallow-to-deep forensic evidence, and the backward scan performs deep-to-shallow contextual refinement to reinterpret low-level cues in light of high-level semantic context. Second, a softmax-weighted aggregator (SWA) computes a learned global summary of all layer tokens as a complementary aggregation path. Third, a sigmoid-gated blend (SGA) adaptively fuses the BSA and SWA outputs via a learnable scalar gate, allowing the model to dynamically balance directional sequential evidence and global layer-wise aggregation. Extensive experiments on UniversalFakeDetect (96.6\% mACC, 99.5\% mAP) and AIGCDetect (95.3\% mACC, 98.1\% mAP) demonstrate that \methodname{} outperforms 18 detectors with superior generalization across diverse generative models, while training only 1.3\% of total parameters (0.13\% for LoRA alone).

Figures

Figures reproduced from arXiv: 2608.07999 by the authors.

Figure 1
Figure 1. Overview of the proposed PE-Mamba framework for AIGI detection. A frozen PE-Core vision transformer with LoRA adaptation on QKV projections extracts layer-wise CLS tokens from all second normalization layers, which are projected into a unified embed￾ding space by the first projection layer. The BSA processes the projected tokens through forward (shallow to deep) and backward (deep to shallow) selective scans to prod… view at source ↗
Figure 2
Figure 2. Grad-CAM visualizations comparing the pretrained PE-Core and the proposed [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Ablation on LoRA hyperparameters: (a) rank [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Ablation on (a) projection layers n, (b), projection dimension d, and (c) training steps. Optimal settings are 2,000 steps, n=2, and d=1024. Gold stars mark the best ACC. As shown in [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 58 canonical work pages

  1. [1]

    CNN detec- tion of GAN-generated face images based on cross-band co-occurrences analysis

    Mauro Barni, Kassem Kallas, Ehsan Nowroozi, and Benedetta Tondi. CNN detec- tion of GAN-generated face images based on cross-band co-occurrences analysis. In IEEE International Workshop on Information Forensics and Security (WIFS), pages 1–6. IEEE, 2020

    work page 2020

  2. [2]

    Perception en- coder: The best visual embeddings are not at the output of the network.arXiv preprint arXiv:2504.13181, 2025

    Daniel Bolya, Po-Yao Huang, Peize Sun, Jang Hyun Cho, Andrea Madotto, Chen Wei, Tengyu Ma, Jiale Zhi, Jathushan Rajasegaran, Hanoona Rasheed, et al. Perception en- coder: The best visual embeddings are not at the output of the network.arXiv preprint arXiv:2504.13181, 2025

    arXiv 2025

  3. [3]

    Large scale GAN training for high fidelity natural image synthesis.arXiv preprint arXiv:1809.11096, 2018

    Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale GAN training for high fidelity natural image synthesis.arXiv preprint arXiv:1809.11096, 2018

    arXiv 2018

  4. [4]

    Emerging properties in self-supervised vision trans- formers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bo- janowski, and Armand Joulin. Emerging properties in self-supervised vision trans- formers. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion (ICCV), pages 9650–9660, 2021

    work page 2021

  5. [5]

    Learning to see in the dark

    Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun. Learning to see in the dark. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 3291–3300, 2018

    work page 2018

  6. [6]

    Photographic image synthesis with cascaded refine- ment networks

    Qifeng Chen and Vladlen Koltun. Photographic image synthesis with cascaded refine- ment networks. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 1511–1520, 2017

    work page 2017

  7. [7]

    ForgeLens: Data-efficient forgery focus for generalizable forgery image detection, 2025

    Yingjian Chen, Lei Zhang, and Yakun Niu. ForgeLens: Data-efficient forgery focus for generalizable forgery image detection, 2025

    work page 2025

  8. [8]

    StarGAN: Unified generative adversarial networks for multi-domain image-to- image translation

    Yunjey Choi, Minje Choi, Munyoung Kim, Jung-Woo Ha, Sunghun Kim, and Jaegul Choo. StarGAN: Unified generative adversarial networks for multi-domain image-to- image translation. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 8789–8797, 2018

    work page 2018

Show all 73 references
  1. [9]

    On the detection of synthetic images generated by diffusion mod- els

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

    2023

  2. [10]

    Raising the bar of AI-generated image detection with CLIP.arXiv preprint arXiv:2312.00195, 2024

    Davide Cozzolino, Giovanni Poggi, Riccardo Corvi, Matthias Nießner, and Luisa Ver- doliva. Raising the bar of AI-generated image detection with CLIP.arXiv preprint arXiv:2312.00195, 2024. 16UDDIN ET AL.: PE-MAMBA FOR AI-GENERA TED IMAGE DETECTION

    2024 arXiv

  3. [11]

    Second-order attention network for single image super-resolution

    Tao Dai, Jianrui Cai, Yongbing Zhang, Shu-Tao Xia, and Lei Zhang. Second-order attention network for single image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11065–11074, 2019

    2019

  4. [12]

    Diffusion models beat GANs on image syn- thesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat GANs on image syn- thesis. InAdvances in Neural Information Processing Systems (NeurIPS), volume 34, pages 8780–8794, 2021

    2021

  5. [13]

    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 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7890–7899, 2020

    2020

  6. [14]

    Deepfake-related fraud forecast to hit $40b by 2027.https://www

    Eftsure. Deepfake-related fraud forecast to hit $40b by 2027.https://www. eftsure.com/statistics/deepfake-statistics, 2025. [Online; ac- cessed 2026]

    2027

  7. [15]

    Transferable adversarial attacks on audio deepfake detection

    Muhammad Umar Farooq, Awais Khan, Kutub Uddin, and Khalid Mahmood Malik. Transferable adversarial attacks on audio deepfake detection. In2025 IEEE/CVF Win- ter Conference on Applications of Computer Vision Workshops (WACVW), pages 1555–

  8. [16]

    Uncertainty- aware deepfake detection via multi-view structural learning.arXiv preprint arXiv:2607.28769, 2026

    Muhammad Umar Farooq, Kutub Uddin, Awais Khan, and Khalid Malik. Uncertainty- aware deepfake detection via multi-view structural learning.arXiv preprint arXiv:2607.28769, 2026

    2026 arXiv

  9. [17]

    Leveraging frequency analysis for deep fake image recog- nition

    Joshua Frank, Thorsten Eisenhofer, Lea Schönherr, Andreas Fischer, Dorothea Kolossa, and Thorsten Holz. Leveraging frequency analysis for deep fake image recog- nition. InInternational Conference on Machine Learning (ICML), pages 3247–3258. PMLR, 2020

    2020

  10. [18]

    Deepfake fraud costs the financial sector an average of $600,000 per company.https://www.businesswire.com/news/home/ 20241031656724/en/Deepfake-Fraud-Costs, 2024

    Globe Newswire. Deepfake fraud costs the financial sector an average of $600,000 per company.https://www.businesswire.com/news/home/ 20241031656724/en/Deepfake-Fraud-Costs, 2024. [Online; accessed 2026]

    2024

  11. [19]

    Mamba: Linear-time sequence modeling with selective state spaces.arXiv preprint arXiv:2312.00752, 2023

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces.arXiv preprint arXiv:2312.00752, 2023

    2023 arXiv

  12. [20]

    Efficiently modeling long sequences with structured state spaces.arXiv preprint arXiv:2111.00396, 2021

    Albert Gu, Karan Goel, and Christopher Ré. Efficiently modeling long sequences with structured state spaces.arXiv preprint arXiv:2111.00396, 2021

    2021 arXiv

  13. [21]

    Vector quantized diffusion model for text-to-image syn- thesis

    Shuyang Gu, Dong Chen, Jianmin Bao, Fang Wen, Bo Zhang, Dongdong Chen, Lu Yuan, and Baining Guo. Vector quantized diffusion model for text-to-image syn- thesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10696–10706, 2022

    2022

  14. [22]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016. UDDIN ET AL.: PE-MAMBA FOR AI-GENERA TED IMAGE DETECTION17

    2016

  15. [23]

    LoRA: Low-rank adaptation of large language models.Proceedings of the International Conference on Learning Representations (ICLR), 2022

    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.Proceedings of the International Conference on Learning Representations (ICLR), 2022

    2022

  16. [24]

    Progressive growing of gans for improved quality, stability, and variation.arXiv preprint arXiv:1710.10196, 2017

    Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gans for improved quality, stability, and variation.arXiv preprint arXiv:1710.10196, 2017

    2017 arXiv

  17. [25]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. InProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 4401–4410, 2019

    2019

  18. [26]

    Analyzing and improving the image quality of StyleGAN

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of StyleGAN. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8110–8119, 2020

    2020

  19. [27]

    Leveraging representations from inter- mediate encoder-blocks for synthetic image detection

    Christos Koutlis and Symeon Papadopoulos. Leveraging representations from inter- mediate encoder-blocks for synthetic image detection. InEuropean Conference on Computer Vision (ECCV), pages 394–411. Springer, 2024

    2024

  20. [28]

    Diverse image synthesis from semantic lay- outs via conditional IMLE

    Ke Li, Tianhao Zhang, and Jitendra Malik. Diverse image synthesis from semantic lay- outs via conditional IMLE. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4220–4229, 2019

    2019

  21. [29]

    Im- proving synthetic image detection towards generalization: An image transformation perspective

    Ouxiang Li, Jiayin Cai, Yanbin Hao, Xiaolong Jiang, Yao Hu, and Fuli Feng. Im- proving synthetic image detection towards generalization: An image transformation perspective. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Dis- covery and Data Mining V . 1, pages 2...

    2025

  22. [30]

    Towards generalizable ai-generated image detection via image-adaptive prompt learning.arXiv preprint arXiv:2508.01603, 2025

    Yiheng Li, Zichang Tan, Guoqing Xu, Zhen Lei, Xu Zhou, and Yang Yang. Towards generalizable ai-generated image detection via image-adaptive prompt learning.arXiv preprint arXiv:2508.01603, 2025

    2025

  23. [31]

    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 Recog- nition (CVPR), pages 10770–10780, 2024

    2024

  24. [32]

    VMamba: Visual state space model.arXiv preprint arXiv:2401.10166, 2024

    Yue Liu, Yunjie Tian, Yuzhong Zhao, Hongtian Yu, Lingxi Xie, Yaowei Wang, Qix- iang Shen, and Yunfan Liu. VMamba: Visual state space model.arXiv preprint arXiv:2401.10166, 2024

    2024 arXiv

  25. [33]

    De- tecting GAN generated fake images using co-occurrence matrices.arXiv preprint arXiv:1903.06836, 2019

    Lakshmanan Nataraj, Tajuddin Manhar Mohammed, Shivkumar Chandrasekaran, Ar- juna Flenner, Jawadul H Bappy, Amit K Roy-Chowdhury, and B S Manjunath. De- tecting GAN generated fake images using co-occurrence matrices.arXiv preprint arXiv:1903.06836, 2019

    1903 arXiv

  26. [34]

    GLIDE: Towards photorealistic image gener- ation and editing with text-guided diffusion models.arXiv preprint arXiv:2112.10741, 2021

    Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. GLIDE: Towards photorealistic image gener- ation and editing with text-guided diffusion models.arXiv preprint arXiv:2112.10741, 2021. 18UDDIN ET AL.: PE-MAMB...

    2021 arXiv

  27. [35]

    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 (CVPR), pages 24480–24489, 2023

    2023

  28. [36]

    Semantic image synthesis with spatially-adaptive normalization

    Taesung Park, Ming-Yu Liu, Ting-Chun Wang, and Jun-Yan Zhu. Semantic image synthesis with spatially-adaptive normalization. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), pages 2337–2346, 2019

    2019

  29. [37]

    Sdxl: Improving latent diffusion models for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. InInternational Conference on Learning Representa- tions, volume 2024, pages 1862–...

    2024

  30. [38]

    Learning transferable visual models from natural lan- guage supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sand- hini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural lan- guage supervision. InInternational ...

    2021

  31. [39]

    Stay-positive: A case for ignoring real image features in fake image detection.Proceedings of the International Conference on Machine Learning (ICML), 2025

    Anirudh Sundara Rajan and Yong Jae Lee. Stay-positive: A case for ignoring real image features in fake image detection.Proceedings of the International Conference on Machine Learning (ICML), 2025

    2025

  32. [40]

    Aligned datasets improve detection of latent diffusion-generated images.Proceedings of the International Conference on Learning Representations (ICLR), 2025

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

    2025

  33. [41]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. InInternational Conference on Machine Learning (ICML), pages 8821–8831. PMLR, 2021

    2021

  34. [42]

    Hi- erarchical text-conditional image generation with CLIP latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hi- erarchical text-conditional image generation with CLIP latents. InarXiv preprint arXiv:2204.06125, 2022

    2022 arXiv

  35. [43]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Om- mer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10684–10695, 2022

    2022

  36. [44]

    FaceForensics++: Learning to detect manipulated facial images

    Andreas Rossler, Davide Cozzolino, Luisa Verdoliva, Christian Riess, Justus Thies, and Matthias Nießner. FaceForensics++: Learning to detect manipulated facial images. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 1–11, 2019

    2019

  37. [45]

    Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra

    Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-CAM: Visual explanations from deep networks via gradient-based localization. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 6...

    2017

  38. [46]

    Mirage: Multimodal discriminative representation learning for generalizable ai-generated image detection

    Kuo Shi, Jie Lu, Shanshan Ye, Guangquan Zhang, and Zhen Fang. Mirage: Multimodal discriminative representation learning for generalizable ai-generated image detection. InProceedings of the 33rd ACM International Conference on Multimedia, pages 353– 361, 2025

    2025

  39. [47]

    Learning on gradients: Generalized artifacts representation for GAN-generated images detec- tion

    Chuangchuang Tan, Yao Zhao, Shikui Wei, Guanghua Gu, and Yunchao Wei. Learning on gradients: Generalized artifacts representation for GAN-generated images detec- tion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12105–12114, 2023

    2023

  40. [48]

    Frequency-aware deepfake detection: Improving generalizability through fre- quency space domain learning

    Chuangchuang Tan, Yao Zhao, Shikui Wei, Guanghua Gu, Ping Liu, and Yunchao Wei. Frequency-aware deepfake detection: Improving generalizability through fre- quency space domain learning. InProceedings of the AAAI Conference on Artificial Intelligence, pages 5052–5060, 2024

    2024

  41. [49]

    Rethinking 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. Rethinking the up-sampling operations in CNN-based generative network for generalizable deepfake detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR...

    2024

  42. [50]

    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 Artifi- cial Intelligence, pages 7184–7192, 2025

    2025

  43. [51]

    Nusrat Tasnim, Kutub Uddin, and Khalid Mahmood Malik. AI-generated image de- tection: An empirical study and future research directions.Proceedings of the BMVC 2025 Workshop on Media Authenticity in the Age of Artificial Intelligence, 2025

    2025

  44. [52]

    Grex-bench: Benchmarking gen- eralization, robustness, and explainability in ai-generated image detection

    Nusrat Tasnim, Khalid Malik, et al. Grex-bench: Benchmarking gen- eralization, robustness, and explainability in ai-generated image detection. https://doi.org/10.21203/rs.3.rs-8633550/v1, 2026

    2026 doi

  45. [53]

    Nusrat Tasnim, Kutub Uddin, and Khalid Malik. A comprehensive survey, large-scale empirical study, and future insights on generalization, robustness, and explainability of ai-generated image detection.SSRN (January 06, 2026), 2026

    2026

  46. [54]

    Nusrat Tasnim, Kutub Uddin, and Khalid Malik. Diversity matters: Dataset diversi- fication and dual-branch network for generalized ai-generated image detection.Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026

    2026

  47. [55]

    Enhanced adversarial attack for avoidance of fake image detection.Journal of Broadcast Engineering, 28(7):859–866, 2023

    Kutub Uddin and Byung Tae Oh. Enhanced adversarial attack for avoidance of fake image detection.Journal of Broadcast Engineering, 28(7):859–866, 2023

    2023

  48. [56]

    Anti-forensic against double jpeg compression detection using adversarial generative network.Proceedings of the Korean Society of Broadcast Engineers Conference, pages 58–60, 2019

    Kutub Uddin, Yoonmo Yang, and Byung Tae Oh. Anti-forensic against double jpeg compression detection using adversarial generative network.Proceedings of the Korean Society of Broadcast Engineers Conference, pages 58–60, 2019

    2019

  49. [57]

    Analysis of generative adversarial network targeting anti-forensic in jpeg compressed domain

    Kutub Uddin, Yoonmo Yang, and Byung Tae Oh. Analysis of generative adversarial network targeting anti-forensic in jpeg compressed domain. InInternational Workshop on Advanced Imaging Technology (IWAIT) 2021, volume 11766, pages 627–631. SPIE, 2021. 20UDDIN ET AL.: PE-MAMBA FOR...

    2021

  50. [58]

    Double compression detection in hevc-coded video with the same coding parameters using picture partitioning informa- tion.Signal Processing: Image Communication, 103:116638, 2022

    Kutub Uddin, Yoonmo Yang, and Byung Tae Oh. Double compression detection in hevc-coded video with the same coding parameters using picture partitioning informa- tion.Signal Processing: Image Communication, 103:116638, 2022

    2022

  51. [59]

    A robust open- set multi-instance learning for defending adversarial attacks in digital image.IEEE Transactions on Information Forensics and Security, 19:2098–2111, 2023

    Kutub Uddin, Yoonmo Yang, Tae Hyun Jeong, and Byung Tae Oh. A robust open- set multi-instance learning for defending adversarial attacks in digital image.IEEE Transactions on Information Forensics and Security, 19:2098–2111, 2023

    2023

  52. [60]

    Deep learning-based counter anti- forensic of gan-based attack in hevc compressed domain using coding pattern analysis

    Kutub Uddin, Yoonmo Yang, and Byung Tae Oh. Deep learning-based counter anti- forensic of gan-based attack in hevc compressed domain using coding pattern analysis. Expert Systems with Applications, 233:120912, 2023

    2023

  53. [61]

    Counter-act against gan-based attacks: A collaborative learning approach for anti-forensic detection.Applied Soft Computing, 153:111287, 2024

    Kutub Uddin, Tae Hyun Jeong, and Byung Tae Oh. Counter-act against gan-based attacks: A collaborative learning approach for anti-forensic detection.Applied Soft Computing, 153:111287, 2024

    2024

  54. [62]

    Adversarial attacks on audio deepfake detection: A benchmark and comparative study

    Kutub Uddin, Muhammad Umar Farooq, Awais Khan, and Khalid Mahmood Malik. Adversarial attacks on audio deepfake detection: A benchmark and comparative study. arXiv preprint arXiv:2509.07132, 2025

    2025 arXiv

  55. [63]

    Advbench: A comprehensive benchmark of adversarial attacks on deepfake detectors in real-world consumer ap- plications

    Kutub Uddin, Muhammad Umar Farooq, Awais Khan, Muhammad Saad Saeed, Ijaz Ul Haq, Nusrat Tasnim, and Khalid Mahmood Malik. Advbench: A comprehensive benchmark of adversarial attacks on deepfake detectors in real-world consumer ap- plications. 2025

    2025

  56. [64]

    SHIELD: A secure and highly enhanced integrated learning for robust deepfake detection against adversarial attacks.arXiv preprint arXiv:2507.13170, 2025

    Kutub Uddin, Awais Khan, Muhammad Umar Farooq, and Khalid Malik. SHIELD: A secure and highly enhanced integrated learning for robust deepfake detection against adversarial attacks.arXiv preprint arXiv:2507.13170, 2025

    2025 arXiv

  57. [65]

    GUARD: Generative unmasking and adversarial-resistant deepfake detection using multi-model knowledge distillation.Authorea Preprints, 2025

    Kutub Uddin, Nusrat Tasnim, Muhammad Saad Saeed, and Khalid Mahmood Ma- lik. GUARD: Generative unmasking and adversarial-resistant deepfake detection using multi-model knowledge distillation.Authorea Preprints, 2025

    2025

  58. [66]

    Do transformations reveal the truth? generative residual learning for generalized ai-generated image detection.arXiv preprint arXiv:2607.08674, 2026

    Kutub Uddin, Nusrat Tasnim, Awais Khan, Mohammad Umar Farooq, and Khalid Ma- lik. Do transformations reveal the truth? generative residual learning for generalized ai-generated image detection.arXiv preprint arXiv:2607.08674, 2026

    2026 arXiv

  59. [67]

    Face2parts: Exploring coarse-to-fine inter-regional facial dependencies for generalized deepfake detection.IEEE Access, 14:55111–55125, 2026

    Kutub Uddin, Nusrat Tasnim, and Byung Tae Oh. Face2parts: Exploring coarse-to-fine inter-regional facial dependencies for generalized deepfake detection.IEEE Access, 14:55111–55125, 2026

    2026

  60. [68]

    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 (CVPR), pages 8695–8704, 2020

    2020

  61. [69]

    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 (ICCV), pages 22445–22455, 2023. UDDIN ET AL.: PE-MAMBA FOR AI-...

    2023

  62. [70]

    A sanity check for AI-generated image detection.Proceedings of the International Conference on Learning Representations (ICLR), 2025

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

    2025

  63. [71]

    Zhiyuan Yan, Jiangming Wang, Zhendong Wang, Peng Jin, Ke-Yue Zhang, Shen Chen, Taiping Yao, Shouhong Ding, Baoyuan Wu, and Li Yuan. EFFORT: Efficient orthog- onal modeling for generalizable AI-generated image detection.Proceedings of the IEEE/CVF Conference on Computer Vision ...

    2024

  64. [72]

    Unpaired image-to- image translation using cycle-consistent adversarial networks

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to- image translation using cycle-consistent adversarial networks. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 2223–2232, 2017

    2017

  65. [73]

    Vision mamba: Efficient visual representation learning with bidirectional state space model.arXiv preprint arXiv:2401.13660, 2024

    Lianghui Zhu, Bencheng Liao, Qian Zhang, Xinlong Wang, Wenyu Liu, and Xinggang Wang. Vision mamba: Efficient visual representation learning with bidirectional state space model.arXiv preprint arXiv:2401.13660, 2024

    2024 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.