Pith. sign in

REVIEW 4 major objections 5 minor 54 references

Generated Images Are Easier to Forget: A Machine Unlearning Perspective for Synthetic Image Detection

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read When a vision model is made to forget, generated images lose their features faster than natural images, and this asymmetry alone can detect AI-generated images without any training.

desk verdict A simple pruning-based feature-similarity score detects generated images surprisingly well, but the theory assumes the very gap it claims to explain. read the letter →

arxiv 2608.00716 v1 pith:LNNHWWST submitted 2026-08-01 cs.CV cs.LG

Jun Nie , Yonggang Zhang , Tongliang Liu , Yiu-ming Cheung , Bo Han , Xinmei Tian This is my paper · ORCID
classification cs.CVcs.LG
keywords generatedimagedetectionmachineunlearningweightpruninglarge-scalevisionmodelssyntheticfeaturesimilaritydata-freedomainshift
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

This paper claims that the act of forgetting, not learning, is what separates AI-generated images from natural ones. Large vision models pretrained on natural-image-dominated data represent both kinds of images with comparably low loss, but when the model is degraded by pruning weights, the representations of generated images shift far more than those of natural images. The authors turn this asymmetry into a detector by scoring an image with the cosine similarity between its features in the original and the pruned model, with a larger drop signaling a generated image. If the claim holds, one pretrained model plus a parameter-pruning step yields a training-free detector that generalizes to generators never seen during development, and using a few generated images during unlearning makes detection stronger. This positions the task as a matter of selectively damaging a model's knowledge of one class rather than drawing a boundary between classes.

What carries the argument

The carrying object is the pruning-induced output shift analyzed through the Jacobian of the representation with respect to model parameters. Assumption 3.3 posits a directional representation sensitivity gap along the pruning direction $\Delta\theta=\theta-\theta'$: generated images have larger expected squared Jacobian-vector norm than natural images, $S_{D_{\mathrm{gen}}}(\Delta\theta)-S_{D_{\mathrm{nat}}}(\Delta\theta)\ge \omega\|\Delta\theta\|_2^2$. Proposition 3.4 then shows, under local smoothness, that the pruning-induced output difference is larger for generated images when $\omega > C_f\|\Delta\theta\|_2$, which is exactly what makes the cosine-similarity score in Eq. (4) discriminative. The data-driven variant replaces passive pruning with an explicit unlearning loss that retains natural-image similarity and enforces a margin $\gamma$ between original and unlearned features for generated images.

What would settle it

Take any vision model pretrained on natural-image-dominated data, prune the same small-magnitude weights, and measure whether images from a held-out generator show larger feature displacement than natural images; if a generator that the paper did not test produces no separation under the cosine-similarity score, or if a model pretrained with substantial synthetic data no longer shows the gap, the central claim fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is disparate forgetting dynamics: during machine unlearning, feature degradation for generated images escalates faster than for natural images. The detector score is $s(x)=\cos(F(x;\theta_F), F(x;\theta'_F))$, the cosine similarity between representations under the original and the unlearned model; natural images keep high similarity while generated images' features drift. In the main data-free setup the unlearned model is obtained by pruning 90% of the smallest-magnitude weights in the fc2 layer of block 16 of DINOv2 ViT-L/14, and the paper reports that this score exceeds conventional detectors on ImageNet, GenImage, Chameleon, LSUN-BEDROOM, DiffusionForensics, DRCT-2M, and on images sampled from Sora and Open Sora videos. When generated images are available, a data-driven variant fine-tunes the model with LoRA to push the same score separation further, with the paper reporting average AUROC of 98.29 versus 92.20 for the data-free version on ImageNet.

Load-bearing premise

The load-bearing premise is that generated images really are more sensitive than natural images to the particular pruning direction; the paper assumes this sensitivity gap ($\omega>0$) instead of proving it, and it notes that the gap may shrink as future pretraining corpora include more synthetic images.

Editorial extensions

If this is right

  • A training-free detector can be instantiated from any pretrained vision model by pruning a layer and comparing features before and after, with no labels and no generator-specific training data.
  • Because the score comes from a distributional asymmetry rather than a learned boundary, it transfers to generators unseen at development time, including video-based image samples the paper tests.
  • Access to a modest set of generated images improves the score further: the data-driven unlearning variant raises average AUROC on ImageNet from 92.20 to 98.29 in the paper's setup.
  • The detection threshold is not fixed at 0.5; it must be calibrated on a validation set, and the paper reports that accuracy varies noticeably with the threshold choice.

Reading between the lines

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

  • In the paper's own terms, the sensitivity gap is about distribution shift, not about generative models specifically; a natural image that lies in a long-tail or out-of-distribution region of the pretraining data should also produce a large similarity drop, so the same score may flag unusual natural images as well.
  • The directional sensitivity gap could be measured directly on a small calibration set, turning hyperparameter selection (which block, which pruning ratio) into an automatic procedure rather than a manual choice.
  • If synthetic images keep entering pretraining corpora, the asymmetry may invert or vanish, so a future detector may need to prune in a direction that specifically amplifies rare natural images or to re-derive the score from a model whose pretraining distribution is known.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes detecting AI-generated images by exploiting the observation that, when a large pretrained vision model is degraded via machine unlearning, its representations of generated images change more rapidly than those of natural images. The authors introduce two detectors built on the cosine similarity between original and pruned (or unlearned) features: a data-free method that prunes 90% of the smallest-magnitude weights in the fc2 layer of the 16th block of DINOv2 ViT-L/14, and a data-driven method that fine-tunes DINOv2 with LoRA using an unlearning objective that pushes generated-image representations away from the original model while preserving natural-image representations. Experiments are reported on ImageNet, GenImage, DiffusionForensics, DRCT-2M, LSUN-BEDROOM, Chameleon, and on Sora/OpenSora video-derived images. The paper also presents a local perturbation analysis, Proposition 3.4, intended to explain why pruning amplifies representation differences for generated images.

Significance. If the empirical finding holds, the proposed approach is practically significant: a single pruning operation on an off-the-shelf pretrained model yields a training-free detector that shows strong generalization to unseen generators, including the Sora evaluation. The paper's strengths are its broad benchmark coverage, the direct test on inaccessible generators (Sora and OpenSora), the clean ablations over backbones, pruning locations, pruning ratios, and pruning strategies, and the explicit threshold-sensitivity analysis. The data-free method is indeed training-free apart from validation-set-based hyperparameter and threshold selection, and the data-driven method's reported gains on several benchmarks are substantial. However, the theoretical contribution is conditional in a way that partly restates the phenomenon, and the empirical claim of universal superiority over prior methods is not supported by all tables. The absence of a non-synthetic out-of-distribution negative set leaves open whether the signal is specific to generated images or is a generic distribution-shift effect.

major comments (4)
  1. [Section 3.3, Assumption 3.3 (Eq. 10) and Proposition 3.4] The formal result does not provide an independent derivation of the detection signal. Assumption 3.3 postulates exactly the phenomenon to be explained: that generated images have larger directional representation sensitivity than natural images along the pruning direction, quantified by omega > 0. Proposition 3.4 then shows that this assumed gap carries over to output differences when omega dominates a higher-order term, but the proof does not establish the assumption or verify condition (12). In the actual method, 90% of the fc2 weights in block 16 are pruned, which is not a small-perturbation regime, and no measurement of S_Dgen - S_Dnat or of the local smoothness constant C_f along the realized pruning direction is provided. Moreover, the paper's own limitation (Appendix A.2) concedes that the gap depends on current natural-dominated pretraining corpora, which is consistent with the assumption being an empirical regularity rather than a robust property. I recommend either removing the formal proposition and presenting Eq. (10) as an empirical observation supported by Figures 2 and 4, or adding direct measurements of the sensitivity gap and validating the local expansion in the deployed pruning regime.
  2. [Section 4.2, Tables 1 and 3] The abstract and Section 4.2 state that the unlearning approach outperforms conventional detectors, but the data-free variant is not consistently better across benchmarks. On ImageNet (Table 1), data-free unlearning averages 92.20 AUROC while FatFormer attains 93.68 and AIDE 93.71. On GenImage (Table 3), data-free unlearning averages 81.8 ACC, below FatFormer (88.9), DRCT (89.4), and SAFE (87.2). Even the data-driven variant is not uniformly superior per generator: for Midjourney it reaches 90.8 versus FatFormer's 92.7 and SAFE's 98.3, and for SD V1.4 it reaches 95.6 versus FatFormer's 100.0. The claims should be qualified to average or selected-metric performance, and because the main tables report no error bars or significance tests, the reported margins need accompanying variability estimates; Appendix A.12 states that results are averaged over five seeds, so these estimates should be reported in the main tables.
  3. [Section 3.3 and Section 4 (OOD natural-image evaluation)] Because Assumption 3.3 frames generated images as distribution-shifted samples, the score s(x) in Eq. (4) may be a general out-of-distribution detector rather than a synthetic-image detector. The paper does not measure false-positive rates on natural images from domains that are also distribution-shifted relative to the pretraining data, such as medical images, aerial images, artwork, or sketches. If such images are forgotten at similar rates, the central claim 'generated images are easier to forget' should be narrowed to 'distribution-shifted images are easier to forget,' and the practical detector would be unreliable for legitimate OOD photographs. I ask for a dedicated negative-set experiment with non-synthetic OOD natural images and a report of the resulting AUROC or ACC, together with a comparison of the score distributions against the generated-image score distributions.
  4. [Section 3.2-3.3, Eq. (4) versus Eq. (8)] The detector uses cosine similarity between original and pruned representations, but Proposition 3.4 is stated for mean squared L2 output differences. A larger L2 displacement does not imply a lower cosine similarity, because the norm of the features can also change; the paper does not provide a mathematical or empirical bridge between Eq. (8) and Eq. (4). Please add a direct analysis of the cosine score, or present the L2 result only as supporting intuition and validate the cosine link empirically, for example by showing that the L2 gap and the cosine gap rank the same images.
minor comments (5)
  1. [Section 4.1 and Appendix A.11] The block numbering is inconsistent: the main text says pruning is applied to the fc2 layer of the 16th transformer block, while Appendix A.11 says 'we prune the weights of block 15' and Table 18 reports results for blocks 16 through 23. Please clarify the indexing convention.
  2. [Eq. (14)] The use of LCE on l2-normalized features, with one normalized vector treated as logits and the other as a soft target, is not a standard cross-entropy operation; please define precisely how the logits and soft targets are formed, including any temperature or smoothing, so that the loss is reproducible.
  3. [Appendix A.13, Table 13] The threshold-sensitivity analysis shows that classification accuracy varies noticeably with the generator used to calibrate the threshold; since the main tables report ACC using thresholds chosen on a ProGAN-based validation set, this caveat should be stated wherever ACC is reported, or threshold-free metrics such as AUROC and AP should be made primary.
  4. [Appendix A.14] There is a typo in the Chameleon paragraph: 'a a very challenging' should be 'a very challenging.'
  5. [Section 3.2] The data-free method is called 'training-free' and 'data-free,' but it uses 1k natural images and ProGAN generated images for hyperparameter and threshold selection; please clarify that no images from the target test generators are used rather than claiming the method has no data dependence at all.

Circularity Check

1 steps flagged · score 4.0 of 10

The theory section restates its key assumption as a proposition, but the detector itself is independently benchmarked on external data.

  1. other [Section 3.3, Assumption 3.3 (Eq. 10) and Proposition 3.4 (Eqs. 11-13); Appendix A.3.1 discussion]
    "Assumption 3.3 ... we assume that generated images exhibit larger directional representation sensitivity than natural images along the pruning direction. Specifically, there exists ω > 0 such that SDgen(Δθ) − SDnat(Δθ) ≥ ω∥Δθ∥2 2. ... Proposition 3.4 formalizes the mechanism behind our unlearning-induced detection signal: pruning amplifies the representation discrepancy whenever generated images are more sensitive to the pruning direction than natural images. ..."

    Proposition 3.4 is not an independent derivation that generated images are easier to forget; it is Assumption 3.3 transported from Jacobian-level directional sensitivity to output-level difference via a Taylor expansion (Eqs. 15-22). The assumption itself encodes the very gap the detector exploits, and the paper says it is motivated by prior observations and empirically supported by the same Figures 2 and 4 used to motivate the method. Thus the theoretical 'prediction' is a restatement of the input assumption rather than a first-principles result. The empirical detector, however, is still validated on external benchmarks, so the circularity is confined to the theoretical framing.

full rationale

The main empirical claim — that pruning a pretrained LVM yields a score separating natural from generated images — is tested on multiple external benchmarks (ImageNet, GenImage, Chameleon, DRCT-2M, LSUN-BEDROOM, DiffusionForensics, and even Sora/OpenSora), so it is not a fitted-input-called-prediction artifact. Hyperparameters (pruning block, ratio, margin, threshold) are selected on a small ProGAN/ImageNet validation set, but the headline AUROC/AP results are threshold-free and evaluated on unseen generators, so this is ordinary model selection rather than circularity. The paper's self-citations (Nie et al. 2025; Zhang et al. 2025) appear only in related-work remarks and are not load-bearing. The one genuine circular step is in the theory: Assumption 3.3 postulates the directional sensitivity gap that is the phenomenon being exploited, and Proposition 3.4 merely restates that gap at the output level under a smoothness condition. The appendix itself concedes the gap is empirical, noting it may weaken as synthetic images enter pretraining corpora (A.2), and that the assumption is 'motivated by prior observations' and 'empirically supported' by the same figures that motivate the method. This makes the theoretical characterization conditional rather than derivational, but it does not undermine the independent experimental evaluation of the detector. A separate rigor concern, not circularity, is that the proof's condition ω > C_f ||Δθ||_2 is asserted for the actual 90%-pruning regime rather than verified; that belongs in correctness risk, not in the circularity score.

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

No new physical entities are introduced. The method depends on several fitted hyperparameters and on an assumption that generated images are more sensitive to pruning; the theory propagates that assumption rather than deriving it. The external benchmark results provide independent support.

free parameters (4)
  • Pruning ratio and location = 90% of smallest-magnitude weights in fc2 layer of block 16 of DINOv2 ViT-L/14
    Selected using 1k ImageNet natural images and 1k ProGAN generated images (Section 4.1 and A.12). Figure 6 shows robustness, but the central detector depends on this chosen configuration.
  • Classification threshold for ACC = 0.94287 for data-free; 0.90178 for data-driven
    Chosen on a validation set of ProGAN generated and ImageNet natural images (A.12). ACC numbers depend on this fitted threshold; AUROC and AP do not. Table 13 shows substantial threshold sensitivity.
  • Margin gamma = 20
    Hand-set in the data-driven loss Eq. 14; Figure 9 shows robustness to gamma, but it is not derived.
  • LoRA hyperparameters = r=8, alpha=8, applied to q_proj and v_proj of blocks 18-20, 3 epochs, lr=1e-5
    Chosen for optimization of data-driven unlearning (A.12); Table 7 ablates rank and alpha, but the configuration is not derived from first principles.
assumptions (4)
  • ad hoc to paper Assumption 3.3: generated images have larger directional representation sensitivity than natural images along the pruning direction (Eq. 10).
    This is the load-bearing premise of Proposition 3.4 and of the data-free detector. It is motivated by prior work and by Figures 2 and 4, but not proven; it essentially states the phenomenon the paper exploits.
  • standard math The representation f(x;theta) is twice differentiable with locally bounded second derivative and locally bounded Jacobian in a neighborhood of theta.
    Used in the proof of Proposition 3.4 (Eqs. 15-19) to bound Taylor remainders; this is a standard smoothness assumption.
  • domain assumption Pre-training corpora of DINOv2 and CLIP are dominated by natural images, so generated images behave as rare or distribution-shifted samples.
    Used in Section 3.2 to connect Hooker et al.'s compression findings to generated images. The paper's own limitation in A.2 says the method weakens if future pretraining data include many synthetic images.
  • domain assumption The validation set of ImageNet natural images and ProGAN generated images is representative enough to select hyperparameters and thresholds that transfer to other test generators.
    Hyperparameters and ACC thresholds are chosen on this small validation set; Table 13 shows accuracy varies with threshold, so transfer to new datasets is not guaranteed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generated Images Are Easier to Forget: A Machine Unlearning Perspective for Synthetic Image Detection." pith.science (2026). https://pith.science/paper/LNNHWWST

@misc{pith2026260800716,
  author       = {Pith},
  title        = {Pith review of: Generated Images Are Easier to Forget: A Machine Unlearning Perspective for Synthetic Image Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LNNHWWST}},
  note         = {Machine review of arXiv:2608.00716}
}
read the original abstract

Robust detection of generated images is critical to counter the misuse of generative models. Existing methods primarily depend on learning from human-annotated training datasets, limiting their generalization to unseen distributions. In contrast, large-scale vision models (LVMs) pre-trained on web-scale datasets exhibit exceptional generalization power through exposure to diverse distributions, offering a transformative paradigm for this task. However, our experimental results reveal that LVMs pre-trained on natural-image-dominated data can effectively capture the features of both natural and generated images, yielding comparably low losses and thus limited discriminative capacity between them. This prompts a key question: When and how do LVMs exhibit different behaviors when capturing features of natural and generated images? This investigation reveals an insight: during unlearning, LVMs exhibit disparate forgetting dynamics with feature degradation for generated images escalating faster than natural ones. Inspired by the disparate dynamics, we introduce two detection methods: 1) data-free detection, which prunes model parameters to induce unlearning without data access, and 2) data-driven detection, which optimizes LVMs to unlearn knowledge tied to generated images. Extensive experiments conducted on various benchmarks demonstrate that our unlearning-based approach outperforms conventional detection methods. By recasting the detection task as a problem of machine unlearning, our work establishes a new paradigm for generated image detection.

Figures

Figures reproduced from arXiv: 2608.00716 by the authors.

Figure 1
Figure 1. LVMs pre-trained on natural images exhibit low losses for both natural and gener￾ated images, thus restricting their discrimina￾tive capacity between the types of images. Regarding the ability of generalization, pre-trained vision models (LVMs) have shown great success in various domains thanks to their large-scale training with extensive data (Radford et al., 2021). Thus, LVMs emerge as promising candidates for pro… view at source ↗
Figure 2
Figure 2. Dynamic illustration of unlearn￾ing. As the model transitions from learned to unlearned state, the feature extraction ability of the model shows a clear discrepancy be￾tween natural and generated images. We com￾pute feature similarity of images on unlearned model and original model, using t-SNE to vi￾sualize feature distribution of images on mod￾els with different unlearning levels. To investigate this fundamental q… view at source ↗
Figure 3
Figure 3. Differences and connections between learning-based detection and unlearning-based [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: The feature shift caused by weight perturbation is more significant in the generated images. Eq. 3 presents a general machine unlearning framework that requires collecting both natural and generated im￾ages to fine-tune the model, incurring additional com￾putational co…
Figure 6
Figure 6. Figure 6: The effect of pruning ratio [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Comparison of feature similarity on learned and unlearned models. The generated images [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 9
Figure 9. Figure 9: The effect of parameter γ. Robustness to image corruptions. In addition to the performance on clean images, the robustness of the detector to various image corruptions is also an important metric. In reality, images may be exposed to various perturbations, e.g., when p…
Figure 8
Figure 8. Figure 8: Robustness to perturbations: (a) JPEG compression; (b) Gaussian blur; (c) Gaussian noise. [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 10
Figure 10. Figure 10: The effect of structured pruning. We obtain an unlearned model by completely dropping a full block. As shown in [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 9
Figure 9. Figure 9: Following CNNspot (Wang et al., 2020), data augmentation techniques including JPEG [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 47 canonical work pages

  1. [1]

    A., Jia, H., Travers, A., Zhang, B., Lie, D., and Papernot, N

    Bourtoule, L., Chandrasekaran, V., Choquette-Choo, C. A., Jia, H., Travers, A., Zhang, B., Lie, D., and Papernot, N. (2021). Machine unlearning. In 2021 IEEE symposium on security and privacy (SP)

    work page 2021

  2. [2]

    Brock, A., Donahue, J., and Simonyan, K. (2019). Large scale GAN training for high fidelity natural image synthesis. In International Conference on Learning Representations, ICLR

    work page 2019

  3. [3]

    Cai, Q., Wu, C., Zhang, Y., Yu, J., and Tian, X. (2025). Towards generalizable detector for generated image. Advances in Neural Information Processing Systems , 38:94773--94800

    work page 2025

  4. [4]

    Caron, M., Misra, I., Mairal, J., Goyal, P., Bojanowski, P., and Joulin, A. (2020). Unsupervised learning of visual features by contrasting cluster assignments. In Advances in Neural Information Processing Systems, NeurIPS

    work page 2020

  5. [5]

    Caron, M., Touvron, H., Misra, I., J \' e gou, H., Mairal, J., Bojanowski, P., and Joulin, A. (2021). Emerging properties in self-supervised vision transformers. In IEEE/CVF International Conference on Computer Vision, ICCV

    work page 2021

  6. [6]

    Chen, B., Zeng, J., Yang, J., and Yang, R. (2024). DRCT: diffusion reconstruction contrastive training towards universal detection of diffusion generated images. In International Conference on Machine Learning, ICML

    work page 2024

  7. [7]

    Chen, R., Xi, J., Yan, Z., Zhang, K.-Y., Wu, S., Xie, J., Chen, X., Xu, L., Guan, I., Yao, T., et al. (2025). Dual data alignment makes ai-generated image detector easier generalizable. arXiv preprint arXiv:2505.14359

    arXiv 2025

  8. [8]

    Deng, J., Dong, W., Socher, R., Li, L., Li, K., and Fei - Fei, L. (2009). Imagenet: A large-scale hierarchical image database. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    work page 2009

Show all 54 references
  1. [9]

    and Nichol, A

    Dhariwal, P. and Nichol, A. Q. (2021). Diffusion models beat gans on image synthesis. In Advances in Neural Information Processing Systems, NeurIPS

    2021

  2. [10]

    Durall, R., Keuper, M., and Keuper, J. (2020). Watch your up-convolution: CNN based generative deep neural networks are failing to reproduce spectral distributions. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2020

  3. [11]

    Frank, J., Eisenhofer, T., Sch \"o nherr, L., Fischer, A., Kolossa, D., and Holz, T. (2020a). Leveraging frequency analysis for deep fake image recognition. In International conference on machine learning , pages 3247--3258. PMLR

    2020

  4. [12]

    Frank, J., Eisenhofer, T., Sch \" o nherr, L., Fischer, A., Kolossa, D., and Holz, T. (2020b). Leveraging frequency analysis for deep fake image recognition. In International Conference on Machine Learning, ICML

    2020

  5. [13]

    Golatkar, A., Achille, A., Ravichandran, A., Polito, M., and Soatto, S. (2021). Mixed-privacy forgetting in deep networks. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR

    2021

  6. [14]

    Golatkar, A., Achille, A., and Soatto, S. (2020). Eternal sunshine of the spotless net: Selective forgetting in deep networks. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2020

  7. [15]

    Han, S., Pool, J., Tran, J., and Dally, W. J. (2015). Learning both weights and connections for efficient neural network. In Advances in Neural Information Processing Systems, NeurIPS

    2015

  8. [16]

    He, K., Fan, H., Wu, Y., Xie, S., and Girshick, R. B. (2020). Momentum contrast for unsupervised visual representation learning. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2020

  9. [17]

    Ho, J., Jain, A., and Abbeel, P. (2020). Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, NeurIPS

    2020

  10. [18]

    Hooker, S., Courville, A., Clark, G., Dauphin, Y., and Frome, A. (2019). What do compressed deep neural networks forget? arXiv preprint arXiv:1911.05248

    2019 arXiv

  11. [19]

    J., Shen, Y., Wallis, P., Allen - Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W

    Hu, E. J., Shen, Y., Wallis, P., Allen - Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W. (2022). Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations, ICLR

    2022

  12. [20]

    Jia, J., Liu, J., Ram, P., Yao, Y., Liu, G., Liu, Y., Sharma, P., and Liu, S. (2023). Model sparsity can simplify machine unlearning. In Advances in Neural Information Processing Systems, NeurIPS

    2023

  13. [21]

    Ju, Y., Jia, S., Ke, L., Xue, H., Nagano, K., and Lyu, S. (2022). Fusing global and local features for generalized ai-synthesized image detection. In International Conference on Image Processing, ICIP

    2022

  14. [22]

    Karras, T., Laine, S., and Aila, T. (2019). A style-based generator architecture for generative adversarial networks. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2019

  15. [23]

    Langley, P. (2000). Crafting papers on machine learning. In Langley, P., editor, Proceedings of the 17th International Conference on Machine Learning (ICML 2000) , pages 1207--1216, Stanford, CA. Morgan Kaufmann

    2000

  16. [24]

    Li, O., Cai, J., Hao, Y., Jiang, X., Hu, Y., and Feng, F. (2025). 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, KDD 2025, Toronto, O...

    2025

  17. [25]

    Liu, B., Yang, F., Bi, X., Xiao, B., Li, W., and Gao, X. (2022). Detecting generated images by real images. In Proceedings of the European Conference on Computer Vision, ECCV

    2022

  18. [26]

    Liu, H., Tan, Z., Tan, C., Wei, Y., Wang, J., and Zhao, Y. (2024a). Forgery-aware adaptive transformer for generalizable synthetic image detection. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR

    2024

  19. [27]

    Liu, H., Tan, Z., Tan, C., Wei, Y., Zhao, Y., and Wang, J. (2024b). Forgery-aware adaptive transformer for generalizable synthetic image detection. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2024

  20. [28]

    Liu, Z., Qi, X., and Torr, P. H. S. (2020). Global texture enhancement for fake face detection in the wild. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2020

  21. [29]

    Luo, Y., Du, J., Yan, K., and Ding, S. (2024). Lare \^ 2: Latent reconstruction error based method for diffusion-generated image detection. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2024

  22. [30]

    Mandelli, S., Bonettini, N., Bestagini, P., and Tubaro, S. (2022). Detecting gan-generated images by orthogonal training of multiple cnns. In International Conference on Image Processing, ICIP

    2022

  23. [31]

    P., Oikawa, R., Divakaran, D

    Nguyen, Q. P., Oikawa, R., Divakaran, D. M., Chan, M. C., and Low, B. K. H. (2022). Markov chain monte carlo-based machine unlearning: Unlearning what needs to be forgotten. In ASIA CCS '22: ACM Asia Conference on Computer and Communications Security

    2022

  24. [32]

    Nie, J., Zhang, Y., Liu, T., Cheung, Y.-m., Han, B., and Tian, X. (2025). Epistemic uncertainty for generated image detection. Advances in Neural Information Processing Systems , 38:101030--101057

    2025

  25. [33]

    Ojha, U., Li, Y., and Lee, Y. J. (2023). Towards universal fake image detectors that generalize across generative models. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2023

  26. [34]

    Sora: Creating video from text

    OpenAI (2024). Sora: Creating video from text

    2024

  27. [35]

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H. V., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El - Nouby, A., Assran, M., Ballas, N., Galuba, W., Howes, R., Huang, P., Li, S., Misra, I., Rabbat, M., Sharma, V., Synnaeve, G., Xu, H., J \' e gou, H., Mair...

    2024

  28. [36]

    Qian, Y., Yin, G., Sheng, L., Chen, Z., and Shao, J. (2020). Thinking in frequency: Face forgery detection by mining frequency-aware clues. In Proceedings of the European Conference on Computer Vision, ECCV

    2020

  29. [37]

    W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., and Sutskever, I

    Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., and Sutskever, I. (2021). Learning transferable visual models from natural language supervision. In International Conference on Machine Learning, ICML

    2021

  30. [38]

    Ricker, J., Lukovnikov, D., and Fischer, A. (2024). AEROBLADE: training-free detection of latent diffusion images using autoencoder reconstruction error. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2024

  31. [39]

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. (2022). High-resolution image synthesis with latent diffusion models. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2022

  32. [40]

    Schuhmann, C., Vencu, R., Beaumont, R., Kaczmarczyk, R., Mullis, C., Katta, A., Coombes, T., Jitsev, J., and Komatsuzaki, A. (2021). Laion-400m: Open dataset of clip-filtered 400 million image-text pairs. arXiv preprint arXiv:2111.02114

    2021 arXiv

  33. [41]

    and Yamasaki, T

    Shiohara, K. and Yamasaki, T. (2022). Detecting deepfakes with self-blended images. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2022

  34. [42]

    C., Hosseinzadeh, R., Sui, Y., Ross, B

    Stein, G., Cresswell, J. C., Hosseinzadeh, R., Sui, Y., Ross, B. L., Villecroze, V., Liu, Z., Caterini, A. L., Taylor, J. E. T., and Loaiza - Ganem, G. (2023). Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion models. In Advances in ...

    2023

  35. [43]

    Tan, C., Liu, H., Zhao, Y., Wei, S., Gu, G., Liu, P., and Wei, Y. (2024). Rethinking the up-sampling operations in cnn-based generative network for generalizable deepfake detection. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2024

  36. [44]

    Tan, C., Zhao, Y., Wei, S., Gu, G., and Wei, Y. (2023). Learning on gradients: Generalized artifacts representation for gan-generated images detection. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2023

  37. [45]

    Wang, S., Wang, O., Zhang, R., Owens, A., and Efros, A. A. (2020). Cnn-generated images are surprisingly easy to spot... for now. In The IEEE / CVF Computer Vision and Pattern Recognition Conference, CVPR

    2020

  38. [46]

    Wang, Z., Bao, J., Zhou, W., Wang, W., Hu, H., Chen, H., and Li, H. (2023). DIRE for diffusion-generated image detection. In IEEE/CVF International Conference on Computer Vision, ICCV

    2023

  39. [47]

    Yan, S., Li, O., Cai, J., Hao, Y., Jiang, X., Hu, Y., and Xie, W. (2024). A sanity check for ai-generated image detection. arXiv preprint arXiv:2406.19435

    2024 arXiv

  40. [48]

    Yu, F., Zhang, Y., Song, S., Seff, A., and Xiao, J. (2015). LSUN: construction of a large-scale image dataset using deep learning with humans in the loop. CoRR

    2015

  41. [49]

    Zhang, X., Karaman, S., and Chang, S. (2019). Detecting and simulating artifacts in GAN fake images. In International Workshop on Information Forensics and Security, WIFS

    2019

  42. [50]

    Zhang, Y., Nie, J., Tian, X., Gong, M., Zhang, K., and Han, B. (2025). Detecting generated images by fitting natural image distributions. Advances in Neural Information Processing Systems , 38:31337--31361

    2025

  43. [51]

    Zheng, Z., Peng, X., Yang, T., Shen, C., Li, S., Liu, H., Zhou, Y., Li, T., and You, Y. (2024). Open-sora: Democratizing efficient video production for all

    2024

  44. [52]

    Zhong, N., Xu, Y., Qian, Z., and Zhang, X. (2023). Rich and poor texture contrast: A simple yet effective approach for ai-generated image detection. arXiv preprint arXiv:2311.12397

    2023 arXiv

  45. [53]

    Zhu, M., Chen, H., Huang, M., Li, W., Hu, H., Hu, J., and Wang, Y. (2023a). Gendet: Towards good generalizations for ai-generated image detection. arXiv preprint arXiv:2312.08880

    2023 arXiv

  46. [54]

    Zhu, M., Chen, H., Yan, Q., Huang, X., Lin, G., Li, W., Tu, Z., Hu, H., Hu, J., and Wang, Y. (2023b). Genimage: A million-scale benchmark for detecting ai-generated image. In Advances in Neural Information Processing Systems, NeurIPS

    2023

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.