pith. sign in

arxiv: 2607.02038 · v1 · pith:OMEUW2KAnew · submitted 2026-07-02 · 💻 cs.CV

Hierarchical Anti-Aesthetics: Protecting Facial Privacy against Customized Diffusion Models

Songping Wang , Yueming Lyu , Shiqi Liu , Chen Zhao , Ziyuan Chen , Ning Li , Jing Dong , Caifeng Shan This is my paper

Pith reviewed 2026-07-03 16:01 UTC · model grok-4.3

classification 💻 cs.CV
keywords facial privacyanti-aestheticsdiffusion modelsidentity protectionadversarial perturbationscustomized generationimage quality degradationprivacy protection
0
0 comments X

The pith

The Hierarchical Anti-Aesthetics framework protects facial privacy by degrading the quality of images from customized diffusion models at global and local levels.

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

Customized diffusion models enable personalized image creation but create risks of facial identity exposure through malicious use. The paper establishes that image aesthetics correlate with human perception of quality, so deliberately lowering aesthetics can limit identity leakage in generated outputs. It introduces the Hierarchical Anti-Aesthetics framework with a global branch that applies a reward mechanism and loss to reduce overall aesthetics and generation quality, plus a local branch that targets facial regions with adversarial perturbations via similar local mechanisms. Integration of the branches produces multi-level degradation during the customization process. Experiments demonstrate that this yields stronger identity removal than prior approaches.

Core claim

The paper claims that the Hierarchical Anti-Aesthetics (HAA) framework, built from global and local anti-aesthetic branches each using dedicated reward mechanisms and losses, reduces facial identity leakage by degrading overall and region-specific aesthetics in images produced by customized diffusion models.

What carries the argument

The Hierarchical Anti-Aesthetics (HAA) framework consisting of a Global Anti-Aesthetics branch and a Local Anti-Aesthetics branch, each driven by an anti-aesthetic reward mechanism and corresponding loss to direct degradation.

Load-bearing premise

Degrading aesthetic quality at both global and local perceptual levels will reliably reduce facial identity leakage in images from customized diffusion models.

What would settle it

A controlled test in which customized diffusion models trained on HAA-protected images still output faces with high identity similarity scores even after aesthetic quality has been measurably lowered.

Figures

Figures reproduced from arXiv: 2607.02038 by Caifeng Shan, Chen Zhao, Jing Dong, Ning Li, Shiqi Liu, Songping Wang, Yueming Lyu, Ziyuan Chen.

Figure 1
Figure 1. Figure 1: Previous image-protection methods overlook aesthetic cues, which limits their ability to remove identity and inevitably leads to privacy leakage. In contrast, our method effectively enhances the ability to eliminate facial identity, guided by the proposed anti-aesthetic mechanisms. The prompt is “a dslr portrait of sks person”. Malicious users may exploit these models to generate forged images or deepfake … view at source ↗
Figure 2
Figure 2. Figure 2: Compared with general training, aesthetic learning enhances the quality and details of generated images by aligning with human aesthetic preferences. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The framework of our proposed HAA. It is an iterative training process where the adversarial noise and the parameters of the surrogate customized [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FID value varies with different λ and β values defined in Eq. 14 on CelebA-HQ. Baseline GAA LAA HAA 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 (A) FDSR Baseline GAA LAA HAA 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 (B) Face Similarity Baseline GAA LAA HAA 0.8 0.6 0.4 0.2 0.0 0.2 (C) Image Reward Baseline GAA LAA HAA 0 100 200 300 400 (D) FID [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Effectiveness Analysis of HAA Components on CelebA-HQ. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: CLIP cosine similarity after masking detected face regions. Lower values indicate less residual non-face identity-related information. Clean Base GAA LAA HAA [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Grad-CAM response visualization for HAA components. Brighter regions indicate stronger optimization pressure for suppressing identity￾supporting information. We first mask all detected face regions and compute CLIP image-embedding cosine similarity between the masked gen￾erated images and clean references. Clean denotes unprotected training, and Base denotes reconstruction-loss-only pertur￾bation without a… view at source ↗
Figure 8
Figure 8. Figure 8: Face-detection-failure stress test. The 0% and 60% settings denote valid-local-detection availability for LAA. Adding GAA consistently improves LAA when local detections are unavailable or partially available. TABLE VII EFFECTIVENESS STUDIES ACROSS VARYING NOISE BUDGETS ON CELEBA-HQ. η FDSR ↓ Face Similarity ↓ Image Reward ↓ FID ↑ 0.00 1.000 0.498 0.599 112.3 0.02 0.425 0.220 -0.428 379.8 0.05 0.281 0.117 … view at source ↗
Figure 9
Figure 9. Figure 9: The relationship between Anti-Aesthetic Score, Face Similarity [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Comparative visualization results on the VGGFace2 dataset. Two [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Visualization of HAA’s protection effectiveness under extreme scenarios such as occlusion, rotation, and motion blur. (A) represents the original [PITH_FULL_IMAGE:figures/full_fig_p013_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Qualitative LPIPS-controlled comparison and ArcFace Grad-CAM analysis. The first two rows compare generic degradations and HAA. The third row shows ArcFace embedding Grad-CAM responses with respect to the original identity; brighter regions indicate stronger identity evidence, while weaker responses indicate stronger de-identification. TABLE XI REWARD-CONTROL STUDY FOR AESTHETIC-MECHANISM-DRIVEN OPTIMIZAT… view at source ↗
Figure 14
Figure 14. Figure 14: Identity-specificity diagnostics of RMl . Top: different identities with similar aesthetic quality, where each tile reports the RMl score / absolute difference from the Aesthetic Scorer. Bottom: same-identity groups under background and expression changes, where each tile reports the mean / standard deviation. indicating stable scoring across non-identity variations. These diagnostics indicate that RMl’s … view at source ↗
Figure 15
Figure 15. Figure 15: Robustness analysis of the global reward model RMg. Under Gaussian Blur, Gaussian Noise, Defocus Blur, and Salt-Pepper Noise, the global aesthetic score consistently decreases as degradation intensity increases [PITH_FULL_IMAGE:figures/full_fig_p015_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Robustness analysis of the local reward model RMl . Under the same four degradation types, the face-local aesthetic score consistently decreases with stronger degradation. S. More details Both reward models use a BLIP-style preference-scorer ar￾chitecture with a ViT-L image encoder, a 12-layer transformer text encoder, cross-modal fusion, and an MLP scalar head. For RMg, we use the public ImageRewardDB da… view at source ↗
read the original abstract

The rise of customized diffusion models has fueled a boom in personalized visual content creation, but it also introduces serious risks of malicious misuse, thereby posing threats to personal privacy. Image aesthetics are strongly correlated with human perception of image quality. Motivated by this observation, we address facial privacy protection from a novel aesthetic perspective by degrading the generation quality of maliciously customized models, thus reducing facial identity leakage. Specifically, we propose a Hierarchical Anti-Aesthetics (HAA) framework that exploits aesthetic cues at multiple perceptual levels. HAA consists of two key branches: (1) Global Anti-Aesthetics, which degrades overall aesthetics and generation quality by constructing a global anti-aesthetic reward mechanism and a corresponding loss; and (2) Local Anti-Aesthetics, which disrupts facial identity by using a local anti-aesthetic reward mechanism and loss to guide adversarial perturbations toward facial regions. By integrating both branches, HAA achieves anti-aesthetic degradation from a global to a local level during customized generation. Extensive experiments show that HAA outperforms existing methods in identity removal, providing an effective tool for protecting facial privacy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes Hierarchical Anti-Aesthetics (HAA), a two-branch framework for facial privacy protection against customized diffusion models. Global Anti-Aesthetics constructs a reward mechanism and loss to degrade overall image aesthetics and generation quality; Local Anti-Aesthetics applies a separate reward and loss to drive adversarial perturbations specifically into facial regions. The central claim is that integrating the branches produces hierarchical anti-aesthetic degradation that reduces facial identity leakage, with extensive experiments asserted to show outperformance over prior methods.

Significance. If the causal link between aesthetic degradation and identity removal is validated, the work would introduce a new aesthetic-based axis for adversarial defense in personalized generative models. It could inform privacy tools that operate without direct access to model weights. The absence of any parameter-free derivation, machine-checked proof, or falsifiable prediction in the presented material limits the immediate technical contribution.

major comments (2)
  1. [Abstract] Abstract: the claim that 'HAA outperforms existing methods in identity removal' is stated without reference to any quantitative metrics, baselines, datasets, or ablation results. Because this is the sole empirical support for the central privacy-protection claim, the absence of evidence prevents assessment of whether the method actually succeeds.
  2. [Abstract] Abstract: the motivation equates correlation ('Image aesthetics are strongly correlated with human perception of image quality') with the causal claim that deliberately lowering aesthetic scores will preferentially disrupt identity manifolds rather than merely reducing perceptual quality metrics. No derivation, auxiliary experiment, or analysis is supplied to show that identity-specific features are more sensitive to the proposed global/local anti-aesthetic losses than other structural cues.
minor comments (1)
  1. [Abstract] The abstract introduces 'global anti-aesthetic reward mechanism' and 'local anti-aesthetic reward mechanism' without even a one-sentence definition or reference to the equations that implement them, making the high-level description difficult to follow.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed comments on the abstract. We respond to each major comment below and indicate where revisions will be made to improve clarity and support for the claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'HAA outperforms existing methods in identity removal' is stated without reference to any quantitative metrics, baselines, datasets, or ablation results. Because this is the sole empirical support for the central privacy-protection claim, the absence of evidence prevents assessment of whether the method actually succeeds.

    Authors: The abstract is intended as a high-level summary; the quantitative metrics, baselines (e.g., existing adversarial and privacy-protection methods), datasets, and ablation studies are presented in detail in the experiments section of the full manuscript. To directly address the concern and allow readers to assess the claim from the abstract itself, we will revise the abstract to include specific quantitative results, such as identity similarity scores or removal rates relative to baselines on standard facial datasets. revision: yes

  2. Referee: [Abstract] Abstract: the motivation equates correlation ('Image aesthetics are strongly correlated with human perception of image quality') with the causal claim that deliberately lowering aesthetic scores will preferentially disrupt identity manifolds rather than merely reducing perceptual quality metrics. No derivation, auxiliary experiment, or analysis is supplied to show that identity-specific features are more sensitive to the proposed global/local anti-aesthetic losses than other structural cues.

    Authors: The manuscript is an empirical study motivated by the established correlation between aesthetics and perceived quality; the hierarchical losses are designed and validated through experiments to reduce identity leakage in customized diffusion outputs. No theoretical derivation or formal proof of preferential sensitivity of identity features is provided, as the contribution centers on the practical effectiveness of the two-branch framework rather than a causal mechanistic analysis. We will add a brief clarifying sentence in the introduction or discussion to explicitly note the empirical basis of the approach. revision: partial

Circularity Check

0 steps flagged

No circularity: method proposal is self-contained and independent of its motivational assumption.

full rationale

The paper motivates HAA from the observed correlation between aesthetics and perceived quality, then defines global and local anti-aesthetic reward mechanisms plus losses to degrade generation. This construction does not reduce to the correlation by definition, nor does any equation or branch rename a fitted input as a prediction. No self-citation load-bearing steps, uniqueness theorems, or ansatzes imported from prior author work appear in the abstract or described framework. The derivation chain consists of an independent engineering proposal whose effectiveness is tested experimentally rather than forced by its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that aesthetics correlate with perceived quality and identity leakage. No free parameters, additional axioms, or invented entities are specified in the abstract.

axioms (1)
  • domain assumption Image aesthetics are strongly correlated with human perception of image quality.
    Explicitly stated as motivation in the first sentence of the abstract.

pith-pipeline@v0.9.1-grok · 5736 in / 1148 out tokens · 37783 ms · 2026-07-03T16:01:50.374053+00:00 · methodology

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

45 extracted references · 9 canonical work pages · 5 internal anchors

  1. [1]

    Vector quantized diffusion model for text-to-image synthesis,

    S. Gu, D. Chen, J. Bao, F. Wen, B. Zhang, D. Chen, L. Yuan, and B. Guo, “Vector quantized diffusion model for text-to-image synthesis,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10 696–10 706

    2022

  2. [2]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10 684–10 695

    2022

  3. [3]

    Hierarchical Text-Conditional Image Generation with CLIP Latents

    A. Ramesh, P. Dhariwal, A. Nichol, C. Chu, and M. Chen, “Hierarchical text-conditional image generation with clip latents,”arXiv preprint arXiv:2204.06125, vol. 1, no. 2, p. 3, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  4. [4]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems, vol. 33, pp. 6840– 6851, 2020

    2020

  5. [5]

    Denoising Diffusion Implicit Models

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” arXiv preprint arXiv:2010.02502, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  6. [6]

    Text2live: Text-driven layered image and video editing,

    O. Bar-Tal, D. Ofri-Amar, R. Fridman, Y . Kasten, and T. Dekel, “Text2live: Text-driven layered image and video editing,” inEuropean conference on computer vision. Springer, 2022, pp. 707–723

    2022

  7. [7]

    Diffusionclip: Text-guided diffusion models for robust image manipulation,

    G. Kim, T. Kwon, and J. C. Ye, “Diffusionclip: Text-guided diffusion models for robust image manipulation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 2426–2435

    2022

  8. [8]

    An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion

    R. Gal, Y . Alaluf, Y . Atzmon, O. Patashnik, A. H. Bermano, G. Chechik, and D. Cohen-Or, “An image is worth one word: Person- alizing text-to-image generation using textual inversion,”arXiv preprint arXiv:2208.01618, 2022. MANUSCRIPT FOR IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY 16

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  9. [9]

    Dreambooth: Fine tuning text-to-image diffusion models for subject- driven generation,

    N. Ruiz, Y . Li, V . Jampani, Y . Pritch, M. Rubinstein, and K. Aberman, “Dreambooth: Fine tuning text-to-image diffusion models for subject- driven generation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 22 500–22 510

    2023

  10. [10]

    Multi- concept customization of text-to-image diffusion,

    N. Kumari, B. Zhang, R. Zhang, E. Shechtman, and J.-Y . Zhu, “Multi- concept customization of text-to-image diffusion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 1931–1941

    2023

  11. [11]

    Svdiff: Compact parameter space for diffusion fine-tuning,

    L. Han, Y . Li, H. Zhang, P. Milanfar, D. Metaxas, and F. Yang, “Svdiff: Compact parameter space for diffusion fine-tuning,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 7323–7334

    2023

  12. [12]

    Bagm: A backdoor attack for manipulating text-to-image generative models,

    J. Vice, N. Akhtar, R. Hartley, and A. Mian, “Bagm: A backdoor attack for manipulating text-to-image generative models,”IEEE Transactions on Information Forensics and Security, 2024

    2024

  13. [13]

    Personalization as a shortcut for few-shot backdoor attack against text-to-image diffusion models,

    Y . Huang, F. Juefei-Xu, Q. Guo, J. Zhang, Y . Wu, M. Hu, T. Li, G. Pu, and Y . Liu, “Personalization as a shortcut for few-shot backdoor attack against text-to-image diffusion models,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 19, 2024, pp. 21 169– 21 178

    2024

  14. [14]

    Leveraging frequency analysis for deep fake image recogni- tion,

    J. Frank, T. Eisenhofer, L. Sch ¨onherr, A. Fischer, D. Kolossa, and T. Holz, “Leveraging frequency analysis for deep fake image recogni- tion,” inInternational conference on machine learning. PMLR, 2020, pp. 3247–3258

    2020

  15. [15]

    A comprehensive overview of deepfake: Generation, detection, datasets, and opportuni- ties,

    J. W. Seow, M. K. Lim, R. C. Phan, and J. K. Liu, “A comprehensive overview of deepfake: Generation, detection, datasets, and opportuni- ties,”Neurocomputing, vol. 513, pp. 351–371, 2022

    2022

  16. [16]

    Deepfake detection by analyz- ing convolutional traces,

    L. Guarnera, O. Giudice, and S. Battiato, “Deepfake detection by analyz- ing convolutional traces,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2020, pp. 666– 667

    2020

  17. [17]

    Understanding and im- proving adversarial attacks on latent diffusion model,

    B. Zheng, C. Liang, X. Wu, and Y . Liu, “Understanding and im- proving adversarial attacks on latent diffusion model,”arXiv preprint arXiv:2310.04687, 2023

    work page arXiv 2023

  18. [18]

    Anti-dreambooth: Protecting users from personalized text-to-image synthesis,

    T. Van Le, H. Phung, T. H. Nguyen, Q. Dao, N. N. Tran, and A. Tran, “Anti-dreambooth: Protecting users from personalized text-to-image synthesis,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 2116–2127

    2023

  19. [19]

    Perturbing attention gives you more bang for the buck: Subtle imaging perturbations that efficiently fool customized diffusion models,

    J. Xu, Y . Lu, Y . Li, S. Lu, D. Wang, and X. Wei, “Perturbing attention gives you more bang for the buck: Subtle imaging perturbations that efficiently fool customized diffusion models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 24 534–24 543

    2024

  20. [20]

    Simac: a simple anti-customization method for protecting face privacy against text-to- image synthesis of diffusion models,

    F. Wang, Z. Tan, T. Wei, Y . Wu, and Q. Huang, “Simac: a simple anti-customization method for protecting face privacy against text-to- image synthesis of diffusion models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 12 047–12 056

    2024

  21. [21]

    Harnessing global- local collaborative adversarial perturbation for anti-customization,

    L. Xu, J. Wang, H. Hao, H. Qin, J. Zhao, and X. Liu, “Harnessing global- local collaborative adversarial perturbation for anti-customization,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp. 13 414–13 423

    2025

  22. [22]

    Image quality and the aesthetic judgment of photographs: Contrast, sharpness, and grain teased apart and put together

    P. P. Tinio, H. Leder, and M. Strasser, “Image quality and the aesthetic judgment of photographs: Contrast, sharpness, and grain teased apart and put together.”Psychology of Aesthetics, Creativity, and the Arts, vol. 5, no. 2, p. 165, 2011

    2011

  23. [23]

    Visual aesthetics and human preference,

    S. E. Palmer, K. B. Schloss, and J. Sammartino, “Visual aesthetics and human preference,”Annual review of psychology, vol. 64, no. 1, pp. 77–107, 2013

    2013

  24. [24]

    Rating image aesthetics using deep learning,

    X. Lu, Z. Lin, H. Jin, J. Yang, and J. Z. Wang, “Rating image aesthetics using deep learning,”IEEE Transactions on Multimedia, vol. 17, no. 11, pp. 2021–2034, 2015

    2021

  25. [25]

    Personalizing text-to-image generation via aesthetic gradi- ents,

    V . Gallego, “Personalizing text-to-image generation via aesthetic gradi- ents,”arXiv preprint arXiv:2209.12330, 2022

    work page arXiv 2022

  26. [26]

    Vmix: Improving text- to-image diffusion model with cross-attention mixing control,

    S. Wu, F. Ding, M. Huang, W. Liu, and Q. He, “Vmix: Improving text- to-image diffusion model with cross-attention mixing control,”arXiv preprint arXiv:2412.20800, 2024

    work page arXiv 2024

  27. [27]

    Cascaded diffusion models for high fidelity image generation,

    J. Ho, C. Saharia, W. Chan, D. J. Fleet, M. Norouzi, and T. Salimans, “Cascaded diffusion models for high fidelity image generation,”Journal of Machine Learning Research, vol. 23, no. 47, pp. 1–33, 2022

    2022

  28. [28]

    Repaint: Inpainting using denoising diffusion probabilistic models,

    A. Lugmayr, M. Danelljan, A. Romero, F. Yu, R. Timofte, and L. Van Gool, “Repaint: Inpainting using denoising diffusion probabilistic models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 11 461–11 471

    2022

  29. [29]

    SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

    D. Podell, Z. English, K. Lacey, A. Blattmann, T. Dockhorn, J. M ¨uller, J. Penna, and R. Rombach, “Sdxl: Improving latent diffusion models for high-resolution image synthesis,”arXiv preprint arXiv:2307.01952, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  30. [30]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning. PMLR, 2021, pp. 8748–8763

    2021

  31. [31]

    Adversarial example does good: Preventing painting im- itation from diffusion models via adversarial examples,

    C. Liang, X. Wu, Y . Hua, J. Zhang, Y . Xue, T. Song, Z. Xue, R. Ma, and H. Guan, “Adversarial example does good: Preventing painting im- itation from diffusion models via adversarial examples,”arXiv preprint arXiv:2302.04578, 2023

    work page arXiv 2023

  32. [32]

    Imagereward: Learning and evaluating human preferences for text-to- image generation,

    J. Xu, X. Liu, Y . Wu, Y . Tong, Q. Li, M. Ding, J. Tang, and Y . Dong, “Imagereward: Learning and evaluating human preferences for text-to- image generation,”Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  33. [33]

    The surprising effectiveness of ppo in cooperative multi-agent games,

    C. Yu, A. Velu, E. Vinitsky, J. Gao, Y . Wang, A. Bayen, and Y . Wu, “The surprising effectiveness of ppo in cooperative multi-agent games,” Advances in neural information processing systems, vol. 35, pp. 24 611– 24 624, 2022

    2022

  34. [34]

    Celebv-hq: A large-scale video facial attributes dataset,

    H. Zhu, W. Wu, W. Zhu, L. Jiang, S. Tang, L. Zhang, Z. Liu, and C. C. Loy, “Celebv-hq: A large-scale video facial attributes dataset,” in European conference on computer vision. Springer, 2022, pp. 650–667

    2022

  35. [35]

    Vggface2: A dataset for recognising faces across pose and age,

    Q. Cao, L. Shen, W. Xie, O. M. Parkhi, and A. Zisserman, “Vggface2: A dataset for recognising faces across pose and age,” in2018 13th IEEE international conference on automatic face & gesture recognition (FG 2018). IEEE, 2018, pp. 67–74

    2018

  36. [36]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” 2021

    2021

  37. [37]

    Scaling rectified flow transformers for high-resolution image synthesis,

    P. Esser, S. Kulal, A. Blattmann, R. Entezari, J. M ¨uller, H. Saini, Y . Levi, D. Lorenz, A. Sauer, F. Boeselet al., “Scaling rectified flow transformers for high-resolution image synthesis,” inForty-first international conference on machine learning, 2024

    2024

  38. [38]

    FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space

    B. F. Labs, S. Batifol, A. Blattmann, F. Boesel, S. Consul, C. Diagne, T. Dockhorn, J. English, Z. English, P. Esser, S. Kulal, K. Lacey, Y . Levi, C. Li, D. Lorenz, J. M ¨uller, D. Podell, R. Rombach, H. Saini, A. Sauer, and L. Smith, “Flux.1 kontext: Flow matching for in-context image generation and editing in latent space,” 2025. [Online]. Available: h...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    From facial parts responses to face detection: A deep learning approach,

    S. Yang, P. Luo, C.-C. Loy, and X. Tang, “From facial parts responses to face detection: A deep learning approach,” inProceedings of the IEEE international conference on computer vision, 2015, pp. 3676–3684

    2015

  40. [40]

    Arcface: Additive angular margin loss for deep face recognition,

    J. Deng, J. Guo, N. Xue, and S. Zafeiriou, “Arcface: Additive angular margin loss for deep face recognition,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 4690– 4699

    2019

  41. [41]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,”Advances in neural information processing systems, vol. 30, 2017

    2017

  42. [42]

    Retinaface: Single-shot multi-level face localisation in the wild,

    J. Deng, J. Guo, E. Ververas, I. Kotsia, and S. Zafeiriou, “Retinaface: Single-shot multi-level face localisation in the wild,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 5203–5212

    2020

  43. [43]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 586–595

    2018

  44. [44]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE transactions on image processing, vol. 13, no. 4, pp. 600–612, 2004

    2004

  45. [45]

    Laion- 5b: An open large-scale dataset for training next generation image-text models,

    C. Schuhmann, R. Beaumont, R. Vencu, C. Gordon, R. Wightman, M. Cherti, T. Coombes, A. Katta, C. Mullis, M. Wortsmanet al., “Laion- 5b: An open large-scale dataset for training next generation image-text models,”Advances in neural information processing systems, vol. 35, pp. 25 278–25 294, 2022

    2022