pith. sign in

arxiv: 2604.21879 · v1 · submitted 2026-04-23 · 💻 cs.CV · cs.AI

Addressing Image Authenticity When Cameras Use Generative AI

Umar Masud , Abhijith Punnappurath , Luxi Zhao , David B. Lindell , Michael S. Brown This is my paper

Pith reviewed 2026-05-09 22:21 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords image authenticitygenerative AI in camerasISP hallucinationsimage recoveryMLP decoderpost-capture processingcamera image signal processor
0
0 comments X

The pith

An image-specific MLP decoder with a modality encoder recovers the pre-hallucination version of camera images after generative AI processing in the ISP.

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

Cameras now incorporate generative AI directly into their image signal processors, which can add or alter content in ways that change how an image is interpreted, especially during AI zoom or low-light enhancement. This paper establishes a way to recover the version of the image that existed before those AI additions by optimizing a small multi-layer perceptron decoder tailored to each image along with a fixed encoder for the type of image data. The entire system fits in 180 kilobytes and works after the photo has been taken without needing any access to the camera's internal hardware. If this holds, it would give users a practical way to check and restore the original content of their camera photos to prevent being misled by AI-generated details.

Core claim

By optimizing an image-specific multi-layer perceptron (MLP) decoder together with a modality-specific encoder, the method recovers the image before hallucinated content was added from the final camera output. The encoder and MLP decoder are self-contained, require only 180 KB of storage, and can be embedded as metadata in standard formats such as JPEG and HEIC, allowing post-capture application without ISP access.

What carries the argument

An image-specific MLP decoder optimized jointly with a modality-specific encoder that inverts the effects of generative AI operations performed during image capture.

If this is right

  • The recovered unhallucinated image helps avoid misinterpretation of semantic content introduced by AI enhancements.
  • The approach operates post-capture on the final image alone, independent of the camera ISP.
  • Minimal storage of 180 KB allows embedding the decoder and encoder in common image file formats.
  • Applies to specific operations like AI-based digital zoom and low-light image enhancement.

Where Pith is reading between the lines

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

  • This could lead to software tools that automatically restore original image details from enhanced camera photos.
  • The method might extend to detecting and correcting other forms of generative modifications in imaging devices.
  • Camera manufacturers could adopt this to provide optional authenticity recovery features in future models.

Load-bearing premise

The hallucinations added by generative operations in the camera ISP are sufficiently invertible from the final image alone through optimization of an image-specific MLP decoder with a modality encoder, without access to ground-truth pre-hallucination data or the ISP itself.

What would settle it

Comparing the MLP-recovered image against a known ground-truth pre-hallucination image (such as a non-AI zoomed capture) and finding that it does not reduce the difference metrics like perceptual loss compared to the original AI-processed image.

Figures

Figures reproduced from arXiv: 2604.21879 by Abhijith Punnappurath, David B. Lindell, Luxi Zhao, Michael S. Brown, Umar Masud.

Figure 1
Figure 1. Figure 1: Modern smartphone image signal processors (ISPs) are [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: GenAI-based super-resolution applied by the ISP to faces can sometimes result in subtle changes to appearance or even identity. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: An overview of our proposed method. (A) At capture time, we (i) run the ISP’s output image [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of various methods against our proposed [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative results for natural image super-resolution on the DIV2K [ [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative results for low-light image enhancement on the LOL [ [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Examples of hallucinations from GenAI-based text [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
read the original abstract

The ability of generative AI (GenAI) methods to photorealistically alter camera images has raised awareness about the authenticity of images shared online. Interestingly, images captured directly by our cameras are considered authentic and faithful. However, with the increasing integration of deep-learning modules into cameras' capture-time hardware -- namely, the image signal processor (ISP) -- there is now a potential for hallucinated content in images directly output by our cameras. Hallucinated capture-time image content is typically benign, such as enhanced edges or texture, but in certain operations, such as AI-based digital zoom or low-light image enhancement, hallucinations can potentially alter the semantics and interpretation of the image content. As a result, users may not realize that the content in their camera images is not authentic. This paper addresses this issue by enabling users to recover the 'unhallucinated' version of the camera image to avoid misinterpretation of the image content. Our approach works by optimizing an image-specific multi-layer perceptron (MLP) decoder together with a modality-specific encoder so that, given the camera image, we can recover the image before hallucinated content was added. The encoder and MLP are self-contained and can be applied post-capture to the image without requiring access to the camera ISP. Moreover, the encoder and MLP decoder require only 180 KB of storage and can be readily saved as metadata within standard image formats such as JPEG and HEIC.

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 / 0 minor

Summary. The manuscript proposes a post-capture method to recover the 'unhallucinated' version of images captured by cameras whose ISP incorporates generative AI modules. The approach optimizes an image-specific MLP decoder together with a modality-specific encoder using only the final camera image, enabling recovery of pre-hallucination content without direct ISP access. The encoder and decoder are claimed to require only 180 KB of storage and can be embedded as metadata in standard formats like JPEG and HEIC.

Significance. If the proposed optimization successfully inverts the effects of ISP-based generative AI operations such as AI zoom and low-light enhancement, the result would be significant for image forensics and authenticity verification in consumer devices. It offers a lightweight, self-contained solution that does not require manufacturer cooperation or access to proprietary ISP pipelines, potentially allowing widespread adoption for preserving original scene content.

major comments (2)
  1. [Abstract] Abstract: The central claim that an image-specific MLP decoder can recover the pre-hallucination image is presented without any quantitative results, validation experiments, error metrics, or comparisons to baselines. This leaves the invertibility assertion unsupported by evidence.
  2. [Abstract] Abstract: The optimization of the MLP decoder is described only at a high level; no loss function, objective, or training procedure is specified for the case where no ground-truth pre-hallucination image is available. This makes the mapping underconstrained, as many plausible pre-images are consistent with the observed output.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments and positive evaluation of the work's potential impact. We respond point-by-point to the major comments below and indicate the revisions we will incorporate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that an image-specific MLP decoder can recover the pre-hallucination image is presented without any quantitative results, validation experiments, error metrics, or comparisons to baselines. This leaves the invertibility assertion unsupported by evidence.

    Authors: The abstract is intentionally concise and does not include numerical results. The full manuscript contains quantitative validation experiments, including error metrics such as PSNR and SSIM, along with comparisons to relevant baselines that support the invertibility claim. We will revise the abstract to include a brief summary of these key quantitative results. revision: yes

  2. Referee: [Abstract] Abstract: The optimization of the MLP decoder is described only at a high level; no loss function, objective, or training procedure is specified for the case where no ground-truth pre-hallucination image is available. This makes the mapping underconstrained, as many plausible pre-images are consistent with the observed output.

    Authors: The abstract provides a high-level overview, but the full manuscript details the self-supervised optimization, including the specific loss function (a combination of reconstruction consistency with the observed image and regularization terms to constrain the solution space) and the training procedure. This addresses the underconstrained nature by enforcing fidelity and preventing arbitrary pre-images. We will revise the abstract to briefly specify the loss function and objective. revision: yes

Circularity Check

0 steps flagged

No significant circularity; method is an independent optimization procedure

full rationale

The paper presents a post-capture optimization of an image-specific MLP decoder and modality-specific encoder to recover a pre-hallucination image from the final camera output. No equations, derivations, or first-principles claims are described that reduce the recovery result to a fitted parameter, self-referential definition, or self-citation chain. The approach is framed as an empirical optimization applied to the observed image alone, without any load-bearing step that renames or tautologically equates the output to the input by construction. The central claim therefore remains self-contained as a proposed method rather than a circular restatement of its own assumptions.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the domain assumption that ISP hallucinations are learnable and invertible from output images alone via per-image optimization, plus standard assumptions about neural network optimization.

free parameters (1)
  • image-specific MLP weights
    The decoder is optimized per image to recover the unhallucinated version, introducing image-dependent parameters fitted during inference.
axioms (2)
  • domain assumption Hallucinations from AI-based ISP operations are invertible using a modality-specific encoder and image-specific decoder without access to the original capture pipeline.
    Invoked in the description of the recovery approach.
  • domain assumption A modality-specific encoder can be trained or provided to enable the decoder to undo hallucinations across similar camera modes.
    Required for the encoder-decoder pair to function post-capture.

pith-pipeline@v0.9.0 · 5570 in / 1455 out tokens · 44004 ms · 2026-05-09T22:21:58.433464+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

61 extracted references · 61 canonical work pages

  1. [1]

    Accessed: 2026-02-26

    DALL·E 3.https://www.openai.com/index/ dall-e-3/. Accessed: 2026-02-26. 1

    work page 2026

  2. [2]

    Make your own deepfakes.https://deepfakesweb. com/. Accessed: 2026-02-26. 1

    work page 2026

  3. [3]

    Accessed: 2026-02-26

    EU Artificial Intelligence Act.https : / / artificialintelligenceact.eu/high-level- summary/. Accessed: 2026-02-26. 1

    work page 2026

  4. [4]

    Accessed: 2026-02-26

    Faceapp: Face editor.https://www.faceapp.com/. Accessed: 2026-02-26. 1

    work page 2026

  5. [5]

    Accessed: 2026-02-26

    AI generated images on FB.https://about.fb.com/ news / 2024 / 02 / labeling - ai - generated - images - on - facebook - instagram - and - threads/. Accessed: 2026-02-26. 2

    work page 2024

  6. [6]

    Adobe Firefly.https : / / www . adobe . com / products/firefly.html. Accessed: 2026-02-26. 1

    work page 2026

  7. [7]

    google / technology / ai / google - gen - ai - content - transparency - c2pa/

    GenAI content transparency on Google.https : / / blog . google / technology / ai / google - gen - ai - content - transparency - c2pa/. Accessed: 2026-02-26. 2

    work page 2026

  8. [8]

    Accessed: 2026-02-26

    MidJourney.https://midjourney.com/. Accessed: 2026-02-26. 1

    work page 2026

  9. [9]

    Protecting world leaders against deep fakes

    Shruti Agarwal, Hany Farid, Yuming Gu, Mingming He, Koki Nagano, and Hao Li. Protecting world leaders against deep fakes. InCVPR workshops, 2019. 1

    work page 2019

  10. [10]

    NTIRE 2017 chal- lenge on single image super-resolution: Dataset and study

    Eirikur Agustsson and Radu Timofte. NTIRE 2017 chal- lenge on single image super-resolution: Dataset and study. InCVPR Workshops, 2017. 5, 6, 7, 8, 9, 10, 11

    work page 2017

  11. [11]

    Michael S. Brown. Color processing for digital cameras. InFundamentals and Applications of Colour Engineering, chapter 4, pages 81–98. Wiley, 2023. 2

    work page 2023

  12. [12]

    Simple baselines for image restoration

    Liangyu Chen, Xiaojie Chu, Xiangyu Zhang, and Jian Sun. Simple baselines for image restoration. InECCV, 2022. 5, 6, 9

    work page 2022

  13. [13]

    Brown, and Peyman Milanfar

    Mauricio Delbracio, Damien Kelly, Michael S. Brown, and Peyman Milanfar. Mobile computational photography: A tour.Annual Review of Vision Science, 7(1):571–604, 2021. 2

    work page 2021

  14. [14]

    Do (microtargeted) deepfakes have real effects on political attitudes?The International Journal of Press/Politics, 26(1):69–91, 2021

    Tom Dobber, Nadia Metoui, Damian Trilling, Natali Hel- berger, and Claes De Vreese. Do (microtargeted) deepfakes have real effects on political attitudes?The International Journal of Press/Politics, 26(1):69–91, 2021. 1

    work page 2021

  15. [15]

    Image super-resolution using deep convolutional net- works.IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(2):295–307, 2016

    Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Image super-resolution using deep convolutional net- works.IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(2):295–307, 2016. 3

    work page 2016

  16. [16]

    Brown, Radu Timofte, Karlo Ko ˇsˇcevi´c, Michael Freeman, Vasily Tesalin, Dmitry Bocharov, et al

    Egor Ershov, Alex Savchik, Denis Shepelev, Nikola Bani ´c, Michael S. Brown, Radu Timofte, Karlo Ko ˇsˇcevi´c, Michael Freeman, Vasily Tesalin, Dmitry Bocharov, et al. NTIRE 2022 challenge on night photography rendering. InCVPR Workshops, 2022. 3

    work page 2022

  17. [17]

    Image forgery detection.IEEE Signal Process- ing Magazine, 26(2):16–25, 2009

    Hany Farid. Image forgery detection.IEEE Signal Process- ing Magazine, 26(2):16–25, 2009. 3

    work page 2009

  18. [18]

    Creating, using, misusing, and detecting deep fakes.Journal of Online Trust and Safety, 1(4), 2022

    Hany Farid. Creating, using, misusing, and detecting deep fakes.Journal of Online Trust and Safety, 1(4), 2022. 3

    work page 2022

  19. [19]

    Implicit diffusion models for continuous super-resolution

    Sicheng Gao, Xuhui Liu, Bohan Zeng, Sheng Xu, Yan- jing Li, Xiaoyan Luo, Jianzhuang Liu, Xiantong Zhen, and Baochang Zhang. Implicit diffusion models for continuous super-resolution. InCVPR, 2023. 2

    work page 2023

  20. [20]

    Generative adversarial nets.NeurIPS, 2014

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets.NeurIPS, 2014. 3

    work page 2014

  21. [21]

    TruFor: Leveraging all-round clues for trustworthy image forgery detection and localiza- tion

    Fabrizio Guillaro, Davide Cozzolino, Avneesh Sud, Nicholas Dufour, and Luisa Verdoliva. TruFor: Leveraging all-round clues for trustworthy image forgery detection and localiza- tion. InCVPR, 2023. 3

    work page 2023

  22. [22]

    You won’t believe what they just said! The effects of political deepfakes embedded as vox populi on social media

    Michael Hameleers, Toni GLA van der Meer, and Tom Dob- ber. You won’t believe what they just said! The effects of political deepfakes embedded as vox populi on social media. Social Media+ Society, 8(3), 2022. 1

    work page 2022

  23. [23]

    DRAW: Defending camera-shooted RAW against image manipulation

    Xiaoxiao Hu, Qichao Ying, Zhenxing Qian, Sheng Li, and Xinpeng Zhang. DRAW: Defending camera-shooted RAW against image manipulation. InICCV, 2023. 3

    work page 2023

  24. [24]

    AIM 2019 chal- lenge on RAW to RGB mapping: Methods and results

    Andrey Ignatov, Radu Timofte, Sung-Jea Ko, Seung-Wook Kim, Kwang-Hyun Uhm, Seo-Won Ji, Sung-Jin Cho, Jun- Pyo Hong, Kangfu Mei, Juncheng Li, et al. AIM 2019 chal- lenge on RAW to RGB mapping: Methods and results. In ICCV Workshops, 2019. 3

    work page 2019

  25. [25]

    AIM 2020 chal- lenge on learned image signal processing pipeline

    Andrey Ignatov, Radu Timofte, Zhilu Zhang, Ming Liu, Haolin Wang, Wangmeng Zuo, Jiawei Zhang, Ruimao Zhang, Zhanglin Peng, Sijie Ren, et al. AIM 2020 chal- lenge on learned image signal processing pipeline. InECCV Workshops, 2020

    work page 2020

  26. [26]

    Learned smartphone ISP on mobile NPUs with deep learning, mobile AI 2021 chal- lenge: Report

    Andrey Ignatov, Cheng-Ming Chiang, Hsien-Kai Kuo, Anas- tasia Sycheva, and Radu Timofte. Learned smartphone ISP on mobile NPUs with deep learning, mobile AI 2021 chal- lenge: Report. InCVPR Workshops, 2021. 3

    work page 2021

  27. [27]

    Efficient and accurate quantized image super-resolution on mobile NPUs, mobile AI & AIM 2022 challenge: report

    Andrey Ignatov, Radu Timofte, Maurizio Denna, Abdel Younes, Ganzorig Gankhuyag, Jingang Huh, Myeong Kyun Kim, Kihwan Yoon, Hyeon-Cheol Moon, Seungho Lee, et al. Efficient and accurate quantized image super-resolution on mobile NPUs, mobile AI & AIM 2022 challenge: report. In ECCV Workshops, 2022. 3

    work page 2022

  28. [28]

    Learned smartphone ISP on mobile GPUs with deep learning, mobile AI & AIM 2022 challenge: Report

    Andrey Ignatov, Radu Timofte, Shuai Liu, Chaoyu Feng, Fu- rui Bai, Xiaotao Wang, Lei Lei, Ziyao Yi, Yan Xiang, Zibin Liu, et al. Learned smartphone ISP on mobile GPUs with deep learning, mobile AI & AIM 2022 challenge: Report. In ECCV Workshops, 2022. 3

    work page 2022

  29. [29]

    AutoDIR: Automatic all-in-one image restoration with latent diffusion

    Yitong Jiang, Zhaoyang Zhang, Tianfan Xue, and Jinwei Gu. AutoDIR: Automatic all-in-one image restoration with latent diffusion. InECCV, 2024. 6

    work page 2024

  30. [30]

    Perceptual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Perceptual losses for real-time style transfer and super-resolution. In ECCV, 2016. 3

    work page 2016

  31. [31]

    Adam: A method for stochastic optimization.arXiv, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv, 2014. 6

    work page 2014

  32. [32]

    Digital image integrity – a survey of protection and verification techniques.Digital Signal Processing, 71:1– 26, 2017

    Paweł Korus. Digital image integrity – a survey of protection and verification techniques.Digital Signal Processing, 71:1– 26, 2017. 3

    work page 2017

  33. [33]

    Content authentication for neural imaging pipelines: End-to-end optimization of photo provenance in complex distribution channels

    Pawel Korus and Nasir Memon. Content authentication for neural imaging pipelines: End-to-end optimization of photo provenance in complex distribution channels. InCVPR,

    work page

  34. [34]

    GamutMLP: A lightweight MLP for color loss re- covery

    Hoang M Le, Brian Price, Scott Cohen, and Michael S Brown. GamutMLP: A lightweight MLP for color loss re- covery. InCVPR, 2023. 4

    work page 2023

  35. [35]

    Photo- realistic single image super-resolution using a generative ad- versarial network

    Christian Ledig, Lucas Theis, Ferenc Husz´ar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo- realistic single image super-resolution using a generative ad- versarial network. InCVPR, 2017. 3

    work page 2017

  36. [36]

    Metadata- based RAW reconstruction via implicit neural functions

    Leyi Li, Huijie Qiao, Qi Ye, and Qinmin Yang. Metadata- based RAW reconstruction via implicit neural functions. In CVPR, 2023. 4

    work page 2023

  37. [37]

    Learning generative structure prior for blind text image super-resolution

    Xiaoming Li, Wangmeng Zuo, and Chen Change Loy. Learning generative structure prior for blind text image super-resolution. InCVPR, 2023. 3, 5, 7

    work page 2023

  38. [38]

    CameraNet: A two-stage framework for effective camera ISP learning.IEEE Transactions on Image Processing, 30: 2248–2262, 2021

    Zhetong Liang, Jianrui Cai, Zisheng Cao, and Lei Zhang. CameraNet: A two-stage framework for effective camera ISP learning.IEEE Transactions on Image Processing, 30: 2248–2262, 2021. 3, 4

    work page 2021

  39. [39]

    Deep-FlexISP: A three- stage framework for night photography rendering

    Shuai Liu, Chaoyu Feng, Xiaotao Wang, Hao Wang, Ran Zhu, Yongqiang Li, and Lei Lei. Deep-FlexISP: A three- stage framework for night photography rendering. InCVPR Workshops, 2022. 3, 4

    work page 2022

  40. [40]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. NeRF: Representing scenes as neural radiance fields for view syn- thesis. InECCV, 2020. 2, 3, 5, 6, 7

    work page 2020

  41. [41]

    Instant neural graphics primitives with a multires- olution hash encoding.ACM Transactions on Graphics, 41 (4):1–15, 2022

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexan- der Keller. Instant neural graphics primitives with a multires- olution hash encoding.ACM Transactions on Graphics, 41 (4):1–15, 2022. 3, 5, 6, 7, 9

    work page 2022

  42. [42]

    Brubaker, and Michael S

    Seonghyeon Nam, Abhijith Punnappurath, Marcus A. Brubaker, and Michael S. Brown. Learning sRGB-to-raw- RGB de-rendering with content-aware metadata. InCVPR,

    work page

  43. [43]

    Rang M. H. Nguyen and Michael S. Brown. RAW image re- construction using a self-contained sRGB-JPEG image with only 64 KB overhead. InCVPR, 2016. 4

    work page 2016

  44. [44]

    Deep learning for deepfakes creation and detection: A survey.Computer Vision and Image Under- standing, 223:103525, 2022

    Thanh Thi Nguyen, Quoc Viet Hung Nguyen, Dung Tien Nguyen, Duc Thanh Nguyen, Thien Huynh-The, Saeid Nahavandi, Thanh Tam Nguyen, Quoc-Viet Pham, and Cuong M Nguyen. Deep learning for deepfakes creation and detection: A survey.Computer Vision and Image Under- standing, 223:103525, 2022. 3

    work page 2022

  45. [45]

    A triple self- embedding fragile watermarking scheme for image tamper detection and recovery.IEEE Access, 12:140082–140096,

    Huseyin Ozkaya and Veysel Aslantas. A triple self- embedding fragile watermarking scheme for image tamper detection and recovery.IEEE Access, 12:140082–140096,

    work page

  46. [46]

    Statistical tools for digi- tal forensics

    Alin C Popescu and Hany Farid. Statistical tools for digi- tal forensics. InThe International Workshop on Information Hiding, 2004. 3

    work page 2004

  47. [47]

    Abhijith Punnappurath, Luxi Zhao, Abdelrahman Abdel- hamed, and Michael S. Brown. Advocating pixel-level au- thentication of camera-captured images.IEEE Access, 12: 45839–45846, 2024. 2, 3, 4, 5, 9

    work page 2024

  48. [48]

    NTIRE 2023 challenge on night photography rendering

    Alina Shutova, Egor Ershov, Georgy Perevozchikov, Ivan Er- makov, Nikola Bani´c, Radu Timofte, Richard Collins, Maria Efimova, Arseniy Terekhin, Simone Zini, et al. NTIRE 2023 challenge on night photography rendering. InCVPR Work- shops, 2023. 3

    work page 2023

  49. [49]

    Implicit neural representa- tions with periodic activation functions.NeurIPS, 2020

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural representa- tions with periodic activation functions.NeurIPS, 2020. 2, 3, 5, 6, 7

    work page 2020

  50. [50]

    Implicit neural representations for image compression

    Yannick Str ¨umpler, Janis Postels, Ren Yang, Luc Van Gool, and Federico Tombari. Implicit neural representations for image compression. InECCV, 2022. 2, 4

    work page 2022

  51. [51]

    Deepfakes and disinformation: Exploring the impact of synthetic political video on deception, uncertainty, and trust in news.Social media+ society, 6(1), 2020

    Cristian Vaccari and Andrew Chadwick. Deepfakes and disinformation: Exploring the impact of synthetic political video on deception, uncertainty, and trust in news.Social media+ society, 6(1), 2020. 1

    work page 2020

  52. [52]

    Media forensics and deepfakes: an overview.IEEE Journal of Selected Topics in Signal Pro- cessing, 14(5):910–932, 2020

    Luisa Verdoliva. Media forensics and deepfakes: an overview.IEEE Journal of Selected Topics in Signal Pro- cessing, 14(5):910–932, 2020. 3

    work page 2020

  53. [53]

    Real-ESRGAN: Training real-world blind super-resolution with pure synthetic data

    Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-ESRGAN: Training real-world blind super-resolution with pure synthetic data. InICCV Workshops, 2021. 3, 5, 7, 9

    work page 2021

  54. [54]

    Raw image reconstruc- tion with learned compact metadata

    Yufei Wang, Yi Yu, Wenhan Yang, Lanqing Guo, Lap-Pui Chau, Alex C Kot, and Bihan Wen. Raw image reconstruc- tion with learned compact metadata. InCVPR, 2023. 4

    work page 2023

  55. [55]

    SinSR: Diffusion-based image super- resolution in a single step

    Yufei Wang, Wenhan Yang, Xinyuan Chen, Yaohui Wang, Lanqing Guo, Lap-Pui Chau, Ziwei Liu, Yu Qiao, Alex C Kot, and Bihan Wen. SinSR: Diffusion-based image super- resolution in a single step. InCVPR, 2024. 3

    work page 2024

  56. [56]

    Deep Retinex decomposition for low-light enhancement

    Chen Wei, Wenjing Wang, Wenhan Yang, and Jiaying Liu. Deep Retinex decomposition for low-light enhancement. In BMVC, 2018. 5, 6, 7, 8

    work page 2018

  57. [57]

    Learning to immunize images for tamper localization and self-recovery.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    Qichao Ying, Hang Zhou, Zhenxing Qian, Sheng Li, and Xinpeng Zhang. Learning to immunize images for tamper localization and self-recovery.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023. 3

    work page 2023

  58. [58]

    ResShift: Efficient diffusion model for image super- resolution by residual shifting.NeurIPS, 2024

    Zongsheng Yue, Jianyi Wang, and Chen Change Loy. ResShift: Efficient diffusion model for image super- resolution by residual shifting.NeurIPS, 2024. 3

    work page 2024

  59. [59]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, 2018. 3

    work page 2018

  60. [60]

    EditGuard: Versatile image watermarking for tamper localization and copyright protection

    Xuanyu Zhang, Runyi Li, Jiwen Yu, Youmin Xu, Weiqi Li, and Jian Zhang. EditGuard: Versatile image watermarking for tamper localization and copyright protection. InCVPR,

    work page

  61. [61]

    Diffusion-based blind text image super-resolution

    Yuzhe Zhang, Jiawei Zhang, Hao Li, Zhouxia Wang, Luwei Hou, Dongqing Zou, and Liheng Bian. Diffusion-based blind text image super-resolution. InCVPR, 2024. 3

    work page 2024