Physically-Guided Optical Inversion Enable Non-Contact Side-Channel Attack on Isolated Screens

Huiyu Zhou; Jin Liu; Shaowei Jiang; Tao Zhang; Wenwen Tang; Xiaoshuai Zhang; Xingru Huang; Yuheng Qiao; Zhao Huang; Zhiwen Zheng

arxiv: 2604.13419 · v1 · submitted 2026-04-15 · 💻 cs.CV

Physically-Guided Optical Inversion Enable Non-Contact Side-Channel Attack on Isolated Screens

Zhiwen Zheng , Yuheng Qiao , Xiaoshuai Zhang , Zhao Huang , Tao Zhang , Huiyu Zhou , Shaowei Jiang , Jin Liu

show 2 more authors

Wenwen Tang Xingru Huang

This is my paper

Pith reviewed 2026-05-10 13:16 UTC · model grok-4.3

classification 💻 cs.CV

keywords non-contact side-channel attackoptical inversionscreen content exfiltrationpassive speckle patternsradiometric inversionphysical regularizationsemantic reprojectionIR4Net

0 comments

The pith

IR4Net recovers electronic screen content from passive speckle patterns in diffuse reflections by embedding physical light transport equations into a neural inversion network.

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

The paper establishes a non-contact optical side-channel method for extracting content from isolated screens. It confronts the near-singular Jacobian of projection mapping and irreversible compression during light transport by exploiting passive speckle patterns. IR4Net integrates a physically regularized irradiance approximation that embeds the radiative transfer equation, a contour-to-detail cross-scale reconstruction to limit noise, and an irreversibility-constrained semantic reprojection module to restore global structure. The resulting reconstructions maintain higher fidelity than prior neural approaches across varied scenes and remain stable under illumination changes.

Core claim

IR4Net performs stable radiometric inversion by fusing a Physically Regularized Irradiance Approximation that embeds the radiative transfer equation in a learnable optimizer, a contour-to-detail cross-scale reconstruction mechanism that arrests noise propagation, and an Irreversibility Constrained Semantic Reprojection module that reinstates lost global structure through context-driven semantic mapping, achieving superior fidelity while retaining resilience to illumination perturbations.

What carries the argument

Physically Regularized Irradiance Approximation (PRIrr-Approximation) that embeds the radiative transfer equation in a learnable optimizer, paired with the Irreversibility Constrained Semantic Reprojection (ICSR) module for semantic structure recovery from passive speckle patterns.

If this is right

Screen content becomes exfiltratable via optical side-channels even when the display is isolated and viewed only through reflections.
Reconstructions remain stable across four scene categories despite illumination perturbations.
The physically regularized modules reduce sensitivity to the singular Jacobian spectrum compared with standard neural inversion.
Global semantic cues lost in light transport can be restored through context-driven reprojection.

Where Pith is reading between the lines

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

The same passive-speckle inversion principle could apply to recovering information from other reflective surfaces such as windows or tabletops.
Integration with existing camera networks might allow remote auditing of air-gapped displays without physical access.
Countermeasures such as screen polarization filters or randomized backlight modulation could be evaluated as direct falsifiers of the attack surface.

Load-bearing premise

Passive speckle patterns formed by diffuse reflection contain enough recoverable information to overcome the near-singular Jacobian and irreversible compression in light transport.

What would settle it

A controlled test in which IR4Net is applied to scenes with progressively weaker diffuse reflections or stronger illumination perturbations, checking whether semantic correctness of the recovered screen content collapses below a measurable threshold.

Figures

Figures reproduced from arXiv: 2604.13419 by Huiyu Zhou, Jin Liu, Shaowei Jiang, Tao Zhang, Wenwen Tang, Xiaoshuai Zhang, Xingru Huang, Yuheng Qiao, Zhao Huang, Zhiwen Zheng.

**Figure 2.** Figure 2: Overall architecture of IR4Net, comprising the PRIrr-Approximation and ICSR modules. The multi-scale frequency separation module, a key component of ICSR, is implemented via concatenation. 3.1 PHYSICALLY-REGULARIZED IRRADIANCE APPROXIMATION Optical-projection side-channel attacks confront a fundamental challenge in the intricate physics of image formation: the observed image arises from a highly nonlinear… view at source ↗

**Figure 3.** Figure 3: Visual comparison of IR4Net and baseline methods on four datasets. Our model yields visually more faithful restorations across various scenes. 4.2 ABLATION EXPERIMENT Inversion behaviour was evaluated across three datasets using four iterative schemes: classical momentum formulations including ADMM, NAG, and Heavy-Ball, and the proposed update strategy [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

**Figure 4.** Figure 4: As screen brightness decreases on the ReSh-Screen dataset, our model’s PSNR degrades [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

**Figure 5.** Figure 5: Experimental setups for camera motion. (a) Camera orbital motion: (a1) motion along a [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

read the original abstract

Noncontact exfiltration of electronic screen content poses a security challenge, with side-channel incursions as the principal vector. We introduce an optical projection side-channel paradigm that confronts two core instabilities: (i) the near-singular Jacobian spectrum of projection mapping breaches Hadamard stability, rendering inversion hypersensitive to perturbations; (ii) irreversible compression in light transport obliterates global semantic cues, magnifying reconstruction ambiguity. Exploiting passive speckle patterns formed by diffuse reflection, our Irradiance Robust Radiometric Inversion Network (IR4Net) fuses a Physically Regularized Irradiance Approximation (PRIrr-Approximation), which embeds the radiative transfer equation in a learnable optimizer, with a contour-to-detail cross-scale reconstruction mechanism that arrests noise propagation. Moreover, an Irreversibility Constrained Semantic Reprojection (ICSR) module reinstates lost global structure through context-driven semantic mapping. Evaluated across four scene categories, IR4Net achieves fidelity beyond competing neural approaches while retaining resilience to illumination perturbations.

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.

Desk Editor's Note private letter to a colleague

The paper introduces named modules for inverting screen reflections into content via physics-guided neural nets, but the abstract supplies no metrics, ablations, or data to check whether the speckle patterns actually carry recoverable information.

read the letter

The core idea is to recover screen content from passive diffuse reflections using IR4Net, which combines a radiative-transfer embedding in the optimizer, a cross-scale contour-to-detail path, and a semantic reprojection step to handle the unstable Jacobian and lost global cues in light transport. Those three pieces look like a fresh combination for this specific inversion task, and the framing of the two instabilities is clear and on-point for a side-channel security setting. The attempt to keep the model anchored in the physics of irradiance rather than pure data fitting is a reasonable direction. That said, the abstract states superior fidelity and illumination robustness across four scene types without any numbers, error bars, ablation tables, or description of how the training data were collected or filtered. The central question—whether the captured speckle actually contains enough high-frequency and semantic signal to support stable inversion—remains unaddressed in the provided text. If the diffuse component is low-rank, the network could be producing plausible outputs rather than true inversions, and nothing in the summary rules that out. The physical regularization is described at a high level but without equations or training details it is hard to judge whether it reduces to post-hoc fitting. This work is aimed at researchers working on optical side channels and light-transport inversion in computer vision. A reader already following that niche could extract the module designs as starting points, but only the full experiments would show whether the approach holds. I would send it to referees so the quantitative claims and information-content analysis can be examined directly rather than desk-rejecting on the abstract alone.

Referee Report

3 major / 1 minor

Summary. The paper introduces IR4Net, a neural architecture for non-contact exfiltration of screen content via inversion of passive speckle patterns arising from diffuse reflection. It addresses two instabilities—the near-singular Jacobian of projection mapping and irreversible compression in light transport—by combining a Physically Regularized Irradiance Approximation (PRIrr-Approximation) that embeds the radiative transfer equation within a learnable optimizer and an Irreversibility Constrained Semantic Reprojection (ICSR) module for context-driven semantic mapping. The abstract asserts that this yields reconstructions of higher fidelity than competing neural methods while remaining resilient to illumination perturbations, evaluated across four scene categories.

Significance. If the quantitative claims hold after proper validation, the work would be significant for the fields of computer vision and hardware security, as it offers a physically guided approach to a classically ill-posed inverse problem and could inform practical side-channel threat models for isolated displays.

major comments (3)

[Abstract] Abstract: the central performance claim that IR4Net achieves fidelity beyond competing neural approaches is unsupported by any quantitative metrics, error bars, ablation studies, or comparison tables; without these, the superiority and illumination-resilience assertions cannot be evaluated.
[Abstract] Abstract: the description of PRIrr-Approximation states that it embeds the radiative transfer equation in a learnable optimizer, yet supplies neither the explicit formulation nor training details, leaving open whether the regularization is truly physics-constrained or reduces to parameters fitted on the same reconstruction data (circularity risk).
[Abstract] Abstract: no analysis is provided to confirm that passive speckle patterns retain sufficient high-frequency and global semantic information to overcome the near-singular Jacobian and irreversible light-transport compression; absent singular-value decay of the forward operator, mutual-information statistics, or ablations that remove the semantic prior, the stability of the reconstructions remains unverified.

minor comments (1)

[Title] The title contains a subject-verb agreement error ('Enable' should be 'Enables').

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. We have revised the abstract to incorporate quantitative support, explicit references to formulations, and pointers to stability analyses as outlined below. These changes strengthen the presentation without altering the core contributions.

read point-by-point responses

Referee: [Abstract] Abstract: the central performance claim that IR4Net achieves fidelity beyond competing neural approaches is unsupported by any quantitative metrics, error bars, ablation studies, or comparison tables; without these, the superiority and illumination-resilience assertions cannot be evaluated.

Authors: We agree that the abstract should directly reference supporting quantitative evidence. The full manuscript reports these results in Section 4, including comparison tables against neural baselines, ablation studies, and error bars computed over repeated trials under controlled illumination variations. To address the concern, we have revised the abstract to include concise quantitative statements summarizing the fidelity gains and illumination resilience demonstrated in the experiments. revision: yes
Referee: [Abstract] Abstract: the description of PRIrr-Approximation states that it embeds the radiative transfer equation in a learnable optimizer, yet supplies neither the explicit formulation nor training details, leaving open whether the regularization is truly physics-constrained or reduces to parameters fitted on the same reconstruction data (circularity risk).

Authors: The abstract provides a high-level summary of the module. The explicit formulation embedding the radiative transfer equation as a regularization term in the optimizer, together with training details that enforce the physics constraint independently of the target reconstruction data, appears in Section 3.2 and Section 4.1. We have updated the abstract to briefly note the RTE-derived nature of the regularization and its separation from data-driven fitting, thereby reducing the risk of perceived circularity. revision: yes
Referee: [Abstract] Abstract: no analysis is provided to confirm that passive speckle patterns retain sufficient high-frequency and global semantic information to overcome the near-singular Jacobian and irreversible light-transport compression; absent singular-value decay of the forward operator, mutual-information statistics, or ablations that remove the semantic prior, the stability of the reconstructions remains unverified.

Authors: Section 3.1 of the manuscript analyzes the information content of passive speckle patterns, including discussion of the Jacobian spectrum and light-transport compression. To directly respond to the request for singular-value decay, mutual-information statistics, and semantic-prior ablations, we have revised the abstract to reference these stability verifications and added explicit pointers to the corresponding figures and tables in the body. revision: partial

Circularity Check

0 steps flagged

No load-bearing circularity; derivation self-contained on available text.

full rationale

The abstract introduces IR4Net with PRIrr-Approximation embedding the radiative transfer equation in a learnable optimizer and ICSR for semantic reprojection, but presents these as architectural choices without equations, self-citations, or derivations that reduce predictions to fitted inputs by construction. No quotes exist showing self-definitional equivalence, fitted parameters renamed as predictions, or uniqueness imported from prior author work. The method is described as fusing physical regularization with cross-scale reconstruction to address instabilities, with evaluation across scenes claimed as empirical validation rather than tautological. Per rules, absent specific reductions or quotes exhibiting circularity, the chain is treated as self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that the radiative transfer equation can be usefully embedded as a learnable regularizer and that semantic context can reliably compensate for lost global cues; no explicit free parameters or invented entities are quantified in the abstract.

axioms (2)

domain assumption Radiative transfer equation governs the observed irradiance from diffuse reflections
Invoked as the basis for the Physically Regularized Irradiance Approximation module
domain assumption Contour information can be reliably extracted and used to guide detail reconstruction across scales
Core to the contour-to-detail cross-scale mechanism

pith-pipeline@v0.9.0 · 5505 in / 1471 out tokens · 32122 ms · 2026-05-10T13:16:34.004200+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

[1]

Side-channel attacks and countermea- sures in cloud services and infrastructures

Abdullah Albalawi, Vassilios Vassilakis, and Radu Calinescu. Side-channel attacks and countermea- sures in cloud services and infrastructures. InNOMS 2022-2022 IEEE/IFIP Network Operations and Management Symposium, pp. 1–4. IEEE,

work page 2022
[2]

Bin Chen, Gehui Li, Rongyuan Wu, Xindong Zhang, Jie Chen, Jian Zhang, and Lei Zhang

doi: 10.1109/BigData.2017.8258268. Bin Chen, Gehui Li, Rongyuan Wu, Xindong Zhang, Jie Chen, Jian Zhang, and Lei Zhang. Ad- versarial diffusion compression for real-world image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,

work page doi:10.1109/bigdata.2017.8258268 2017
[3]

Poster: Acoustic side-channel attack on robot vacu- ums

Peter Chen, Guannan Liu, and Haining Wang. Poster: Acoustic side-channel attack on robot vacu- ums. InProceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security, pp. 5027–5029,

work page 2024
[4]

Side-channel attack in internet of things: A survey

Mampi Devi and Abhishek Majumder. Side-channel attack in internet of things: A survey. In Applications of Internet of Things: Proceedings of ICCCIOT 2020, pp. 213–222. Springer,

work page 2020
[5]

Channel consis- tency prior and self-reconstruction strategy based unsupervised image deraining.arXiv preprint arXiv:2503.18703,

Guanglu Dong, Tianheng Zheng, Yuanzhouhan Cao, Linbo Qing, and Chao Ren. Channel consis- tency prior and self-reconstruction strategy based unsupervised image deraining.arXiv preprint arXiv:2503.18703,

work page arXiv
[6]

Towards a novel ultrasonic side-channel attack on mobile devices

Yingli Duan, Weizhi Meng, Wei-Yang Chiu, and Yu Wang. Towards a novel ultrasonic side-channel attack on mobile devices. InProceedings of the ACM SIGCOMM 2024 Conference: Posters and Demos, pp. 101–103,

work page 2024
[7]

Gcfsr: a generative and controllable face super resolution method without facial and gan priors

11 Published as a conference paper at ICLR 2026 Jingwen He, Wu Shi, Kai Chen, Lean Fu, and Chao Dong. Gcfsr: a generative and controllable face super resolution method without facial and gan priors. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 1889–1898,

work page 2026
[8]

Image-to-image translation with conditional adversarial networks

Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. InComputer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on,

work page 2017
[9]

A side-channel attack on a masked hardware implementation of crystals-kyber

Yanning Ji and Elena Dubrova. A side-channel attack on a masked hardware implementation of crystals-kyber. InProceedings of the 2023 Workshop on Attacks and Solutions in Hardware Security, pp. 27–37,

work page 2023
[10]

Imaging privacy threats from an ambient light sensor.Science Advances, 10(2):eadj3608, 2024a

Yang Liu, Gregory W Wornell, William T Freeman, and Fr ´edo Durand. Imaging privacy threats from an ambient light sensor.Science Advances, 10(2):eadj3608, 2024a. Zhibo Liu, Yuanyuan Yuan, Yanzuo Chen, Sihang Hu, Tianxiang Li, and Shuai Wang. Deepcache: Revisiting cache side-channel attacks in deep neural networks executables. InProceedings of the 2024 on ...

work page 2024
[11]

Coa: Towards real image dehazing via compression-and-adaptation.arXiv preprint arXiv:2504.05590,

Long Ma, Yuxin Feng, Yan Zhang, Jinyuan Liu, Weimin Wang, Guang-Yong Chen, Chengpei Xu, and Zhuo Su. Coa: Towards real image dehazing via compression-and-adaptation.arXiv preprint arXiv:2504.05590,

work page arXiv
[12]

U-net: Convolutional networks for biomed- ical image segmentation

12 Published as a conference paper at ICLR 2026 Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomed- ical image segmentation. InMedical image computing and computer-assisted intervention– MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceed- ings, part III 18, pp. 234–241. Springer,

work page 2026
[13]

Stealthy inference attack on dnn via cache-based side-channel attacks

Han Wang, Syed Mahbub Hafiz, Kartik Patwari, Chen-Nee Chuah, Zubair Shafiq, and Houman Homayoun. Stealthy inference attack on dnn via cache-based side-channel attacks. In2022 Design, Automation & Test in Europe Conference & Exhibition (DATE), pp. 1515–1520. IEEE, 2022a. Penghao Wang, Jingzhi Hu, Chao Liu, and Jun Luo. Reflexnoop: Passwords snooping on nlo...

work page 2024
[14]

Learning hazing to dehazing: Towards realistic haze generation for real-world image dehazing.arXiv preprint arXiv:2503.19262,

Ruiyi Wang, Yushuo Zheng, Zicheng Zhang, Chunyi Li, Shuaicheng Liu, Guangtao Zhai, and Xiao- hong Liu. Learning hazing to dehazing: Towards realistic haze generation for real-world image dehazing.arXiv preprint arXiv:2503.19262,

work page arXiv
[15]

Dnlut: Ultra-efficient color image denoising via channel-aware lookup tables.arXiv preprint arXiv:2503.15931,

Sidi Yang, Binxiao Huang, Yulun Zhang, Dahai Yu, Yujiu Yang, and Ngai Wong. Dnlut: Ultra-efficient color image denoising via channel-aware lookup tables.arXiv preprint arXiv:2503.15931,

work page arXiv
[16]

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

Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networkss. InComputer Vision (ICCV), 2017 IEEE Interna- tional Conference on, 2017a. Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Darrell, Alexei A Efros, Oliver Wang, and Eli Shechtman. Toward multimodal image-to-image...

work page 2017

[1] [1]

Side-channel attacks and countermea- sures in cloud services and infrastructures

Abdullah Albalawi, Vassilios Vassilakis, and Radu Calinescu. Side-channel attacks and countermea- sures in cloud services and infrastructures. InNOMS 2022-2022 IEEE/IFIP Network Operations and Management Symposium, pp. 1–4. IEEE,

work page 2022

[2] [2]

Bin Chen, Gehui Li, Rongyuan Wu, Xindong Zhang, Jie Chen, Jian Zhang, and Lei Zhang

doi: 10.1109/BigData.2017.8258268. Bin Chen, Gehui Li, Rongyuan Wu, Xindong Zhang, Jie Chen, Jian Zhang, and Lei Zhang. Ad- versarial diffusion compression for real-world image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,

work page doi:10.1109/bigdata.2017.8258268 2017

[3] [3]

Poster: Acoustic side-channel attack on robot vacu- ums

Peter Chen, Guannan Liu, and Haining Wang. Poster: Acoustic side-channel attack on robot vacu- ums. InProceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security, pp. 5027–5029,

work page 2024

[4] [4]

Side-channel attack in internet of things: A survey

Mampi Devi and Abhishek Majumder. Side-channel attack in internet of things: A survey. In Applications of Internet of Things: Proceedings of ICCCIOT 2020, pp. 213–222. Springer,

work page 2020

[5] [5]

Channel consis- tency prior and self-reconstruction strategy based unsupervised image deraining.arXiv preprint arXiv:2503.18703,

Guanglu Dong, Tianheng Zheng, Yuanzhouhan Cao, Linbo Qing, and Chao Ren. Channel consis- tency prior and self-reconstruction strategy based unsupervised image deraining.arXiv preprint arXiv:2503.18703,

work page arXiv

[6] [6]

Towards a novel ultrasonic side-channel attack on mobile devices

Yingli Duan, Weizhi Meng, Wei-Yang Chiu, and Yu Wang. Towards a novel ultrasonic side-channel attack on mobile devices. InProceedings of the ACM SIGCOMM 2024 Conference: Posters and Demos, pp. 101–103,

work page 2024

[7] [7]

Gcfsr: a generative and controllable face super resolution method without facial and gan priors

11 Published as a conference paper at ICLR 2026 Jingwen He, Wu Shi, Kai Chen, Lean Fu, and Chao Dong. Gcfsr: a generative and controllable face super resolution method without facial and gan priors. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 1889–1898,

work page 2026

[8] [8]

Image-to-image translation with conditional adversarial networks

Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. InComputer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on,

work page 2017

[9] [9]

A side-channel attack on a masked hardware implementation of crystals-kyber

Yanning Ji and Elena Dubrova. A side-channel attack on a masked hardware implementation of crystals-kyber. InProceedings of the 2023 Workshop on Attacks and Solutions in Hardware Security, pp. 27–37,

work page 2023

[10] [10]

Imaging privacy threats from an ambient light sensor.Science Advances, 10(2):eadj3608, 2024a

Yang Liu, Gregory W Wornell, William T Freeman, and Fr ´edo Durand. Imaging privacy threats from an ambient light sensor.Science Advances, 10(2):eadj3608, 2024a. Zhibo Liu, Yuanyuan Yuan, Yanzuo Chen, Sihang Hu, Tianxiang Li, and Shuai Wang. Deepcache: Revisiting cache side-channel attacks in deep neural networks executables. InProceedings of the 2024 on ...

work page 2024

[11] [11]

Coa: Towards real image dehazing via compression-and-adaptation.arXiv preprint arXiv:2504.05590,

Long Ma, Yuxin Feng, Yan Zhang, Jinyuan Liu, Weimin Wang, Guang-Yong Chen, Chengpei Xu, and Zhuo Su. Coa: Towards real image dehazing via compression-and-adaptation.arXiv preprint arXiv:2504.05590,

work page arXiv

[12] [12]

U-net: Convolutional networks for biomed- ical image segmentation

12 Published as a conference paper at ICLR 2026 Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomed- ical image segmentation. InMedical image computing and computer-assisted intervention– MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceed- ings, part III 18, pp. 234–241. Springer,

work page 2026

[13] [13]

Stealthy inference attack on dnn via cache-based side-channel attacks

Han Wang, Syed Mahbub Hafiz, Kartik Patwari, Chen-Nee Chuah, Zubair Shafiq, and Houman Homayoun. Stealthy inference attack on dnn via cache-based side-channel attacks. In2022 Design, Automation & Test in Europe Conference & Exhibition (DATE), pp. 1515–1520. IEEE, 2022a. Penghao Wang, Jingzhi Hu, Chao Liu, and Jun Luo. Reflexnoop: Passwords snooping on nlo...

work page 2024

[14] [14]

Learning hazing to dehazing: Towards realistic haze generation for real-world image dehazing.arXiv preprint arXiv:2503.19262,

Ruiyi Wang, Yushuo Zheng, Zicheng Zhang, Chunyi Li, Shuaicheng Liu, Guangtao Zhai, and Xiao- hong Liu. Learning hazing to dehazing: Towards realistic haze generation for real-world image dehazing.arXiv preprint arXiv:2503.19262,

work page arXiv

[15] [15]

Dnlut: Ultra-efficient color image denoising via channel-aware lookup tables.arXiv preprint arXiv:2503.15931,

Sidi Yang, Binxiao Huang, Yulun Zhang, Dahai Yu, Yujiu Yang, and Ngai Wong. Dnlut: Ultra-efficient color image denoising via channel-aware lookup tables.arXiv preprint arXiv:2503.15931,

work page arXiv

[16] [16]

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

Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networkss. InComputer Vision (ICCV), 2017 IEEE Interna- tional Conference on, 2017a. Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Darrell, Alexei A Efros, Oliver Wang, and Eli Shechtman. Toward multimodal image-to-image...

work page 2017