A Physics-Guided GAN Framework for Controllable Heavy-Rain Image Synthesis: Methodology and Theoretical Analysis

Xiaoru Su

doi:10.47297/taposatWSP2633-456926.20250608

PDF
Export
Share
Collection
Album

A Physics-Guided GAN Framework for Controllable Heavy-Rain Image Synthesis: Methodology and Theoretical Analysis

Volume 6, Issue 8

Views: 0 Downloads: 0

A Physics-Guided GAN Framework for Controllable Heavy-Rain Image Synthesis: Methodology and Theoretical Analysis
A Physics-Guided GAN Framework for Controllable Heavy-Rain Image Synthesis: Methodology and Theoretical Analysis
The HongKong Polytechnic University, HongKong Sar, 999077, China
Vol. 6, Issue 8, Pages: 141-146(2025)
Published：30 August 2025
DOI：10.47297/taposatWSP2633-456926.20250608
Accepted：

Scan QR Code

Su Xiaoru.A Physics-Guided GAN Framework for Controllable Heavy-Rain Image Synthesis: Methodology and Theoretical Analysis[J].Theory and Practice of Science and Technology,2025,06(08):141-146.
DOI：

Su Xiaoru.A Physics-Guided GAN Framework for Controllable Heavy-Rain Image Synthesis: Methodology and Theoretical Analysis[J].Theory and Practice of Science and Technology,2025,06(08):141-146. DOI： 10.47297/taposatWSP2633-456926.20250608.

摘要

Abstract

Autonomous-driving vision systems suffer substantial performance degradation in heavy-rain scenarios due to severe image degradation

while the scarcity of real-world rainy data and the prohibitive cost of pixel-level annotation hinder the training of robust perception models. To address this issue

we propose a Physics-Guided Generative Adversarial Network (PG-GAN) that enables controllable translation from clear-weather images to high-fidelity heavy-rain imagery.The model incorporates a physically grounded rain formation process—encompassing rain streaks

atmospheric haze

and windshield droplets—as explicit priors and is optimized via a composite objective integrating adversarial loss

perceptual loss

and physics-consistency loss. The synthetic rainy images are subsequently employed to augment the training set of a YOLOv5 object detector. Theoretical analysis based on prior works [5

20] predicts: (1) rain-streak orientation error reduction to <5° (3) surpassing clear-weather baseline by 15.3% and falling short of the oracle trained on real heavy-rain images by merely 0.8 %. This framework establishes a physics-grounded synthesis paradigm

with efficacy validation pending community collaboration.

关键词

Keywords

references

Y. Chen et al., "The Impact of Rain on LiDAR and Camera Performance in Autonomous Vehicles," IEEE Trans. Intell. Transp. Syst., vol. 24 , no. 7 , pp. 7230 – 7242 , Jul. 2023 .

K. Sakaridis, D. Dai, and L. Van Gool, "Map-Guided Curriculum Domain Adaptation for Semantic Segmentation of Adverse Weather Images," IEEE Trans. Pattern Anal. Mach. Intell., vol. 45 , no. 4 , pp. 4504 – 4520 , Apr. 2023 .

F. Yu et al., "BDD100K: A Diverse Driving Video Database with Scalable Annotation Tooling," arXiv: 1805.04687 , 2018.

A. Dosovitskiy et al., "CARLA: An Open Urban Driving Simulator," in Proc. Conf. Robot Learn., 2017, pp. 1 – 16 .

H. Zhang et al., "Unsupervised Rain Synthesis via Physics-Guided GAN," in Proc. IEEE ICCV, 2021, pp. 12327 – 12336 .

P. Isola et al., "Image-to-Image Translation with Conditional Adversarial Networks," in Proc. IEEE CVPR, 2017.

J. Zhu et al., "Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks," in Proc. IEEE ICCV, 2017.

R. Yasarla et al., "Uncertainty Guided Multi-Scale Residual Learning for Rain Removal," IEEE Trans. Image Process., vol. 30 , pp. 4828 – 4841 , 2021 .

Y. Wang et al., "Rain Generation and Removal as Controllable Tasks," in Proc. IEEE WACV, 2023.

J. Ho et al., "Denoising Diffusion Probabilistic Models," in Proc. NeurIPS, 2020.

K. Garg et al., "Vision and Rain," Int. J. Comput. Vis., vol. 75 , no. 1 , pp. 3 – 27 , 2007 .

S. G. Narasimhan et al., "Vision and the Atmosphere," Int. J. Comput. Vis., vol. 48 , no. 3 , pp. 233 – 254 , 2002 .

X. Li et al., "Recurrent Squeeze-and-Excitation Context Aggregation Net for Rain Removal," IEEE Trans. Image Process., vol. 30 , pp. 2548 – 2561 , 2021 .

A. Geiger et al., "Vision Meets Robotics: The KITTI Dataset," Int. J. Robot. Res., vol. 32 , no. 11 , pp. 1231 – 1237 , 2013 .

M. Johnson et al., "Transferring Simulated Images to Real-World for Autonomous Driving," in Proc. IEEE ITSC, 2020.

H. Porav et al., "Adversarial Training for Adverse Conditions: Robust Perception from Synthetic Data," in Proc. IEEE ICRA, 2020.

V. Dumoulin et al., "Feature-wise Transformations," Distill, 2018.

R. Ranftl et al., "Vision Transformers for Dense Prediction," in Proc. IEEE ICCV, 2021.

J. Johnson et al., "Perceptual Losses for Real-Time Style Transfer," in Proc. ECCV, 2016.

Y. Zhang et al., "SynRain: A Synthetic Rain Dataset for Robust Detection," in Proc. IEEE ICIP, 2022.

ISO 11452-17:2020, "Road vehicles — Component test methods for electrical disturbances from narrowband radiated electromagnetic energy — Part 17: Direct RF power injection," 2020.

W. Yang et al., "RainSyn: Realistic Rain Synthesis for Autonomous Driving Simulation," IEEE Robot. Autom. Lett., vol. 7 , no. 2 , pp. 1529 – 1536 , Apr. 2022 .

M. Heusel et al., "GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium," in Proc. NeurIPS, 2017, pp. 6626 – 6637 .

G. Jocher et al., "Ultralytics YOLOv5," 2021. [Online]. Available: https://github.com/ultralytics/yolov5.

Alert me when the article has been cited

Submit

No data

Related Author

No data

Related Institution

No data

⁰