Proceedings of the International Workshop on Advances in Deep Learning for Image Analysis and Computer Vision (IWADIC 2025)

<Previous Article In Volume
Next Article In Volume>

Multi-Scale Patch Discriminator for Cycle-Consistent Unpaired Image Translation

Authors
Yichen Liu1, *
1School of Intelligent Systems Science and Engineering/JNU-Industry School of Artificial Intelligence, Jinan University, Guangdong, China
*Corresponding author. Email: lyc3407518322@stu2023.jnu.edu.cn
Corresponding Author
Yichen Liu
Available Online 24 April 2026.
DOI
10.2991/978-94-6239-648-7_84How to use a DOI?
Keywords
CycleGAN; Multi-Scale Discriminator; PatchGAN; Unpaired Image Translation; Adversarial Training
Abstract

Unpaired image-to-image translation often relies on a single-scale PatchGAN discriminator that emphasizes local textures while providing limited guidance on global structure, which can lead to shape distortion and unstable optimization at higher resolutions. This paper investigates a drop-in Multi-Scale Patch Discriminator (MSD) for Cycle-consistent translation that aggregates patch-wise scores from parallel critics operating at coarse and fine resolutions, while keeping generators and training losses unchanged. Experiments on horse ↔ zebra (2562 and 5122) under identical settings demonstrate consistent gains: at 2562, FID drops from 58.3 → 47.2 (−19%), KID from 0.046 → 0.035 (−24%), SSIM rises 0.671 → 0.709 (+5.7%), and LPIPS decreases 0.314 → 0.274 (−12.7%). The improvements arrive with sub-linear complexity growth (Params 2.7M → 5.6M; FLOPs 8.1G → 10.8G), and a compact shared-trunk variant further reduces parameters by ≈43% and FLOPs by ≈21% with negligible FID change. Multi-seed runs indicate smoother convergence and lower variance, evidencing enhanced stability. The experimental results indicate that multi-scale adversarial supervision can provide complementary guidance from global to local levels, enhancing the realism and structural fidelity of the results without modifying the generator, while also offering a practical trade-off between accuracy and efficiency for unpaired translation tasks.

Copyright
© 2026 The Author(s)
Open Access
Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

<Previous Article In Volume
Next Article In Volume>
Volume Title
Proceedings of the International Workshop on Advances in Deep Learning for Image Analysis and Computer Vision (IWADIC 2025)
Series
Advances in Computer Science Research
Publication Date
24 April 2026
ISBN
978-94-6239-648-7
ISSN
2352-538X
DOI
10.2991/978-94-6239-648-7_84How to use a DOI?
Copyright
© 2026 The Author(s)
Open Access
Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Cite this article

risenwbib
TY  - CONF
AU  - Yichen Liu
PY  - 2026
DA  - 2026/04/24
TI  - Multi-Scale Patch Discriminator for Cycle-Consistent Unpaired Image Translation
BT  - Proceedings of the International Workshop on Advances in Deep Learning for Image Analysis and Computer Vision (IWADIC 2025)
PB  - Atlantis Press
SP  - 774
EP  - 785
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6239-648-7_84
DO  - 10.2991/978-94-6239-648-7_84
ID  - Liu2026
ER  -