Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
Jun-Yan Zhu empty Taesung Park Phillip Isola Alexei A. Efros Berkeley AI Research (BAIR) laboratory, UC Berkeley
Abstract
Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain $X$ to a target domain $Y$ in the absence of paired examples. Our goal is to learn a mapping $G:X\rightarrow Y$ such that the distribution of images from $G(X)$ is indistinguishable from the distribution $Y$ using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping $F:Y\rightarrow X$ and introduce a cycle consistency loss to enforce $F(G(X))\approx X$ (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach.
中文速览
让图像在两种风格或域之间互相转换（图像翻译，image-to-image translation）通常需要大量成对的"原图—目标图"训练数据，但现实中这类配对数据往往难以获取。这篇论文提出了 CycleGAN，其核心思路是同时训练两个方向的生成器（G: X→Y 和 F: Y→X），并引入"循环一致性损失"（cycle consistency loss）来约束两个生成器互为逆映射——也就是说，一张图从 X 翻译到 Y 再翻译回来，应该还原为原图，反之亦然。这一设计解决了单纯对抗训练容易出现模式崩塌、且无法保证输入输出有意义对应的问题。实验表明，该方法在莫奈画风迁移、马↔斑马转换、季节变换、照片增强等多项无配对数据任务上均优于此前方法。这项工作的重要性在于，它让图像风格或内容的跨域转换不再依赖昂贵的人工标注配对数据，极大拓宽了图像生成与编辑技术的实用范围。
原文 arXiv:1703.10593;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1703.10593v7