基于CycleGan的山水画风格迁移
Landscape Painting Style Migration Based on CycleGan
摘要: 山水画作为中国传统绘画的独特艺术形式，一直以来都备受关注。在互联网流行的今天，通过社交媒体分享的图片，与用于各种场合的图集占据了互联网很大一部分空间。如何有效利用这一庞大的图片资源，并与现有的技术结合，设计出具有实际意义的作品是本文的主要目的。利用GANs技术设计生成的自主学习模型机，可以在大量的同类别的数据集中寻找到人眼难以发现的规律，并加以学习，生成以分析得到的规律为基础的新图像。本研究旨在利用CycleGAN方法进行山水画风格的迁移。我们构建了一个包含大量山水画和普通照片的训练数据集，并使用CycleGAN模型进行训练。通过生成器和判别器之间的对抗学习，我们实现了从普通照片到山水画和从山水画到普通照片的双向风格迁移。该技术在数据集充裕的情况下，可以对数据进行预测，对人眼难以发现的特征进行提取，对大数据进行整合等功能，有着很大的应用前景。实验结果表明，我们的方法在实现山水画风格迁移方面取得了显著的成果。生成的图像在保持了原始照片的内容的同时，成功地融合了山水画的独特风格，具有艺术性和视觉吸引力。并且还通过定量和定性的评估指标对生成图像进行了评估，验证了方法的有效性和稳定性。此外，本研究创新性具体表现在动态风格迁移：基于CycleGAN的山水画风格迁移还可以实现动态风格迁移，即将不同时间段的山水画进行转换，以呈现出不同的季节或氛围。这一方法在艺术创作、设计和虚拟现实等领域具有广泛的应用潜力。通过将传统山水画风格与现代照片图像进行结合，为研究图像生成技术的发展和艺术应用领域的拓展带来了新的可能性。
Abstract: As a unique art form of traditional Chinese painting, landscape painting has always attracted much attention. With the popularity of the Internet today, pictures shared through social media and atlases used in various occasions occupy a large part of the Internet space. How to make effective use of this huge picture resource and combine it with the existing technology to design works with practical significance is the main purpose of this article. The self-learning model machine designed and generated by using GANs technology can find rules that are difficult for human eyes to find in a large number of data sets of the same category, and learn them to generate new images based on the rules obtained by analysis. This study aims to use the CycleGAN method for landscape painting style transfer. We construct a training dataset containing a large number of landscape paintings and general photos, and use the CycleGAN model for training. Through adversarial learning between the generator and the discriminator, we achieve bi-directional style transfer from ordinary photos to landscape paintings and from landscape paintings to ordinary photos. In the case of abundant data sets, this technology can predict data, extract features that are difficult for human eyes to find, and integrate big data, which has great application prospects. Experimental results show that our method achieves remarkable results in achieving landscape painting style transfer. The resulting image successfully incorporates the unique style of landscape painting for artistry and visual appeal while maintaining the content of the original photo. And the generated images are also evaluated by quantitative and qualitative evaluation indicators, verifying the effectiveness and stability of the method. In addition, the innovation of this research is embodied in the dynamic style transfer: the landscape painting style transfer based on CycleGAN can also realize the dynamic style transfer, that is, to convert the landscape paintings of different time periods to present differ-ent seasons or atmospheres. This approach has potential for a wide range of applications in areas such as art creation, design, and virtual reality. By combining the traditional landscape painting style with modern photographic images, it brings new possibilities for the development of research image generation technology and the expansion of artistic application fields.
文章引用：吕泽华, 唐嘉晨, 王琦媛, 谢文誉, 刘懿峤. 基于CycleGan的山水画风格迁移[J]. 计算机科学与应用, 2023, 13(5): 973-980. https://doi.org/10.12677/CSA.2023.135095

参考文献

[1] Zhu, J.Y., Park, T., Isola, P. and Efros, A.A. (2017) Unpaired Image-to-Image Translation Using Cycle-Consistent Ad-versarial Networks. Proceedings of the IEEE International Conference on Computer Vision (ICCV), Venice, 22-29 Oc-tober 2017, 2223-2232. [Google Scholar] [CrossRef] 
[2] Kim, T., Cha, M., Kim, H. and Lee, J.K. (2017) Learning to Discover Cross-Domain Relations with Generative Adversarial Networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, 21-26 July 2017, 3799-3807.
[3] Liu, M.Y., Breuel, T. and Kautz, J. (2017) Unsupervised Image-to-Image Translation Networks. Advances in Neural Infor-mation Processing Systems (NIPS), Long Beach, 4-9 December 2017, 700-708.
[4] Huang, X., Liu, M.Y., Belongie, S. and Kautz, J. (2018) Multimodal Unsupervised Image-to-Image Translation. Proceedings of the European Conference on Computer Vision (ECCV), Munich, 8-14 September 2018, 172-189. [Google Scholar] [CrossRef] 
[5] Zhang, H., Xu, T., Li, H., Zhang, S., Wang, X., Huang, X. and Metaxas, D. (2018) StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 41, 2513-2527. [Google Scholar] [CrossRef] 
[6] Bousmalis, K., Silberman, N., Dohan, D., Erhan, D. and Krishnan, D. (2018) Unsupervised Pixel-Level Domain Adaptation with Generative Adversarial Networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, 18-22 June 2018, 3722-3731. [Google Scholar] [CrossRef] 
[7] Liu, M.Y. and Tuzel, O. (2018) Coupled Generative Ad-versarial Networks. Advances in Neural Information Processing Systems (NIPS), Montréal, 3-8 December 2018, 469-477.
[8] Yi, Z., Zhang, H., Tan, P. and Gong, M. (2017) DualGAN: Unsupervised Dual Learning for Im-age-to-Image Translation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, 21-26 July 2017, 2868-2876. [Google Scholar] [CrossRef] 
[9] Zhang, X., Zhu, C. and Heng, P.A. (2019) Translating and Segmenting Multimodal Medical Volumes with Cycle- and Shape-Consistency Generative Adversarial Network. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, 16-20 June 2019, 9242-9251. [Google Scholar] [CrossRef] 
[10] Wang, X., Girshick, R., Gupta, A. and He, K. (2018) Non-Local Neural Networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, 18-22 June 2018, 7794-7803. [Google Scholar] [CrossRef] 
[11] Li, Y., Fang, C., Yang, J., Wang, Z. and Lu, X. (2019) Diverse Image Synthesis from Semantic Layouts via Conditional IMLE. Proceedings of the IEEE Conference on Computer Vi-sion and Pattern Recognition (CVPR), Long Beach, 16-20 June 2019, 8784-8793. [Google Scholar] [CrossRef] 
[12] Liu, X., Shen, X. and Huang, T.S. (2018) A Neural Representation of Sketch Drawings. Proceedings of the European Conference on Computer Vision (ECCV), Munich, 8-14 September 2018, 430-446.
[13] Choi, Y., Choi, M., Kim, M., Ha, J.W., Kim, S. and Choo, J. (2018) StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, 18-22 June 2018, 8789-8797. [Google Scholar] [CrossRef] 
[14] Zhang, R., Isola, P. and Efros, A.A. (2018) Colorful Image Color-ization. Proceedings of the European Conference on Computer Vision (ECCV), Munich, 8-14 September 2018, 649-666. [Google Scholar] [CrossRef] 
[15] Wu, Y. and He, K. (2018) Group Normalization. Proceedings of the European Conference on Computer Vision (ECCV), Munich, 8-14 September 2018, 3-19. [Google Scholar] [CrossRef] 
[16] Liu, X., Liang, X., Liu, L., Shen, X., Yang, M.H. and Huang, T.S. (2017) Generative Adversarial Training for Visual Storytelling. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, 21-26 July 2017, 834-842.
[17] Zhang, H., Goodfellow, I., Metaxas, D. and Odena, A. (2018) Self-Attention Generative Adversarial Networks. Advances in Neural Information Processing Systems (NIPS), Montréal, 3-8 December 2018, 7424-7435.
[18] Yu, J., Lin, Z., Yang, J., Shen, X., Lu, X. and Huang, T.S. (2018) Generative Image Inpainting with Contextual Attention. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, 18-22 June 2018, 5505-5514. [Google Scholar] [CrossRef]