Research on Clothing Sketch Conversion Based on Generative Adversarial Networks
(1.School of Artificial Intelligence, Wuhan Technology and Business University, Wuhan 430065, China; 2.Institute of Information and Intelligent Engineering Applications, Wuhan 430065, China; 3.Academic Affairs Division, Wuhan University of Science and Technology, Wuhan 430081, China)
Abstract
The image conversion of clothing sketches has enormous application value in fields such as visual editing, clothing design, and human-computer interaction. At present, most of the objects converted from sketches are relatively simple shapes, such as faces, human postures, animal postures, shoes, and bags. The work based on sketch conversion is difficult to handle images with complex texture structures, making it challenging to solve the task of image conversion for clothing sketches. To solve this problem, an application research on sketch to clothing image conversion is proposed based on the above work. Through the reverse generation of clothing simulation sketches using the U2-Net network model, the construction of the clothing sketch dataset is completed. Then, the improved BicycleGAN network is used for training to achieve sketch conversion. Finally, the experimental effects before and after the improvement are compared. The experiment shows that the improved BicycleGAN network model has significant improvement compared to before, and with the improved network proposed in this paper, the task of converting clothing sketches to real clothing images can be effectively completed.
Keywords
image conversion; clothing sketch; Generative Adversarial Networks
References
- [1] 陈健.基于手绘草图的图像检索研究 [D].杭州：浙江工业大学，2020.
- [2] 杨金凯，王国中，范涛.基于神经网络的手绘草图的识别与匹配 [J].智能计算机与应用，2021，11（6）：148-152.
- [3] ISOLA P，ZHU J Y，ZHOU T，et al. Image-to-Image Translation with Conditional Adversarial Networks [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu：IEEE，2017：1125-1134.
- [4] ZHU J Y，ZHANG R，PATHAK D，et al. Toward Multimodal Image-to-Image Translation [J]. Advances in Neural Information Processing Systems，2017，30：465-476.
- [5] 李林，王家华，周晨阳，等.标检测数据集研究综述 [J].数据与计算发展前沿：中英文，2024，6（2）：177-193.
- [6] 李翠锦，瞿中.基于深度学习的图像边缘检测算法综述 [J].计算机应用，2020，40（11）：3280-3288.
- [7] QIN X，ZHANG Z，HUANG C，et al. U2-Net: Going Deeper with Nested U-Structure for Salient Object Detection [J/OL].Pattern Recognition，2020，106：107404[2024-09-03]. https://doi.org/10.1016/j.patcog.2020.107404.
- [8] 任欢，王旭光.注意力机制综述 [J].计算机应用，2021，41（S1）：1-6.
- [9] 周青伟. 基于自注意力机制的图像识别技术研究与实现 [D].北京：北京邮电大学，2023.
- [10] OKTAY O，SCHLEMPER J，FOLGOC L L，et al. Attention U-Net: Learning Where to Look for the Pancreas [J/OL].arXiv:1804.03999 [cs.CV].[2024-11-06].https://arxiv.org/abs/1804.03999.
Funding
武汉工商学院校级基金项目（A2023017）
About the authors
林豪发（1994.01—），男，汉族，浙江台州人，助教，硕士，研究方向：计算机视觉；罗艳玲（1982.06—），女，汉族，湖北宜昌人，副教授，硕士，研究方向：计算机视觉；卢娜（1983.04—），女，汉族，湖北武汉人，讲师，硕士，研究方向：计算机视觉；胡定兴（1996.10—），男，汉族，江西南昌人，助教，硕士，研究方向：深度学习、自然语言处理；朱大琴（1981.06—），女，汉族，湖北天门人，讲师，硕士，研究方向：思想政治教育研究、教学研究与教学改革。