Research On Coloring Garment Line Drafts Based On Generative Adversarial Networks

The garment line draft is the most eloquent tool a professional has to communicate design ideas.Garment line draft with colorization is an essential and popular clothing illustration technique.In recent years,with the rapid development of deep learning,generative adversarial networks have been trained using large-scale data samples and exploiting their powerful feature extraction and learning capabilities.Great success has been achieved on the task of coloring line art images.As a result,we investigate the method of coloring garment line drafts based on generative adversarial networks.Firstly,the existing open source line drawing dataset has a simple line structure and a large gap with real photos.In contrast,the object of this paper is a show dress line image.The data we need for our experiments are the real show outfit images with a completely white background.The paired garment-line image dataset was constructed in response to the lack of a garment line draft image dataset.It is dedicated to the task of coloring garment-line drafts.In this paper,the inverse generation of real show clothing images is performed using an excellent edge detection algorithm.High-quality garment-line images were selected according to the data samples required for the experiment.This is used to construct a paired dataset of1330 pairs of real show dress-line images with pure white backgrounds.This is a good basis for later algorithmic research.Secondly,the paper propose an automated method for coloring garment line drafts.Automated methods of coloring garment line drafts can increase the efficiency of the fashion designer and reduce production costs.However,it often faces the problem of color leaks at the edges.To solve this problem,we propose an Line Colorization Generative Adversarial Networks(LGClo GAN)module for coloring,while introducing a region segmentation mechanism.This mechanism is able to provide region segmentation information for the colorization process explicitly so that the colorization model can learn to avoid assigning the same color across regions with different semantics or inconsistent colors inside an individual region.Furthermore,an Line Colorization Shadow Generative Adversarial Network(LGShd GAN)module is proposed for generating lighting effects to better depict 3D garments.The experimental results show that the introduction of a region segmentation fusion mechanism effectively mitigates the problem of edge colour leakage.It makes the background of the generated colorization images cleaner.For coloured images with low colour saturation,the LGShd GAN module generates better 3D effects.It improves the overall quality of the colorization image.Finally,we propose a user-guided approach to coloring garment line drafts.This method can effectively solve the problem that the synthesis results in a single coloring effect in the automated coloring method,which cannot meet the diverse needs of users.In this paper,we present a garment line draft colorization method based on conditional generative adversarial networks,which can support user interaction by adding scribbles to guide the colorization process.It meets the diverse needs of users.Based on the generated adversarial network architecture,the method introduce a new local feature extraction network.To enhance the generalization ability of networks trained with synthetic data.A special convolution layer is designed to increase the acceptance field and network capacity.To make the generated colorized images more natural and realistic.Furthermore,the loss function in our method is specially designed that can reduce blend and overflow in the result.At last,we use joint bilateral filter to smooth the out-put and generate a cleaner and vivid coloring sketch.Experimental results show that every module in our method can make a contribution to the final results.Moreover,the comparison with the classic approach demonstrates that our method can avoid large areas of leakage in the background and have cleaner edge for clothes.