Researches On Garment Appearance Intelligent Design Based On Deep Learning

The clothing appearance design is one of the key part of clothing design,which makes an important contribution to the popularity and overall sales of clothing.It mainly includes three aspects: the local attribute design of clothing,the local decoration design of clothing and the overall style design of clothing.However,the current clothing local attribute design method based on WGAN has poor attribute generation performance;The local pattern design method of clothing relies heavily on designers and lacks an end-to-end intelligent aided design scheme;The DCGAN based fashion overall style design method shows a fuzzy style transfer performance.Aiming at the above key technical problems,this paper is focus on the following research.A clothing local attribute editing algorithm based on the improved WGAN is proposed.This paper analyzes the problem of poor attribute editing performance of the original WGAN algorithm due to the insufficient ability of feature extraction and network representation.In this paper,the residual structure is used to optimize the feature extraction ability,and the perceptual loss function is used to optimize the attribute learning process,which significantly improves the performance of attribute generation.The experiments show that our algorithm can generate and edit a variety of clothing local attributesA local pattern generation algorithm based on the improved style transfer network is proposed.This paper analyzes the problem that the current fashion pattern design not only relies heavily on the designer's experience and inspiration,but lacks intelligent generation.We use the real-time style transfer network to generate new fashion patterns intelligently,and then seamlessly fit them to the style-less clothing,to realize the endto-end intelligent design of clothing local patterns.This paper improves the normalization method and loss function of the real-time style transfer network to make the generated local pattern more realistic.Finally,the experiments show that the proposed algorithm can effectively realize the intelligent design of local patterns.An improved overall style transfer of clothing based DCGAN is proposed.This paper analyzes the problem that the original DCGAN algorithm is difficult to learn the class differences of various styles by using the cross entropy loss function,which result in the ambiguity of style transfer effect.Therefore,we use the contrastive loss to fully guide the differences between the learning styles.Then we strengthen the feature extraction network to improve the overall style transfer effect.In the last,the comparative experiments verify the feasibility of this algorithm.An intelligent aided clothing appearance design software is developed.The demand analysis and system design of this software are introduced.It shows the construction process and related functions of the clothing appearance intelligent design software.Then we verify these clothing appearance intelligent design algorithms proposed in our paper.We prove these algorithms can help the designer carry out clothing appearance intelligent design conveniently.