Real-Time Body Modeling and Clothing Adaptation Based on Generative Adversarial Network

In order to solve the time-consuming and inconvenient problems in the traditional fitting process, this study focuses on the virtual fitting technology in embodied design, aiming to achieve real-time body modeling and clothing adaptation through GAN (generative adversarial network). This paper first collects a large amount of three-dimensional data of human bodies with different heights, weights, body shapes, etc. Then, Blender is used to build a three-dimensional human body model that is highly similar to the real human body based on these three-dimensional data. Then, the GAN model is trained with the collected data, so that the generator learns to generate images that are no different from the real fitting scene, and the discriminator is responsible for identifying the authenticity of the generated images. The simulation effect is improved through the confrontation optimization between the two; the OpenPose human posture estimation model is used to quickly and accurately extract posture key points and posture information from the input human body image; and U-Net is used to divide the human body area, so that the clothing can be accurately positioned and naturally fit during virtual fitting. According to the human body posture information and key point coordinates, the size, shape and position of the clothing model are adjusted. Finally, the mass spring model is used to simulate the texture and drape of the real fabric, and the clothing model and the human body model are realistically rendered to generate a virtual fitting effect. The virtual fitting technology based on GAN is superior to the traditional fitting method in terms of real-time, size adaptation and fit. The average fitting time of GAN fitting technology is 3.06 seconds, which is much lower than the 14.84 seconds of traditional fitting; the size deviation range is between 0.55 cm and 1.29 cm, which is smaller than the 1.53 cm to 2.4 cm of traditional fitting; the fit score generally exceeds 9 points, which is excellent and stable. This study not only provides new ideas and methods for the development of virtual fitting technology but also provides strong technical support and theoretical reference for the digital transformation and intelligent upgrading of the apparel industry.