BabyGAN for Facial Contour Reversion: AI Course Applications Using U-Net Architecture
Huan Yao, Wanying Bao, Hequn Wu · 2024
BabyGAN is a generative adversarial network model based on the U-net architecture and residual networks, designed for the transformation of adult facial contours into baby facial contours. This model is primarily used in the artificial intelligence practical courses of vocational colleges, focusing on information technology education. A key feature of BabyGAN is its ability to help students deeply understand the logical structure of generative adversarial networks while also fostering their enthusiasm. Compared to other GAN models used in teaching, BabyGAN has the advantage of allowing students to easily construct training sets, enabling immediate use of class-specific datasets in the classroom. The working principle of BabyGAN involves integrating a special keypoint constraint loss function in the generator to produce high-quality baby facial contours, thereby transforming adult facial contours into corresponding high-quality infant facial contours. BabyGAN plays a significant role in the future work scenarios of students. Incorporating the BabyGAN course into AI curricula effectively blends information technology education with students' future career prospects. The application of GANs to meet user needs is a current research hotspot. Introducing students to BabyGAN in a fun and educational setting also aligns with market demands, equipping students with practical skills and the ability to apply these skills in future work contexts.