Towards superior control in automatic face editing with generative adversarial networks

Generative Adversarial Networks (GANs) have been widely used in image manipulation tasks such as local editing and image interpolation. This project examines StyleMapGAN, a novel approach that evolves from StyleGAN by replacing AdaIN with intermediate latent space carrying information on spatial dim...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Zhang, Xijue
مؤلفون آخرون: Chen Change Loy
التنسيق: Final Year Project
اللغة:English
منشور في: Nanyang Technological University 2022
الموضوعات:
الوصول للمادة أونلاين:https://hdl.handle.net/10356/156775
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
الوصف
الملخص:Generative Adversarial Networks (GANs) have been widely used in image manipulation tasks such as local editing and image interpolation. This project examines StyleMapGAN, a novel approach that evolves from StyleGAN by replacing AdaIN with intermediate latent space carrying information on spatial dimensions, hence capable of performing high-quality local editing. In addition, by introducing a BiSeNet-based face parsing model, this project develops a fully automated process in local editing of human faces that only takes a few seconds. This project demonstrates that the face parsing model outputs masks that rivals manually labelled face datasets. Furthermore, this project explores more controls in local editing by introducing a pair of unaligned masks during stylemap mixing in W+ space in the generator. Local editing with interpolation is achieved and a demo application is developed to demonstrate the local editing process.