Enhancing Face Angle Transformation through Progressive Face Pose Transfer Using Multiple Discriminators

This paper aims to tackle the challenge of face angle transfer. In this research, we introduce a novel approach utilizing a progressive generator in conjunction with multiple discriminators to iteratively transform faces to desired angles. The proposed architecture leverages the Pose-Attentional Transfer Network for progressive face angle manipulation and integrates three discriminators. These discriminators have distinct purposes. The first discriminator focuses on learning angle discrepancies. The second discriminator enhances diversity in facial structures and local perceptual information. The third discriminator elevates the generation quality of key facial regions. With this framework, users can rotate a face image to the desired orientation by providing the original face image and a target face angle. Experimental results demonstrate the efficacy of our method on various performance metrics.