Differential Activation-Based Facial Attribute Editing Using Semantic Segmentation

Face attribute editing aims to change various attributes in the face image and save the specific information of the image, such as hair accessories, jewelry, and background. The most effective approach is differential activation, which aims to accurately activate and edit specific regions of face images during editing. The activation area of the existing differential activation modules often exceeds the scope of the face, resulting in the loss of image information around the face. To solve the above problems, we propose a novel segmentation-based differential activation module, which can more accurately lock the activation position within the face and effectively preserve other information. Experimental results outperform state-of-the-art editing methods in both qualitative and quantitative evaluations.