DeltaAge:一种高保真人脸年龄编辑网络OA

Current mainstream facial age editing methods predominantly rely on generative adversarial networks(GANs)or diffusion models,which typically necessitate training on paired age-annotated datasets.These approaches are often constrained by suboptimal generation quality and the ineffective decoupling of age-related attributes from other facial features.To address these limitations,this paper proposes a diffusion autoencoder-based approach for facial age editing.The method employs a dual-branch architecture to effectively disentangle identity-specific and age-related attributes,coupled with a novel diversity loss to enhance the expressiveness of synthesized aging patterns.Experimental results on the CelebA-HQ dataset demonstrate that DeltaAge achieves superior performance in term of the blur index,robustly validating its high-fidelity editing capabilities.