GHR-2D: Gaze and head redirection via disentanglement and diffusion for gaze estimation

Open Webpage

Daosong Hu, Mingyue Cui, Kai Huang

Published: 2025, Last Modified: 12 Nov 2025Eng. Appl. Artif. Intell. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•We propose to use a latent diffusion model to edit gaze-related features and generate high-quality data with wide coverage.•We propose a training pipeline for disentanglement of gaze, head posture, and identity information embedded in latent space.•We design encoder and decoder structures for aligning vector projection and embedding.•Relative gaze is used to eliminate differences in anatomical structure and achieve gaze-invariant head rotation.