Understand this story
Apple Machine Learning Research proposes GHARP, a real-time Gaussian head animation method that can generate 3D human head animations in real time from a small number of input images and driving expression signals. The method splits the problem into two stages: offline construction of identity geometry and appearance representation, and at runtime a lightweight network predicts expression-related residuals, while a body alignment network is introduced to eliminate ambiguity caused by body pose and clothing position. On the Ava-256 benchmark, GHARP achieves SOTA, with runtime speed improved by up to 13 times on an A100 GPU and the number of Gaussians reduced by 8 times. There are currently no follow-up developments.
AI-generated from public reports
Latest development · Report
GHARP: Real-time Gaussian Head Animation Based on Large-scale Reconstruction PriorsDevelopments
1 confirmed developments
Apple proposes GHARP, a real-time Gaussian head animation method
Source:Apple Machine Learning Research
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 27
Apple Machine Learning ResearchFirst-party
GHARP: Real-time Gaussian Head Animation Based on Large-scale Reconstruction Priors
Apple proposes GHARP, which can generate 3D human head animations in real time from a small number of input images and driving expression signals, achieving SOTA on the Ava-256 benchmark, with runtime speed up to 13 times faster on an A100 GPU and 8 times fewer Gaussians. The method splits the problem into two stages: offline construction of identity geometry and appearance representation, and runtime prediction of expression-related residuals using a lightweight network, and introduces a body alignment network to eliminate ambiguity caused by body pose and clothing position.