GHARP: Real-Time Gaussian Head Animation via Reconstruction Prior

1 reports archived1 independent sourcesupdated 10/26/2026, 08:00:00

Synthesis & Latest Updates

1 Sources Cross-Validated

GHARP introduces a two-stage decoupled architecture that builds identity representations offline and predicts expression residuals online, enabling mobile-friendly real-time 3D head animation. By constraining expression changes within the semantic latent space of a pretrained reconstruction model, it achieves state-of-the-art quality on Ava-256 while running 13x faster with 8x fewer Gaussians.

LATEST/GHARP introduces a two-stage decoupled architecture that builds identity representations offline and predicts expression residuals online, enabling mobile-friendly real-time 3D head animation. By constraining expression changes within the semantic latent space of a pretrained reconstruction model, it achieves state-of-the-art quality on Ava-256 while running 13x faster with 8x fewer Gaussians.

Hourly heat curve

3 fully observed hours

Now

11.2

Peak

11.807:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

Apple Machine Learning ResearchT1Oct 26, 08:00·68 pts
GHARP: Real-Time Gaussian Head Animation via Reconstruction Prior
- Adopts an identity/animation decoupling design, allowing complex offline processing to enable lightweight, mobile-ready online inference
- Leverages the semantic latent space of a pretrained reconstruction model as a prior to make expression residual prediction compact and efficient
- Achieves SOTA visual quality on Ava-256 benchmark with 13x speedup on A100 GPU and 8x reduction in Gaussian primitives
View Related Reports & Analysis Original Source