Efficient 3D Gaussian Head Avatars for Edge Devices
Generative 3D Gaussian head avatars provide high-quality, efficient rendering, but synthesising the Gaussian representation remains computationally expensive, limiting deployment on resource-constrained and edge devices.
Key points
- We introduce an efficient generator architecture for unconditional 3D Gaussian head synthesis, based on a parameter-efficient synthesis block and depth-wise separable convolutions while retaining style-based conditioning.
- Our architecture reduces generator complexity without requiring model compression or quantisation.
- We further demonstrate practical CPU inference and browser-based execution on mobile devices using ONNX Runtime, enabling 3D Gaussian avatar synthesis without dedicated GPU hardware or application-specific software.
- In addition to conventional image-quality metrics, we evaluate multi-view consistency, training cost, and deployment performance.
Sources (1)
- [1]Efficient 3D Gaussian Head Avatars for Edge DevicesarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 10:44 AM
Generative 3D Gaussian head avatars provide high-quality, efficient rendering, but synthesising the Gaussian representation remains computationally expensive, limiting deployment on resource-constrained and edge devices.
We introduce an efficient generator architecture for unconditional 3D Gaussian head synthesis, based on a parameter-efficient synthesis block and depth-wise separable convolutions while retaining style-based conditioning.
Extractive summary: sentences quoted from the sources.