# Instant Volumetric Head Avatars

**Authors:** Wojciech Zielonka, Timo Bolkart, Justus Thies  
**arXiv:** [2211.12499](https://arxiv.org/abs/2211.12499)  
**Date:** 2022-11-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2211.12499v2/

## Key points

- The paper introduces a method called INSTA for instantly reconstructing photo-realistic digital avatars.
- INSTA utilizes a dynamic neural radiance field based on neural graphics primitives integrated around a parametric face model.
- Unlike traditional methods, INSTA can generate a digital avatar in less than 10 minutes using modern GPU hardware, significantly faster than previous solutions.
- Key innovation of INSTA: ability to render interactive poses and expressions in real-time by training on a single monocular RGB portrait video capturing various facial expressions and viewpoints.
- INSTA leverages the geometry prior of the underlying parametric face model to extrapolate to unseen poses, creating highly realistic avatars with unparalleled rendering quality and efficiency.
- In studies conducted, INSTA outperformed state-of-the-art methods in rendering quality and training time, demonstrating its effectiveness in generating detailed digital avatars reflecting human facial expressions and movements accurately.
- With speed, accuracy, and versatility, INSTA represents a significant advancement in digital avatar reconstruction with potential to revolutionize applications from virtual reality to teleconferencing.

## Summary

The paper "Instant Volumetric Head Avatars" introduces a groundbreaking method called INSTA for instantly reconstructing photo-realistic digital avatars. Developed by Wojciech Zielonka, Timo Bolkart, and Justus Thies, this approach utilizes a dynamic neural radiance field that is based on neural graphics primitives integrated around a parametric face model. Unlike traditional methods that can take several days to train an avatar, INSTA can generate a digital avatar in less than 10 minutes using modern GPU hardware, making it significantly faster than previous solutions. The key innovation of INSTA lies in its ability to render interactive poses and expressions in real-time. By training the pipeline on a single monocular RGB portrait video capturing various facial expressions and viewpoints of the subject, INSTA leverages the geometry prior of the underlying parametric face model to extrapolate to unseen poses. This capability allows for the creation of highly realistic avatars with unparalleled rendering quality and efficiency. In quantitative and qualitative studies conducted on diverse subjects, INSTA outperformed state-of-the-art methods in terms of both rendering quality and training time. The results demonstrate the effectiveness of this novel approach in rapidly generating detailed digital avatars that accurately reflect the nuances of human facial expressions and movements. With its speed, accuracy, and versatility, INSTA represents a significant advancement in the field of digital avatar reconstruction and has the potential to revolutionize applications ranging from virtual reality to teleconferencing.