# MobileNeRF: Exploiting the Polygon Rasterization Pipeline for Efficient Neural Field Rendering on Mobile Architectures

**Authors:** Zhiqin Chen, Thomas Funkhouser, Peter Hedman, Andrea Tagliasacchi  
**arXiv:** [2208.00277](https://arxiv.org/abs/2208.00277)  
**Date:** 2022-07-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2208.00277v1/

## Key points

- The paper introduces a novel approach called MobileNeRF to address limitations of existing Neural Radiance Fields (NeRFs) in synthesizing images of 3D scenes from novel views.
- NeRFs rely on specialized volumetric rendering algorithms based on ray marching, which are not well-suited for widely deployed graphics hardware.
- MobileNeRF proposes a NeRF representation based on textured polygons, enabling efficient synthesis of novel images using standard rendering pipelines.
- The NeRF is represented as a collection of polygons with textures encoding binary opacities and feature vectors.
- These polygons are rendered using a z-buffer, resulting in an image with features at every pixel.
- A view-dependent Multi-Layer Perceptron (MLP) running in a fragment shader interprets these features to generate the final pixel color.
- By leveraging the traditional polygon rasterization pipeline, MobileNeRF allows for efficient rendering on various compute platforms, including mobile phones.
- It provides interactive frame rates and real-time rendering capabilities for NeRF-based applications.

## Summary

The paper titled "MobileNeRF: Exploiting the Polygon Rasterization Pipeline for Efficient Neural Field Rendering on Mobile Architectures" introduces a novel approach to address the limitations of existing Neural Radiance Fields (NeRFs) in synthesizing images of 3D scenes from novel views. NeRFs have demonstrated impressive capabilities but rely heavily on specialized volumetric rendering algorithms based on ray marching which are not well-suited for widely deployed graphics hardware. To overcome this limitation, the authors propose a NeRF representation based on textured polygons. This representation enables efficient synthesis of novel images using standard rendering pipelines. The NeRF is represented as a collection of polygons with textures that encode binary opacities and feature vectors which are rendered using a z-buffer resulting in an image with features at every pixel. These features are then interpreted by a small view-dependent Multi-Layer Perceptron (MLP) running in a fragment shader to generate the final pixel color. By leveraging the traditional polygon rasterization pipeline which offers massive pixel-level parallelism, this new approach allows NeRFs to be rendered efficiently on various compute platforms including mobile phones providing interactive frame rates and real-time rendering capabilities for NeRF-based applications. Overall, this paper presents MobileNeRF as an innovative solution that exploits the polygon rasterization pipeline to enhance efficiency and performance of Neural Radiance Fields for synthesizing images of 3D scenes from novel viewpoints.