NeRF stands for Neural Radiance Fields. It is a computer-graphics technique for 3D scene reconstruction and rendering. It was introduced in a 2020 paper by Benjamin Mildenhall and others. NeRF aims to solve some problems in traditional multi-view-image-based 3D reconstruction, such as view dependency and unrealistic synthesis of novel viewpoints.

Summary PPT: NeRF and Human Reconstruction

Based on online materials and my own understanding, I made a PPT:

Instant Volumetric Head Avatars Paper Analysis

Overview

This paper introduces work that combines Instant-NGP-accelerated NeRF with a parametric human model.

This article tries to introduce its principle and run the code provided by the authors.

INSTA paper homepage: https://zielon.github.io/insta/

Code: https://github.com/Zielon/INSTA

Principle Analysis

Based on online materials and my own understanding, I made an analysis PPT:

Compiling instant-ngp

https://blog.csdn.net/zfjbit/article/details/129918801

Principles

Deformation gradient:

https://www.continuummechanics.org/deformationgradient.html

Calling RT cores through CUDA for acceleration:

GTC 2020: Working with Tensor Cores and RTCore for compute

Code

I planned to try running it in WSL2.

Configure the Vulkan environment:

https://zhuanlan.zhihu.com/p/576320322

In the end, I ran it on Windows.

References

What Is NeRF?

https://zhuanlan.zhihu.com/p/380015071

Volume Rendering

NeRF introduction: Volume Rendering

https://zhuanlan.zhihu.com/p/260540982

Instant NGP

From NeRF to Instant-NGP

https://zhuanlan.zhihu.com/p/575269971

Applications of NeRF in Human Reconstruction

INSTA

https://zielon.github.io/insta/

https://zhuanlan.zhihu.com/p/590379734

3D Gaussian

3D Gaussian source-code reading

https://blog.csdn.net/Sakuya__/article/details/135376331

Training:

https://blog.csdn.net/gootyking/article/details/134762792