This is the paper explanation lecture of "DeFace: Learning to Decouple the Lights for 3D Face Texture modeling", recently presented at NeurIPS 2024. Authors introduce a novel approach to model 3D facial textures under challenging illumination affected by occlusions.

00:00 - Intro
00:29 - Background / problem
02:07 - Related works
05:10 - Overall architecture
07:38 - Spherical harmonics & ray-tracing
10:28 - How to handle multiple lighting conditions on face
13:06 - Constraints & loss functions
17:28 - Validation / result
22:24 - Ablation study

Arxiv: https://arxiv.org/abs/2412.08524
Project Page: https://tianxinhuang.github.io/projec...
Code: https://github.com/Tianxinhuang/DeFace
NeurIPS Page: https://nips.cc/virtual/2024/poster/9...

Image & Equation References:
All the images or videos are taken from the project page / paper of DeFace

Sony has launched a video channel to introduce information related to Neural Network Libraries (https://nnabla.org/, https://github.com/sony/nnabla/), an open source deep learning framework software provided by Sony (https://nnabla.org/). In addition to tutorials and tips on Neural Network Libraries, we will be providing information on the latest deep learning technologies (lectures, introduction of cutting-edge papers). Please subscribe to our channel and support us!
The Neural Network Console (https://dl.sony.com/), an intuitive GUI-based deep learning development environment also provided by Sony, has a very popular YouTube channel (https://www.youtube.com/channel/UCRTV.... Do check that out as well!