Xinge Yang (杨辛格)
I am a Ph.D. student at VCC Computational Imaging Group at KAUST, working with Prof. Wolfgang Heidrich. My research interests include differentiable optical simulation, computational imaging/photography, and ray-tracing based rendering.
I like cats, cameras, new techs, and most of sports (especially basketball, hiking and marathon). I have lived in China, Singapore, Saudi Arabia, and the US.
I am maintaining an awesome differentiable ray tracer DeepLens. It can simulate raw captures of a camera lens and optimize it. Feel free to drop me an email if you are interested!
Email  / 
CV  / 
Github  / 
LinkedIn  / 
知乎
|
|
News
03/2024: DeepLens is open-sourced, build your End-to-End lens design pipeline with 5 lines of Python code!
10/2023: I started my internship at Meta, working on gradient-based optical design for AR waveguide.
09/2023: We released an automated lens design demo AutoLens .
07/2023: Our paper, "Aberration-Aware Depth-from-Focus" is accepted by ICCP and TPAMI.
|
Experience
05/2022 - present: Ph.D. student in Computer Science, KAUST, Thuwal, Saudi Arabia.
10/2023 - 01/2024: Research Scientist Intern, Optics and Display Research, Meta Reality Lab, Redmond, WA.
08/2020 - 05/2022: M.Sc. in Computer Science, KAUST, Thuwal, Saudi Arabia.
09/2016 - 06/2020: B.Sc. in Physics (major) and Computer Science (minor), USTC, Hefei, China.
|
Research
I do differentiable optical simulation, and apply it to End-to-End lens design and computational photography.
|
|
Image Quality Is Not All You Want: Task-Driven Lens Design for Image
Classification
Xinge Yang, Qiang Fu, Yunfeng Nie, Wolfgang Heidrich
Arxiv prepint. paper, supp
A new End-to-End optical design methedology: use a well-trained network as objective.
Lens design can start from scratch and gets a larger design space. Our lenses have higher image classification accuracy with fewer elements.
|
|
Aberration-Aware Depth-from-Focus
Xinge Yang, Qiang Fu, Mohamed Elhoseiny, Wolfgang Heidrich
ICCP & TPAMI 2023. paper, supp, project, code
Train a network to represent spatially-varying focus-dependent (4D) PSFs for a real camera lens.
Improve the generalizability of depth-from-focus network models by considering optical aberrations during the training.
|
|
Curriculum Learning for ab initio Deep Learned Refractive Optics (DeepLens)
Xinge Yang, Qiang Fu, Wolfgang Heidrich
Arxiv prepint. paper, supp, code, video
OSA Imaging and Applid Optics Congress - Computational Optical Sensing and Imaging (COSI) 2022 (oral): paper, project, code
Differentiable (gradient-based) ray tracing will be a game-changer for optical design.
Fully automated design for both classical and computational lenses from scratch.
|
|