James Noeckel's Homepage

James Noeckel

Email: jamesn8@cs.washington.edu

CV: [pdf]

About

I am a PhD student working at the GRAIL lab in the Paul G. Allen School of Computer Science and Engineering at the University of Washington, advised by Brian Curless and Adriana Schulz. My research has focused reverse engineering 3D designs of objects through the use of domain-specific geometry representations. In particular, I am fascinated by the ability of such representations to enable precise reconstruction with incomplete observations, especially alongside advances in deep learning techniques for 3D vision / shape modeling. In previous work, for example, I have used a parts-based model of carpentry to recover carpentry assembly instructions from photographs of objects; more recently, I’ve applied a view-centric boundary representation to facilitate single-view reconstruction of CAD models  from RGB-D images. More generally, I am interested in techniques to facilitate computer-aided design, which includes analyzing motion degrees of freedom from geometry, and building robust referencing schemes to facilitate collaborative CAD workflows.

I obtained my B.A. in 2017 from Cornell University in computer science with a minor in physics, where I did research with Kavita Bala on photorealistic cloth rendering (providing the realtime shader implementation) and Timur Dogan on large-scale light radiance simulation in urban environments. During my internships at NVIDIA, I have worked on problems ranging from 3D reconstruction and robotics navigation to volume path tracing, and later at Meta, I developed a geometry processing pipeline to generate 3D-printable personalized smart glasses. I have retained a particular interest in realtime graphics and physics simulation techniques thanks to my graphics / physics background at Cornell, with various side projects which can be seen on my Shadertoy account, linked below.

[ CV ]

[ ORCID ]

[ Github ]

[ Shadertoy ] [my shader which was featured as shader of the week]

[ Linkedin] 

 

Research

Reverse Engineering B-Reps from (imperfect) single-view RGB-D Images
[under review; not the working title]

I have recently submitted work on image-based reconstruction of CAD B-Reps using deep learning in conjunction with geometric optimization. More info soon…

B-Rep Matching for Collaborating Across CAD Systems

Collaboration across CAD systems requires maintaining consistent references to topological entities (faces, vertices, edges); however, the internal referencing schemes CAD systems use are not usable in collaborative workflows involving the sharing of exported B-rep geometry. We developed a machine learning-guided matching algorithm to reconstruct these references across different versions of exported CAD models in the B-rep format to facilitate such workflows, and enables new directions in CAD editing/manipulation and shape correspondence.

  • *Ben Jones, *James Noeckel, *Milin Kodnongbua, Ilya Baran, Adriana Schulz. “B-rep Matching for Collaborating Across CAD Systems”. ACM Transactions on Graphics (SIGGRAPH 2023). [Paper]

Inferring Motion in Assemblies

I presented my work on inferring motion degrees of freedom in mechanical assemblies at the ICML Workshop on Machine Learning in Computational Design (and also at a K-12 computer science outreach event). [Paper]

Reverse Engineering Carpentry

I have developed a method for recovering a rich, part-based description of carpented objects, requiring only a collection of images, such as those taken by a phone camera, as input. This work was published at SGP and featured in the New Scientist magazine.

  • James Noeckel, Haisen Zhao, Brian Curless, and Adriana Schulz. “Fabrication-Aware Reverse Engineering for Carpentry”. Eurographics Symposium on Geometry Processing 2021. [Project page]

Diminished Reality

A collaboration with Edward Zhang, this project aims to realistically re-render indoor environments with objects removed, which involves synthesizing the missing background of the scene (inpainting) and accounting for effects on lighting (inverse illumination). I contributed an improved inpainting technique for reconstructing occluded objects and textures.

Cloth Rendering (undergraduate research assistant)

I developed a realtime visualization tool for cloth renderings using scanned volumes of cloth microstructure.

  • Pramook Khungurn, Rundong Wu, James Noeckel, Steve Marschner, and Kavita Bala. “Fast Rendering of Fabric Micro-Appearance Models Under Directional and Spherical Gaussian Lights”. ACM Transactions on Graphics (SIGGRAPH Asia 2017). [Paper]

Memberships/Awards

  • UW Reality Lab Researcher – 2019-2022
  • Wissner-Slivka Endowed Fellowship – 2017-2018
  • Member of the Phi Beta Kappa Society