LightFields.jl: Fast 3D image reconstruction for VR applications
summary
Claims a novel sub-Nyquist reconstruction that recovers depth maps from a limited linear sequence of views, not just more examples. Method uses shearlet inpainting to fill sparse epipolar-plane images, then line detection to get slopes→depth; demonstrates on a 101-image Disney Research sequence, detecting about 340 trackable points. Implementation mixes Julia + Python + OpenCV and a Raspberry Pi prototype with a motorized linear stage; prototype works but is early and crashes sometimes.
AI-written summary · about
chapters (10)
- 0:00LightFields.jl intro
- 1:51What is a light field
- 5:32Epipolar geometry & depth
- 8:44Sub-Nyquist inpainting
- 13:20Shearlet compression
- 16:59Depth map pipeline
- 22:24Project motivation & demo
- 27:07Raspberry Pi cost
- 28:54Future rendering plans
- 29:38Q&A
related talks
41mComputer Vision for Remote AR
35mFast Multidimensional Signal Processing using Julia with Shearlab.jl
15mDepthLab: Real-time 3D Interaction with Depth Maps for Mobile Augmented Reality
15mDepthLab: Real-time 3D Interaction with Depth Maps for Mobile Augmented Reality
38m3D Face Tracking and Reconstruction using Modern C++
18mLightning Talk: Turning the Internet Into 3D: 3D Content with NeRF and Gaussian Splat...
20mLightning Talk: Beyond Polygons: The Future of 3D Graphics with Implicit Models
48m(Deep) Learning to Fly