Computer Vision
1 language| Type | Interactive demo |
|---|---|
| Origin | CSE 152A at UC San Diego |
| Developer | David |
| Written in | Python, NumPy, OpenCV, PyTorch, TypeScript |
| Tests | Live math fixture-tested against the original NumPy solutions |
| Built with | CSE 152A coursework ported to a browser stage |
| Hosted on | David-Internet/demos/vision |
| Website | https://david-internet.vercel.app/demos/vision |
Computer Vision is an interactive demonstration of David's CSE 152A coursework at UC San Diego. It is hosted on David's Internet and presents classical computer-vision methods in course order.[1]
The page covers photometric stereo, epipolar geometry, corner and feature matching, bag-of-words face classification, and archived CNN and transfer-learning results. The live mathematical components are TypeScript ports checked against the original NumPy solutions.
Overview
The geometry panel applies the normalized eight-point algorithm to the course's dino image pair, so clicks in either view draw the matching epipolar line. A corner detector and an SSD versus NCC comparison use the same images. Separate panels show a 100-word visual vocabulary and stored results from FashionMNIST and STL-10 coursework.
Photometric stereo
Four photographs of a face under known lighting are used to recover normals, albedo, and depth through per-pixel least-squares solves. A three-image or four-image comparison is available. In this view, a draggable light relights the recovered face, using the reconstructed surface rather than a new source image.
Development
The demo presents coursework from winter and spring 2025. Its photometric-stereo solver, Horn integration, eight-point solver, corner detector, and SSD and NCC implementations are fixture-tested against the NumPy solutions. Live math fixture-tested against the original NumPy solutions.
See also
References
- ^
content/vision/README.md, David's Internet.
- Interactive demos
- Computer vision
- 2026 establishments