I'm an undergrad studying computer science and a bit of math at UC Irvine interested in all aspects of computer vision from software to research. Particularly, at the intersection of vision, graphics, and geometry, 3D understanding presents some of the most interesting yet challenging problems (e.g., reconstruction), with many applications ranging from autonomous vehicles to medical imaging.
A real-time visual simultaneous localization and mapping (vSLAM) system for an unmanned drones. The 2D reconstruction (mapping) of the camera feed from the drone relies on visual odometry (VO) to precisely place the image together using SIFT and only uses GPS pose for initial rough placement. Well, as long it's not in a simulation with smooth textures, or else SIFT or most VO system fails. Checkout a more realistic mapping from real images:
map
A simple inverse renderer that uses signed-distance function to reconstruct the object of interest (in this case, a calculator on a newel.) The reflection on the calculator screen was hard to detect, as structure from motion (using COLMAP) failed to generate significant amount of points on there, creating a hole in the object (check out the object file below). Side note, the image is in a blue shade because of OpenCV's reverse RGB coloring scheme.
repo | calculator.obj
A Vulkan renderer out of fun to emulate Minecraft's blocky terrain through a custom graphics pipeline to learn more about how computers render 3D stuff procedurally. Speaking about procedurally, the terrain is generated by Perlin noise with multiple octaves to make it a bit more realistic.
repo
GitHub
LinkedIn
Google Scholar