Portfolio

Manifold

A fully integrated 360◦ video camera supporting 6DoF head motion parallax requires overcoming many technical hurdles, including camera placement, optical design, sensor resolution, system calibration, real-time video capture, depth reconstruction, and real-time novel view synthesis.
Code available at: https://github.com/facebook/facebook360_dep

Intertect

An interactive learning tool for computer architecture. Through this, student will build a MIPS emulator from scratch through a web UI.
Code available at: https://github.com/yashpatel5400/intertect

Neuropath

Branch prediction has become an essential part of the CPU pipeline. This work explored using a perceptron for doing branch prediction in the Gem5 x86 CPU simulator. LTAGE and existing branch predictors were found to generally have better performance overall, albeit with slightly less accuracy in prediction in some cases tested on, than a perceptron.
Code available at: https://github.com/yashpatel5400/neuropath

MIS Ray Tracing

Implementation from scratch of multiple importance sampling for ray tracing. The renderer currently supports rendering an outdoor collection of balls and the Cornell box.
Code available at: https://github.com/yashpatel5400/raytrace-montecarlo

Synalyze

App used for analyzing business meetings and/or conversations that produces analyses on participants. Best Use of Machine Learning: HackPrinceton Spring 2017
Devpost: https://devpost.com/software/synergy (Note: the project was originally named Synergy)
Code available at: https://github.com/yashpatel5400/synalyze

Wizard Chess

Interactive 3D chess game built atop three.js.
Code available at: https://github.com/yashpatel5400/chesswizard

DeepGIF

Video style transfer using convolutional networks, with tracking and masks for GIFs
Code available at: https://github.com/yashpatel5400/DeepGIF

Manhattan Project Visualization

Visualization of the involvement of people from Princeton University in the Manhattan Project. All moving dots (after settling in their final locations) are interactive: clicking them will give a short audio clip of the person’s involvement in their final location. Note: rendering is pretty compute-intensive, so it may be choppy before settling down.
Code available at: https://github.com/yashpatel5400/manhattan