projects

NeRF!

Novel view synthesis using neural radiance fields. Re-implemented in PyTorch.

Grab-Cut

Segmented foreground objects automatically, as seen in Powerpoint, using color histogram-based graph-cuts (GrabCut)

Food Detection

Ranked 5/200 on an online food identification contest, using EfficientNets with MixUp data augmentation

Scene Detection

Detected outdoor scenes with with 90% accuracy using a bag of words classifier and SIFT features, on the SUN dataset

Cat GAN

Realistic cat image generation using DCGAN

2D SLAM

Recovered accurate positional information from noisy location data using non-linear pose-graph optimization

Visual Odometry

Successfully reconstructed 3D environment from a moving car using just a monocular video stream, on KITTI dataset. Triangulated image pairs for depth estimation with ICP and PnP for accurate pose estimation

Robot Navigation

Simulated RRT and MPC planning on an indoor agent and implemented collision cone multi-agent obstacle avoidance

Key-Value Storage

Built a compressed trie with memory optimizations for key-value API transactions

Multiproc Syncing

Several synchronization and locking techniques for multi-process and multi-thread applications using mutex locks and semaphores

Shopping App

Developed an intuitive online shopping app with MERN stack, enabling bulk group purchases and multiple retailers

Modified xv6

Extended MIT's xv6 operating system to include new scheduling algorithms

Bash Shell

Reimplemented POSIX bash shell functionality, supporting process management, redirection, built-in commands & more

Citation Intent Detector

Automatically detected intent behind citations in literature using transfromer models, based on the 3C SDP shared task

Wikipedia Search Engine

Developed a highly efficient search engine for Wikipedia articles, returning results in < 1 second for 80GB of XML

vision