San Jose, California, United States
Siri: Currently working on leveraging foundation model applications to enhance Siri experience. SPG: As part of the Perception Team at SPG, I worked on solving perception problems for Autonomous Systems using a combination of Computer Vision, Deep Learning and Machine Learning techniques.
Publication: Parita Pooj, Michael Grossberg, Peter N. Belhumeur, and Shree K. Nayar. "The Minimalist Camera." In British Machine Vision Conference (BMVC), p. 141. 2018. Project: Object Permanence: Tracking Hidden Objects in Videos (Jan 2019 – Jun 2019) • Designed a memory module for a self-supervised video colorization model, to learn and predict the behavior of hidden objects before, after and while they are concealed in a video, using ResNet-feature associations from grayscale frames. • Achieved qualitative success in saving hidden objects in the memory module; demonstrated using the YouTube-VOS dataset. Learning Intrinsic Image Decomposition using Self-supervised Video Colorization (Sep 2018 – Dec 2018) • Designed a self-supervised model to separate videos into their reflectance (object color and texture) and shading (shadows and depth variations) components. • Achieved an error rate of 23.47% (compared to the 25.9% baseline) for reflectance prediction on the Intrinsic Image in the Wild (IIW) dataset. Demonstrated successful separation of shadows using the Shading Annotations in the Wild (SAW) dataset. Lens-less Imaging (Feb 2016 – Jun 2017) • Designed a low-resource (lens-less) camera which captures sparse modulated data from the scene, without image construction, using a coded aperture mask based on Johnson-Lindenstrauss lemma and dimensionality reduction principles. • Demonstrated theoretical limitations of recovering the image from the captured sparse signal using compressive sensing and Fourier slice theorem.
TA, Introduction to Databases - Spring 2016 TA, Introduction to Databases - Summer 2016 TA, Computer Vision - Fall 2016 TA, Computational Imaging - Spring 2017 TA, Database Systems - Spring 2020
Developed an algorithm that allows users to change the depth of focus of a short video, post-capture, using object tracking, optical flow and depth estimation. Developed a generic framework for generating synthetic video datasets for training and evaluation of supervised algorithms. The framework enables developers to synthesize and use labeled datasets for learning, in the absence of relevant videos.
As a developer for Sybase ASE Database Management System in the 'Database Storage' team, I helped develop and maintain ASE. Project – “In-memory row storage” for Sybase ASE Database - Implemented in-memory row caching and multi-version concurrency control, rolled out as flagship features of the ASE 16.0 SP03 release, which increased throughput and reduced the processing latency of database transactions. - Optimized the memory usage of the ‘bucketpool’ fragment memory manager for non-contiguous memory allocations. - Developed Database Console Commands (DBCC) to monitor the memory usage statistics of in-memory row store cache. - Developed and automated tests for transaction concurrency, triggers, stored procedures and user-defined functions. The tests helped developers establish the stability of features added for the in-memory row store.