I am a Research Scientist at Aware Inc. working on Machine Learning and Computer Vision projects. I have a passion for solving data-driven problems and learning new technologies in this rapidly growing field. Specifically, I have played around with spatial and temporal data and have done projects on Activity Recognition from Video, Image Segmentation and Time Series Analysis. In my free time, I love tinkering some code or reading technology blogs.
GPA: 3.66/4.0
GPA: 7.72/10.0
Implemented VINS-Mono and ORB SLAM2 algorithms, separately, for an Arial robot blimp for mapping a construction site. Used NVIDIA Jetson TX1 development board and monocular camera with an IMU for localizing and mapping the surrounding which used Point Cloud Library (PCL) and Robot Operating System (ROS). Co-ordinated with Hardware and Context Awareness teams to integrate their work to develop a fully functional working prototype. The reports can be found here.
Implemented activity recognition from video using Convolutional Neural Nets and LSTMs. The project aimed at implementing the paper 'Long Recurrent Convolutional Network' for benchmarking a hardware accelerator. Used Transfer Learning on VGG-16 network and obtained an accuracy of 70% for 7 different activities on UCF-101 dataset.
This project was aimed at recommending solutions to users investing in stock market by utilizing fundamental and technical analyses. The best companies to invest in was chosen using fundamental analysis and the user's portfolio was optimized to distribute the investment in different stocks to maximize returns and reduce risks. Technical analysis was implemented separately using Neural Networks and Reinforcement Learning to suggest actions (buy, sell or hold) to the user to maximize his/her portfolio.
This project was focused on estimating the respiratory rate of an individual using accelerometer data, heart rate and body temperature. 3 different models were implemented separately for estimation: Ridge Regression, Neural Networks and Hidden Markov Model.
In this project, I implemented a paper titled 'Deep Visual Attention Prediction' which focuses on predicting human-eye fixation on view-free scenes incorporating multi-saliency predictions. I was able to successfully port the original code in Caffe to Keras with 64% accuracy.
Used Markov Random Field basd approach called Graph-Cut to segment the chambers of a Foraminifera(marine species) from its edge probability map. Segmented the chambers and the aperture in the Foraminifera images to identify different types of species.
Obtained the 3D model of an object from its 2D image. Used vanishing points to obtain homography matrices and the projection matrix. Applied affine transformations to the image using the obtained matrices. Finally, the 3-sides of the object were cropped from the images and fed to a simple wrl file.
Developed an approach to teach a task to a robotic arm which could be replicated as and when required. Interfaced industrial robotic arm Scorbot-ER VII with NI-myRio development board with a user-friendly LabVIEW interface. Got shortlisted in the top 20 teams for a National Level Contest, NIYANTRA, organized by National Instruments, India.