I am a graduate student at University of California, San Diego pursuing a masters in Intelligent Systems, Robotics and Control.My research interests revolve around deep learning with applications in computer vision and natural language processing. I have experience working with data analysis and visualization.
Decemeber 2017 - August 2018
Explored the possibility of estimating sizes of objects from monocular images and thier possible applications. For this purpose, my team and I worked with several state of the art object detection and instance segmentation systems. This was inspired by the fact that humans, while estimating sizes of objects rely on how visible the object is. To extract features usable for this purpose, we used the Mask RCNN model which would provide an estimate of the mask of the objects (which was used for a naive size comparison model). We also used features extracted from the penultimate layer of the Mask RCNN to train a separate Siamese network to compare objects. To explore the usefulness of depth maps for size estimation, we also designed another model which takes depths from the point of view of the image into consideration.
Through the course of this internship, we designed a baseline for size comparison of objects in a scene and tested viability of several models for the purpose of size estimation of objects.
April 2018 - June 2018
Supervised several data analysis and visualization projects for an undergraduate course. As part of my responsibilities, I helped students with difficulty in writing and debugging code in Python2.7. I received and 100% recommendation from the students that I mentored. An official evaluation will be provided on request.
May 2016 - July 2016
As a Research Intern at the Insight Center for Data Analytics, I worked on interactive art that could react to the presence of people in an image. This project was carried out under the supervision of Prof. Noel O'Connor and Prof. David Monoghan. Using concepts in image processing and computer vision, I designed a system that could detect the presence of people in the field of view of the camera. This system was applied to the "Dotman" electronic art work created by Anne Cleary and Dennis Connolly.
The "Dotman" was an electronic art piece showcased at Lumiere London 2016 and at the Euro Cup 2016.
January 2015 - May 2015
As part of course requirements, our group worked with a start up- Cardiac Design Labs, Bangalore, India in looking for means to zoom in and out of ECG signals with minimum loss of data.
2017 - Present
University of California,San Diego
I am currently pursuing a Masters in Intelligent Systems, Robotics and Control. This stream focuses on a wide variety of concepts ranging from machine learning to stochastic and dynamic processes. My primary focus interms of courses revolves around machine learning and computer vision.
2013 - 2017
National Institute of Technology, Karnataka, India
My Bachelor's degree was in Electronics and Communication. This stream covered topics varying from hardware electronics to communication theory. My primary focus was in the fields of signal and image processing in which I focused my thesis on.
This course covers the major ideas behind deep learning ranging from building vanilla dense neural networks extending to concepts in Reinforcement Learning.
This course covers the mathematical intuitions behind deep learning and explored different application in the field of computer vision.
These courses covered the fundamental ideas behind machine learning extending from generative models in 271A to discriminative models in 271B. They followed a practical approach in class applying many of these machine learning methods to computervision problems like object detection and optical character recognition.
This course served as a study into various computer vision problems and solutions. As a part of this course, we explored various topics such as CLAHE, SPM, SSD, YOLO etc.
This course serves to show how python acts as a valuable tool in the hands of a data scientist. The course was taught by Dr. Jose Unpingco and I was offered an opportunity to work as a TA for this course in the following quarter.
This course served as an introduction to the applications of probability and statics has in data science.
- Trained an unsupervised and a supervised depth estimation model for estimating depth from RGB images and compared their performances.
- Each model was trained on indoor and outdoor images from the KITTI dataset and the NYU dataset.
- The trained models were tested against images in both indoor and outdoor environments to test their viability and usefulness.
- The unsupervised model used was used as a part of a demonstration.
- This project was done as a part of the SVCL internship at UCSD.
- Used the Mask-RCNN built by matterport in Tensorflow as feature extractor to feed into a Simese network to compare sizes of objects in an image.
- Used state-of-the-art depth models (Monodepth and Megadepth) to explore utility of depth estimates in estimating sizes of objects.
- Created sevaral baseline models to compare performance of object size comparison models.
- Built a multiagent model from scratch that plays a game of pong against itself to explore co-operative and competitive behavior when training reinforcement learning models against another reinforcement learning model
- Used Pygame to build game engine for training and visualizing the model
- Used policy gradients to train the model.
- Designed and developed a system for diagnosis of Osteoporosis from X-ray images.
- Used Active Appearance Models to segment given image using the menpo library achieving 90% success in segmentation
- Achieved a success rate of 87% in diagnosis
- Implemented javasript application using Processing(programming framework for digital art), to develop interactive art
- Used background subtraction to detect moving objects in a scene. The scene is further filtered using morphological image processing to remove noise.
- Designed code to cause the dots forming the ”Dotman”(an art installation by Anne Cleary and Dennis Connolly) to track large objects in the scene and revert to prior activity in case of a lack of movement
- Designed a system to analyze and visualize the IMDB dataset that contain information about 4.3 million titles and 8 million artists.
- The system can be used to obtain hard to collect information like popularity of a particular genre or a director throughout the year, most popular movies of a particular year, average rating and popularity of a particular genre etc.
- Also designed a movie recommender system based on collaborative filtering.
- Compared performance of different algorithms on evaluating handwritten expressions.
- Trained convolutional neural networks, multi layer perceptrons, adaboost classifiers and random forests to recognize symbols and numbers and evaluated their comparative performance
C23 ,3950, Mahaila Ave., San Diego, California