I am a graduate student at University of California, San Diego (UCSD) pursuing a masters degree in Intelligent Systems and Robotics. My research interests lie in the domain of computer vision and deep learning and solving real world challenges using it.
Please find my resume here.
Contact me at i2gupta@eng.ucsd.edu.
Master of Science in Intelligent Systems and Robotics
QPA : 3.875
Courses : Computer Vision, Statisical Learning, Probabalisitc Reasoning and Learning, Neural Networks and Pattern Recognition, Parameter Estimation, Special Topics on Intelligent Vehicles and Systems
Bachelor of Engineering in Electrical and Electronics Engineering, 2014
CGPA : 8.9/10
Applied Scientist Intern (Jun 2017, Sep 2017)
R&D Video Engineer (Jun 2014, Jun 2016)
R&D Intern (Jan 2014, Jun 2014)
Part Time Computer Vision Intern (Mar 2017, Jun 2017)
I have been working on projects related to computer vision, deep learning, video understanding and compression from the past few years. Here are some of the maintained projects which I worked on.
Developed a Deep Neural Network using Soft Attention to predict the latex markup from latex-rendered images of mathematical formulae. Achieved 75% accuracy in exactly reproducing the input image from the generated latex syntax. Introduced a novel technique for regularizing the attention model by using a running average of the attention weights. Created a web interface in bootstrap for better user experience and cross platform performance validation.
The project involved solving one of the major challenges involved in accounting driver vigilance for safe driving systems. I was involved in data collection and deciding basic activities performed by drivers while driving. We analyzed different approaches like frame by frame classification and RNN based attention model for activity classification. This projects helped in proving major cues to generate self alarms for when should the autonomous car switch from autonomous to manual mode.
We explored a multi-task learning frame-work for face analysis. It is the first tensorflow implementation of one of the state of the arts in the domain of face analysis and facial landmark localization. The training was performed on AFLW datset and algorithms for iterative region proposal and landmark based NMS were writtten from scratch.
Developed Computer Vision Algorithms to automate analysis of videos of SPPB(Short Physical Performance Battery) tests for measuring risk of fall in senior adults. Came up with a novel algorithm to ascertain gait speed and gait speed variability of a subject from videos using Mixture of Gaussians. Collaborated with UCSD Health Care for data collection and other experiments.
Developed Computer Vision Algorithms to automate analysis of videos of SPPB(Short Physical Performance Battery) tests for measuring risk of fall in senior adults. Came up with a novel algorithm to ascertain gait speed and gait speed variability of a subject from videos using Mixture of Gaussians. Collaborated with UCSD Health Care for data collection and other experiments.
Currently working on developing a 3D bounding box estimation module as a part of complete computer vision based visual tracker system.The module will be able to predict the projection of the bottom face center of surrounding vehicles with respect to ego vehicle.The final dimension and visual yaw predictions can parametrize the 3D bounding of the surrounding vehicles.The proposed architecture will remove the dependence on LIDAR of current multi object tracking systems.
I am an IOT enthusiast
The idea of this project was to design a device through which your doctor can analyze your body vitals remotely. Although today's ambulances are very well advanced, we came up with a cheaper solution to monitor body vitals even before the ambulance reaches the patient and can be used even in the ambulance so that the doctor will be informed about the patient's condition for better treatment. E-Health Glove detects all your body vitals through various sensors interfaced to it and displays the readings on the a small LCD display on the glove. The glove also logs all the data onto Intel's IoT Dashboard. The hack got great recognition in Intel IOT Events and was presented among top 10 finalists in the event.