Research
I'm interested in computer vision pertinent to 3D objects and leveraging machine learning for efficient representations of the same. I am interested in leveraging learning techniques that employ an explicit understanding of the underlying object-object and object-human relations, topology, differential geometry and the physics governing the underlying scene configurations.
|
|
Normal Assisted Stereo Depth Estimation
Uday Kusupati,
Shuo Cheng,
Rui Chen, and
Hao Su
CVPR, 2020
We study how to enforce the consistency between surface normal and depth at training time to improve the performance. We couple the learning of a multiview normal estimation module and a multi-view depth estimation module which can improve both the prediction of normal and depth.
|
|
Learning 3D Human Pose from Structure and Motion
Rishabh Dabral,
Anurag Mundhada,
Uday Kusupati,
Safeer Afaque,
Abhishek Sharma, and
Arjun Jain
ECCV, 2018
We propose two anatomically inspired loss functions and use them with a weakly-supervised learning framework to jointly learn from large-scale in-the-wild 2D and indoor/synthetic 3D data. We also present a simple temporal network that exploits temporal and structural cues present in predicted pose sequences to temporally harmonize the pose estimations.
|
|
Visual Parsing and Reconstruction with Stochastic Grammars and Recursive Networks
Undergraduate Thesis, IIT Bombay
Advisor: Prof. Siddhartha Chaudhuri
|
|
Affordance based Furniture Generation
Advisor: Prof. Siddhartha Chaudhuri
|
 |
Object Recognition for Robot Cleaner
Samsung Research, Seoul, South Korea
Worked on efficient memory and processing time reduction for deep learning on an automated Robot Cleaner without significant losses in accuracy.
|
 |
Computing Delaunay Complexes using distance-only computations
DATASHAPE, Inria Sophia Antipolis
Advisor: Prof. Jean-Daniel Boissonnaut
Contributed to the GUDHI library with an implementation of computing Delaunay complexes, with a reduced computational complexity of polynomial in dimension from exponential.
|
 |
Robust Cloth Simulation
Physical Simulation, Spring 2019, UT Austin
Advisor: Prof. Etienne Vouga
We implement cloth simulation, handling collisions between the cloth and rigid bodies as well as itself along withfriction along the contact surfaces by combining Bridson et al.'s method with a position based dynamics model of cloth.
|
 |
Topology aware Mesh Reconstruction: Solving for 2D Data
Numerical Optimization for Graphics & AI, Fall 2018, UT Austin
Advisor: Prof. Qixing Huang
We propose a topology aware method for 3D reconstructions from single RGB images using graph convolutions and reinforcement learning. We test the idea on a smaller task of 1D to 2D reconstruction.
|
 |
Role of structured information for answering questions on data visualizations
Deep Learning Seminar, Fall 2018, UT Austin
Advisor: Prof. Philipp Krähenbühl
We exploit the structural information within bar graphs and use a multi head attention on top of them to give state of the art results on the DVQA dataset.
|
 |
Natural Language to Code using Transformers
Natural Language Processing, Fall 2018, UT Austin
Advisor: Prof. Greg Durrett
We analyse the effectiveness of Transformers for the natural language to code translation task and also experiment with other techniques like back translation and cycle consistency.
|
 |
CS 388 - Natural Language Processing - Fall 2019 - UT Austin
CS 378H - Computer Graphics Honors - Spring 2019 - UT Austin
CS 213/293 - Data Structures and Algorithms - Spring 2018 - IIT Bombay
CS 101 - Computer Programming & Utilization - Fall 2017 - IIT Bombay
CS 101 - Computer Programming & Utilization - Spring 2018 - IIT Bombay
|
|