Paritosh Mittal

Paritosh Mittal

MSCV Student at CMU RI

Carnegie Mellon University

Biography

I am an MS student at the Robotics Institute of Carnegie Mellon University, advised by Professor Shubham Tulsiani. I have a natural affinity towards Computer Vision, Machine Learning and great confidence in their ability to uplift human society. Recently, I am working towards developing generic autoregressive 3D shape priors and exploring their utility for shape completion and reconstruction tasks.

Before CMU, I worked as a Sr. Machine Learning Engineer with the Advanced Technology Labs in Samsung Research, India. I worked closely with Dr. Shankar Venkatesan towards developing AI systems that can remove obstructions from real-world images. For my undergraduate thesis, I was fortunate to be advised by Professor Arijit Sur on the problem of Image Memorability Prediction.

Download my resumé.

Interests
  • Computer Vision
  • Deep Learning
  • Photography
Education
  • M.S. Computer Vision, 2022

    Carnegie Mellon University

  • B.Tech. Computer Science & Technology, 2018

    Indian Institute of Technology, Guwahati

Skills

pytorch
Pytorch

90%

python
Python

90%

opencv
OpenCV

90%

tensorflow
Tensorflow

75%

scikit-learn
Scikit-Learn

75%

pandas
Pandas

75%

Experience

 
 
 
 
 
Verisk Analytics
Research Collaborator (MSCV Capstone)
Verisk Analytics
Sep 2021 – Present Pittsburgh, PA
Working on generative 3D models in collaboration with CMU and Verisk
 
 
 
 
 
NVIDIA
Computer Vision Software Intern
NVIDIA
May 2021 – Aug 2021 Santa Clara, CA
Worked with the Stereo Perception team of Nvidia’s Autonomous Driving Group
 
 
 
 
 
Samsung Research
Sr. Machine Learning Engineer
Samsung Research
Mar 2020 – Dec 2020 Bangalore, IN
Proposed and Developed a stroke based input modality with Input Method Intelligence team
 
 
 
 
 
Samsung Research
Machine Learning Engineer
Samsung Research
Jul 2018 – Mar 2020 Bangalore, IN
Innovated [Patented] a stereo-image based pipeline for Obstruction Removal

News

  • [11/2021] Invited to speak in Verisk AI Frontline (VAI-FI) Seminar Series. Slides
  • [08/2021] Joined Dr. Shubham Tulsiani’s group in CMU RI
  • [08/2021] Undertook TAship in Computer Vision (16-720 B) with Dr. Kris Kitani
  • [05/2021] Interned with Nvidia’s Autonomous Driving team
  • [02/2021] Started graduate studeies at CMU RI
  • [07/2018] Joined the Advanced Technology Labs in Samsung Research, India
  • [07/2014] Started undergraduate studies at IIT-Guwahati

Academic Projects

.js-id-Deep-Learning
Non-sequential Autoregressive Shape Priors for 3D Completion, Reconstruction and Generation

Non-sequential Autoregressive Shape Priors for 3D Completion, Reconstruction and Generation

Capstone project on exploring autoregressive shape priors for 3D objects and developing a unified framework for shape completion, single-view reconstruction and language guided generation

Multi-Modal Multi-Hop Source Retrieval using Graph Convolutions

Multi-Modal Multi-Hop Source Retrieval using Graph Convolutions

A graph convolution based approach to select multiple relevant sources (text or images) of information for multi-hop question answering.

Inverting 3D Deep Learning Architectures

Inverting 3D Deep Learning Architectures

Project on inverting models of 3D object recognition and classification in order to analyze interpretability

3D reconstruction using Stereo Correspondence

3D reconstruction using Stereo Correspondence

Course Assignment on Eight-Point Algorithm, finding epipolar correspondence and bundle adjustment for 3D reconstruction from noisy stereo correspondence

Homography and Panaroma

Homography and Panaroma

Course assignment to compute homography matrix, develop AR application and create panaromas using multiple images

Scene Classification using Visual Words

Scene Classification using Visual Words

Course assignment to use conventional filter responses (Gaussian, Laplacian of Gaussian etc.) to represent images and develop spatial pyramid based approach to classify scenes.

Video Recommender System

Flask based WebApp to recommend and view videos. Uses MySQL, MongoDB and Neo4j for tracking, storing and recommending videos

Hunger Games

A realtime IoT based outdoor shooting game

Java Multi-Threading

Implemented distributed merge-sort and android based client server application using the ForkJoin principle

Logic Programming

Uses logic programming language SWI-Prolog to develop database retireval and maze solver

Publications

More publications »

(2021). Method and electronic device for obtaining reconstructed image. US Patent.

(2020). A method for enhanced recognition of continuous inputs from multiple modalities. US Patent.

(2020). Deep Fence Estimation using Stereo Guidance and Adversarial Learning. In Arxiv.

PDF Cite

(2020). Photo-realistic emoticon generation using multi-modal input. In ACM IUI.

PDF Cite DOI

(2018). Misbehavior Detection in C-ITS Using Deep Learning Approach. In ISDA.

PDF Cite DOI

(2018). Image Memorability: The Role of Depth and Motion. In ICIP.

PDF Cite DOI