cv

Computer Vision and Graphics Intern

NVIDIA Maxine AI

• Developed end-to-end deep learning models for controllable portrait video animation as part of NVIDIA Maxine.

Software Engineering Intern

NVIDIA Maxine AI

• Developed end-to-end visual and audio-visual deep learning models for high-fidelity facial animation of avatars as part of Maxine ARSDK.

Research Software Engineer

IBM Research India

• Developed an end-to-end soil moisture extraction system from satellite images by incorporating image interpolation techniques as a part of IBM Geospatial Analytics suite.
• Developed explainable deep learning models in the domains of image classification and visual search as a part of retail and operations effort.

Research intern

Indian Statistical Institute, Kolkata

• Developed a key recovery scheme based on the properties of Slid Pairs for stream cipher Salsa20.

Ph.D.

University of Southern California

Ming Hsieh Department of Electrical and Computer Engineering

• Grounding Natural Language
• Advanced Computer Vision
• Affective Computing
• Mathematics of High Dimensional Data

M.Tech

Indian Institute of Technology, Bombay

Electrical Engineering

• Matrix Computations
• Machine Learning
• High-performance Computing
• Optimal Control

Bachelor of Engineering

Jadavpur University

Electronics and Telecommunication Engineering

Automated analysis of advertisement videos

Introduced large-scale advertisement benchmark dataset (MM-AU) and multimodal models for semantic video understanding tasks.

• Keywords: multimodal learning, media understanding, advertisements
• Work published in ACM MM 2023 proceedings.

Context driven human affect perception

Developed multimodal context fusion module for emotion recognition in context-driven scenarios.

• Keywords: multimodal fusion , emotion recognition
• Work published in ICASSP 2023 proceedings.

Multimodal federated learning

Co-developed multimodal benchmark tasks and baseline models for federated learning applications

• Keywords: multimodal fusion, federated learning
• Work done in collaboration with Amazon Alexa AI
• Work published in KDD 2023 proceedings.

Visual scene understanding

Proposed a large-scale weakly labeled movie-centered scene dataset (MovieCLIP) and knowledge transfer to scene and genre classification tasks across diverse domains.

• Keywords: visual scene recognition, automatic labeling
• Work done in collaboration with Google Research
• Work published in WACV 2023 proceedings.

Automated analysis of facial paralysis patients

Developed a facial-landmark based video pipeline involving novel asymmetry measures for predicting standardized scores in a mixed effects modeling setup.

• Keywords: facial landmarks, automated analysis, linear mixed effects model
• Work done in collaboration with Keck School of Medicine and Pacific Neuroscience Institute.
• Work published in Facial Plastic Surgery & Aesthetic Medicine proceedings.

Understanding emotion perception in art work

Developed multimodal transformer based architectures with configurable image features for evoked emotion recognition in art images.

• Keywords: multimodal transformer, emotion recognition, art images
• Work published in ICCV CLVL workshop 2021.

Computational analysis of gender portrayal in media

Analyzed emerging trends in TV shows and advertisements across the dimensions of age, perceived skintone and gender

• Keywords: representation in media, media understanding
• Work done in collaboration with Geena Davis Institute on Gender in Media and Google Research.