Prateksha Udhayanan

I am a second-year PhD student in the Department of Computer Science at the University of Maryland, advised by Prof. Abhinav Shrivastava. Before joining UMD, I worked as a Research Associate at Adobe Research, where I worked on projects spanning retrieval, editing, and generation for images, videos, and graphic designs.

I received my Bachelor’s and Master’s in Computer Science Engineering with a specialization in Artificial Intelligence and Machine Learning from IIIT Bangalore. During my undergraduate studies, I spent three months as a research intern at Adobe Research, working with Balaji Vasan Srinivasan and Stefano Petrangeli on document-to-video transformation. I also worked on anomaly detection in procedural videos, with Varghese Alex and Vinay Sudhakaran during my summer internship at Siemens. I had the opportunity to explore various projects during my undergraduate studies with Prof. Dinesh Babu Jayagopi, including analyzing speaker behaviour in videos for virtual meeting scenarios and hand-gesture estimation for Indian Sign Language synthesis on human avatars.

CV  /  Google Scholar  /  LinkedIn

Research

My research interests are in computer vision, with a primary focus on generative models. I am currently working on image generation models, and have previously worked on adding controllability to the image generation process and evaluating motion quality in video generation.

We propose Design-o-meter, a data-driven methodology to score and refine graphic designs.

We propose Contextualized Prompt Learning (CoPL), a prompt learning method that adapts prompt weights dynamically and aligns the prompt vectors with local image features.

We present a training-free framework for iterative, multi-granular image editing, along with IMIE-Bench, a new benchmark dataset for evaluating the proposed task.

We present a novel deep-learning driven system - Recipe2Video that automatically converts a recipe document into a multimodal illustrative video.

We propose a gradient-attention-based learning objective for composed image retrieval, that explicitly forces the model to focus on the local regions of interest being modified in each retrieval step.

We propose a multimodal unsupervised domain adaptation method to predict the persuasiveness and expertise of the speaker in a video

We propose an AST-based method to compute similarity score between two source codes, focusing on their structure rather than their functional outputs.

Source Code from Jon Barron