Divya Kothandaraman

I am a senior researcher at Dolby Laboratories. I earned my PhD in Computer Science from the University of Maryland, College Park. My dissertation, "Learning from Less Data: Perception and Synthesis," explores data-efficient AI. Prior to this, I was an undergraduate at the Indian Institute of Technology Madras, where I obtained a bachelors degree in Electrical Engineering, and masters degree in Data Sciences.

My research interests encompass the intersection of generative AI, computer vision, and multi-modal learning. My recent works range from developing novel methods for generative AI tasks in controllable image and video generation such as personalization, novel-view synthesis, and prompt mixing to developing deep learning based solutions for computer vision tasks such as domain adaptation and video action recognition.

Email / CV / Google Scholar / Twitter / Github

Latest News

(Jan 2025) Joined Dolby Laboratories as a Senior Researcher!
(Nov 2024) ImPoster has been accepted to COLING 2025!
(Nov 2024) Defended my PhD!
(Sep 2024) Gave a talk on novel view synthesis at the ECCV 2024 Wild3D Workshop!
(May 2024) New paper on prompt mixing using the Black Scholes model is on ArXiv.
(Mar 2024) Gave a talk at UCL! Slides here.
(Nov 2023) HawkI is on ArXiv!
(Sep 2023) Aerial Diffusion has been accepted to Siggraph Asia 2023.
(May 2023) Interning at Google DeepMind.
(Jan 2023) Differentiable FAR has been accepted to ICRA 2023.
(Oct 2022) Two papers have been accepted to WACV 2023.
(July 2022) FAR: Fourier Aerial Video Recognition has been accepted to ECCV 2022.

Research Highlights

	Prompt Mixing in Diffusion Models using the Black Scholes Algorithm Divya Kothandaraman, Ming Lin, Dinesh Manocha ArXiv arXiv / GitHub An approach for prompt mixing using novel perspectives from the Black Scholes model in economics and finance.
	HawkI: Homography and Mutual Information Guidance for 3D-free Single Image to Aerial View Divya Kothandaraman, Tianyi Zhou, Ming Lin, Dinesh Manocha ArXiv arXiv / GitHub Mutual information and inverse perspective mapping guidance for text-controlled aerial view synthesis from a single input image using diffusion models.
	ImPoster: Text and Frequency Guidance for Subject Driven Action Personalization using Diffusion Models Divya Kothandaraman, Kuldeep Kulkarni, Sumit Shekhar, Balaji Vasan Srinivasan, Dinesh Manocha COLING 2025 arXiv / GitHub An approach for subject and action personalization using prompting techniques and concepts from image and signal processing.
	Text Prompting for Multi-Concept Video Customization by Autoregressive Generation Divya Kothandaraman, Kihyuk Sohn, Ruben Villegas, Paul Voigtlaender, Dinesh Manocha, Mohammad Babaeizadeh AI4CC Workshop at CVPR 2024 arXiv Sequential and controlled autoregressive generation of the desired custom concepts for multi-concept customized video generation with transfoermer models.
	Aerial Diffusion: Text Guided Ground-to-Aerial View Translation from a Single Image using Diffusion Models Divya Kothandaraman, Tianyi Zhou, Ming Lin, Dinesh Manocha Siggraph Asia 2023 (Conference Proceedings, Technical Communications) arXiv / GitHub A text-guided image to image diffusion model to generate aerial views from a single ground-view image.
	Differentiable Frequency-based Disentanglement for Aerial Video Action Recognition Divya Kothandaraman, Ming Lin, Dinesh Manocha ICRA 2023 arXiv / GitHub A differentiable feature disentanglement method to learn "static salient" and "dynamic salient" regions for aerial video action recognition.
	SALAD: Source-free Active Label Agnostic Domain Adaptation Divya Kothandaraman, Sumit Shekhar, Abhilasha Sancheti, Manoj Ghuhan, Tripti Shukla, Dinesh Manocha WACV 2023 arXiv / GitHub A generic source-free active domain adaptation method that can handle shifts in output label space.
	FAR: Fourier Aerial Video Recognition Divya Kothandaraman, Tianrui Guan, Xijun Wang, Sean Hu, Ming Lin, Dinesh Manocha ECCV 2022 Project Page / arXiv / GitHub An efficient aerial video action recognition method, with novel frequency domain techniques, vis-a-vis, Fourier object disentanglement and Fourier attention.
	GANav: Group-wise Attention Network for Classifying Navigable Regions in Unstructured Outdoor Environments Tianrui Guan Divya Kothandaraman, Rohan Chandra Dinesh Manocha IROS 2022 and RSS 2022 Project Page / arXiv / bibtex An attention-based segmentation method for identifying safe and navigable regions in off-road terrains.
	SS-SFDA : Self-Supervised Source-Free Domain Adaptation for Road Segmentation in Hazardous Environments Divya Kothandaraman, Rohan Chandra Dinesh Manocha ICCV Workshops 2021 Project Page / arXiv / YouTube / GitHub / bibtex A self-supervised learning approach for source free unsupervised road segmentation in adverse weather environments and low light conditions.
	BoMuDA: Boundless Multi-Source Domain Adaptive Segmentation in Unconstrained Environments Divya Kothandaraman, Rohan Chandra Dinesh Manocha ICCV Workshops 2021 Project Page / arXiv / YouTube / GitHub / bibtex A multi-source boundless unsupervised domain adaptation algorithm for semantic segmentation in unstructured environments.
	Domain Adaptive Knowledge Distillation for Driving Scene Semantic Segmentation Divya Kothandaraman, Athira Nambiar Anurag Mittal WACV Workshops 2021 Paper / YouTube / GitHub / bibtex An approach for domain adaptive semantic segmentation in models with limited memory.
	Deep Atrous Guided Filter for Image Restoration in Under Display Cameras Varun Sundar , Sumanth Hegde, Divya Kothandaraman* , Kaushik Mitra ECCV Workshops, 2020 ArXiv / YouTube / Project Page / bibtex Guided Filters when incorporated in a deep network can efficiently recover severely degraded, mega-pixel resolution images.

Website template borrowed from Jon Barron.