Rohit Choudhary

I am a M.S. research scholar at department of Electrical Engineering, IIT Madras, where I am working under the guidance of Dr. Kaushik Mitra and Dr. Mansi Sharma. I am working on developing deep learning methods for low-light image enhancement and scene depth estimation. Prior to joining the M.S program at IIT Madras, I completed my undergraduate studies in Electronics and Communication Engineering from the National Institute of Technology Kurukshetra, India.

Email / CV / Google Scholar / LinkedIn / Github

News

(June 17, 2024) See you at Arch Building Exhibit Hall, Board 56 (Poster Presentation) at 3DMV, CVPR 2024.
(October 9, 2023) I will be presenting ELEGAN (Poster Presentation) at ICIP 2023.
(June 22, 2023) Paper on unsupervised low light image enhancement accepted in IEEE ICIP 2023.
(April 27, 2023) MEStereo-Du2CNN paper accepted for publication in The Visual Computer Journal.
(September, 2021) Paper on stereo depth estimation accepted in IEEE VCIP 2021.

Research

I'm interested in computer vision, machine learning and image processing. Most of my research has been focused on advancing scene depth estimation and low light image enhancement.

	Reflective Teacher: Semi-Supervised Multimodal 3D Object Detection in Bird’s-Eye-View via Uncertainty Measure Saheli Hazra, Sudip Das, Rohit Choudhary , Arindam Das, Ganesh Sistu, Ciar´an Eising and Ujjwal Bhattacharya. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2025. Arxiv The framework uses Reflective Teacher to enhance pseudo-labeling in semi-supervised 3D object detection while preventing catastrophic forgetting. Additionally, Geometry-Aware BEV Fusion aligns LiDAR and camera features for precise geometric and semantic mapping.
	GANESH: Generalizable NeRF for Lensless Imaging Rakesh Raj Madavan, Akshat Kaimal, Badhrinarayanan K V, Vinayak Gupta, Rohit Choudhary, Chandrakala Shanmuganathan, Kaushik Mitra. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2025. Arxiv/ Project page Framework for simultaneous 3D refinement and novel view synthesis from multi-view lensless images, supporting on-the-fly inference and scene-specific tuning.
	2T-UNET: A Two-Tower UNet with Depth Clues for Robust Stereo Depth Estimation Rohit Choudhary, Mansi Sharma, Rithvik Anil, Second Workshop for Learning 3D with Multi-View Supervision (3DMV), CVPR 2024. Paper Link/ ArXiv Robust stereo depth estimation perform incredibly well on complex natural scenes.
	MEStereo-Du2CNN: a dual-channel CNN for learning robust depth estimates from multi-exposure stereo images for HDR 3D applications Rohit Choudhary, Mansi Sharma, T.V. Uma, Rithvik Anil, The Visual Computer Journal, 2023 Paper link / Github page / arXiv / Supplementary file Exploring the utilization of a High Dynamic Range (HDR) reconstruction process to forecast depth, thereby opening up the potential of achieving 3D HDR through a unified pipeline.
	ELEGAN: An efficient low light enhancement for unpaired supervision Rohit Choudhary, T Harshith Reddy, Mansi Sharma IEEE International Conference on Image Processing (ICIP), 2023 Paper link / Presentation video ELEGAN, a lightweight attention-guided generative adversarial network for fast low-light image enhancement in a fully unsupervised manner.
	SDE-DualENet: A Novel Dual Efficient Convolutional Neural Network for Robust Stereo Depth Estimation Rithvik Anil, Mansi Sharma, Rohit Choudhary IEEE Visual Communications and Image Processing (VCIP), 2023 Paper link Eliminating cost-volume construction in stereo disparity estimation.
	A High Resolution Multi-exposure Stereoscopic Image & Video Database of Natural Scenes Rohit Choudhary, Mansi Sharma, Aditya Wadaskar arXiv, 2022 Webpge / ArXiv A diversified stereoscopic multi-exposure dataset captured within the campus of IIT Madras.

Website credits to Jon Barron.