Mithun Parab

Nice to meet you! I’m Mithun, and my work revolves around computer vision, image and video enhancement, and AI. I’m particularly interested in making deep learning models more effective for real-world applications. Let’s connect if you’d like to discuss opportunities or collaborations!

Email / CV / Blog / Scholar / Github

Research Experience

Sejong University, Seoul, South Korea (Research Intern) Dr. Y.G. Kim & Dr. Palash Ingle. Advance video anomaly detection technologies by developing robust models to identify and analyze anomalies in video data.	04/24 — 10/24
Indian Institute of Information Technology, Sri City, India (Research Intern) Dr. Pavan Kumar B.N. , Department of CSE Improved 3D SLAM precision and effectiveness using uncalibrated image-based algorithms.	09/23 — 03/24

Selected Publications

	A Comprehensive Study on LLM Agent Challenges Palash Ingle ^* , Mithun Parab^, Pranay Lendave, Pavan Kumar B N AAAI 2024 Spring Symposia on User-Aligned Assessment of Adaptive AI Systems, Stanford University, Stanford, CA, USA.* Paper \| YouTube
	MT3DNet: Multi-Task learning Network for 3D Surgical Scene Reconstruction Mithun Parab^, Pranay Lendave ^ , Jiyoung Kim, Palash Ingle, Thi Quynh Dan Nguyen DOI:10.48550/arXiv.2412.03928
	Innovative Method for Camouflaged Wildlife Segmentation in Agricultural Practices Mithun Parab, Palash Ingle IEEE Xplore Digital Library, International Conference on Advancement in Computation & Computer Technologies. DOI: 10.1109/InCACCT61598.2024.10551184
	Image Enhancement and Exposure Correction Using Convolutional Neural Network Mithun Parab, Amisha Bhanushali, Palash Ingle, Pavan Kumar B N SN Computer Science, Volume 4, Number 2, 2023. DOI: 10.1007/s42979-022-01608-w

Research Projects

	Tetris-Inspired Video Synopsis A Monte Carlo Tree Search algorithm with a shared-backbone neural network for policy and value optimization, efficiently packing 3D tubes to reduce the time dimension in video synopsis.
	DINO-v2-based Method for Video Anomaly Detection A DINO-v2-based approach for video anomaly detection, using feature embeddings and prototype learning to group similar patterns while pushing apart different ones. Anomalies are detected based on deviations from learned prototypes.
	Monte Carlo Tree Search with Neural Network for 3D Bin Packing A Monte Carlo Tree Search algorithm integrated with policy and value networks for solving 3D bin packing problems, optimizing placement in dynamic packing environments.
	3D Novel View Synthesis from Un-calibrated Images A system for synthesizing new 3D views from un-calibrated images, using a NeRF model optimized for Structure from Motion challenges via distinct MLP modules.
	3D Video Synopsis with Multi-task Learning A condensed video synopsis algorithm and Multi-Task Learning network for abnormal activity segmentation and depth mapping, facilitating 3D video summary reconstruction.

Yep it's another Jon Barron's website .
Last updated on 13 February 2025