Selected graduate research and industry projects across computer vision, generative modeling, and intelligent systems.
Investigates diffusion sampling efficiency for image inpainting by reformulating the reverse diffusion process with Masked-Variance Diffusion (MVD) and Exact Masked-Markov Diffusion (EMMD), cutting inference steps while preserving boundary fidelity.
Single-view sports camera calibration algorithm exploiting projective invariants and cross-ratios to reliably estimate planar homographies under severe occlusion and missing court corner keypoints.
Automated urban traffic signal control using Deep Reinforcement Learning (DQN, DDQN, A2C) in SUMO-RL to minimize emergency vehicle response times while preventing congestion cascade.
Context-aware AI bookmark management system utilizing natural language embeddings and semantic similarity to automate webpage topic extraction, contextual categorization, and conversational search.
Retrieval-augmented conversational agent over Wikipedia articles, combining automated web scraping, TF-IDF/BM25 inverted indexing, and conversational relevance scoring for multi-turn dialogue.
Archived coursework and early capstone projects completed during undergraduate B.Tech studies.
Mobile brand logo detection system targeting the LogoDet-3k benchmark, tackling severe class imbalance via scaled loss formulations and optimized for arm64-v8a mobile edge inference with ArmNN.
Interactive assistive vision wearable for individuals with visual impairment and color vision deficiency, featuring dual-mode power-adaptive object detection and tactile pointer-guided color recognition.
Edge vision-based smart parking system deploying MobileNet-SSD and YOLOv2 on Raspberry Pi 3B+ for real-time slot occupancy tracking, polygon ROI mapping, and cloud state telemetry.
Line-of-sight Free-Space Optical (FSO) vehicle-to-vehicle communication system featuring a custom error-resilient framing protocol, bit stuffing, and stop-and-wait telemetry transmission.