Jacob Krucinski

Jacob Krucinski

Northeastern University — Master's Thesis (Advisors: Alireza Ramezani, Rajagopal Venkatesaramani) · 2026

A real-time image preprocessing framework for visual odometry on the Aerobat flapping-wing MAV consisting of two modular ROS 1 components: a RobustL1-based electronic image stabilizer and an IMU/CNN-guided non-uniform Wiener deblurring pipeline. They both address camera jitter and motion blur from wing-stroke dynamics. The stabilizer runs at 100+ FPS and the deblurring pipeline at 21.4 FPS end-to-end at 480p, with a +1.02 dB PSNR improvement on Aerobat sequences and a 72.8% RMSE reduction in ORB-SLAM3 localization error on the V2_03_difficult EuRoC sequence.

Jacob Krucinski

Lund University — EDAN70 Course Project (Advisor: Sonja Aits) · 2023

Explored BioGPT and SciBERT for relation extraction on annotated biomedical corpora (BC5CDR, ChemProt, DrugProt). Fine-tuned SciBERT on a combined ChemProt+DrugProt corpus, improving precision, recall, and F1 by 5%.

Jacob Krucinski, Adam Bienkowski, Krishna Pattipati

IEEE Aerospace Conference · 2021

Computationally efficient CNNs for detecting and localizing streaking targets from an optical sensor's focal plane array. The ML models were ~340× faster for detection and ~360× faster for localization on 256×256 images vs SOTA probabilistic methods. Detection ROC AUC: 0.94 (ML) vs 0.85 (SOTA); localization MSE: 0.035 (ML) vs 0.243 (SOTA).