David Provencher

• Redesigned and shipped various tire inspection algorithms using semantic segmentation (including data generation, annotation tooling, architecture tuning, training, integration and testing), achieving 30-50% false positive rate reduction across multiple clients’ manufacturing lines • Supervised R&D company-wide for all machine learning and classical computer vision projects • Identified and prioritized improvement opportunities and technical debt in large computer vision codebase to improve ease of use, scalability as well as long-term maintainability • Advised CTO and R&D/Software directors to steer acquisition system design, establish R&D priorities and assess project scope/feasibility to ensure proposed projects are profitable and meet customers’ requirements • Revised processes to better track projects, facilitate knowledge sharing, encourage challenging of ideas and improve developer engagement

• Developed tooling for generating high-quality pseudo-labels from existing tire inspection algorithms and significantly contributed to an in-house, label-efficient semantic segmentation framework based on PyTorch, reducing the need for manual data labelling and accelerating machine learning projects • Migrated legacy algorithms written in multiple languages to Python, improving maintainability and readability • Rescued a failing project by stepping in as team lead, revising planning, redirecting resources and steering R&D efforts to meet client expectations and achieve production-level quality at launch

• Conducted R&D for fisheye stereo sensors exploration (requirements, testing, calibration, sensor fusion, GPU-accelerated stereo vision, etc.) to design and de-risk a next-generation sensor suite for autonomous robots • Created a custom Unity module to accurately simulate RGBD images for calibrated fisheye cameras, reducing real-world data dependency for semantic segmentation and depth perception research • Reviewed literature on various autonomous robotics and machine perception topics (depth estimation, efficient Vision Transformers, etc.) to identify promising research ideas for improving and expanding products

• Supervised a team of software developers working on several concurrent industrial inspection projects, establishing a culture of knowledge sharing and algorithm peer-reviewing • Led computer vision and machine learning R&D projects (object detection, robust template matching, fuzzy text matching, adapting algorithms for different sensor type, etc.) to expand capabilities of systems in the field • Developed, shipped and maintained several inspection algorithms used in tire manufacturing production lines • Reviewed literature on various topics (text quality assessment, unsupervised anomaly detection, etc.) to assess project feasibility and establish realistic roadmaps before undertaking R&D projects

• Conducted multiple R&D projects to solve complex computer vision problems (artifact detection/filtering, template matching, metrology, text inspection, etc.) for an automated tire inspection machine, achieving near-zero false negative rate, while keeping false positive rate and inspection time within constraints • Contributed to numerous collaborative problem-solving sessions, code reviews and design reviews

• Studied blood flow response to electrical activity in the healthy human brain using functional magnetic resonance imaging (fMRI) and electroencephalography (EEG) • Conducted medical imaging studies from start to finish (planning, design, acquisition, data processing, group-level statistics, neuroscientific interpretation) • Optimized data processing pipelines from raw EEG signals and MRI images to co-registered maps of brain activity (signal denoising, deconvolution, clustering, non-rigid registration, etc.) • Published 2 journal articles and presented results in 4 international conferences

• Developed an image acquisition protocol in live mice to capture timecourses of whole-body fluorescent activity using a single camera cycling between calibrated positions • Designed an organ segmentation algorithm based on clustering of fluorescence time-activity curves and custom stereo vision setup

Designed and implemented a low-cost, portable computer vision system for a bioreactor used in arterial tissue growth research