Hire Collins R. - Computer Vision Engineer

Leading a team of computer vision engineers focused on safety, quality, and productivity across Georgia-Pacific’s manufacturing operations. Our work spans three priorities: building vision systems that keep our sites safe by detecting hazards and unsafe conditions in real time, developing automated inspection systems that catch product defects at scale, and designing solutions that streamline operations to boost throughput and reduce downtime. Alongside the technical work, I set our team’s direction, mentor engineers, and align our computer vision strategy with Georgia-Pacific’s operational goals.

I’m a Computer Vision Engineer at Georgia-Pacific, where I help identify industrial problems that can be solved with computer vision and design practical, end-to-end solutions around them. I work with cross-functional teams and site stakeholders to understand operational challenges, evaluate where image processing and model-based systems can create value, and help drive those ideas from concept to production. My work spans site safety, quality inspection, defect detection, automation, and manufacturing applications across paper, lumber, and other product areas. Along the way, I contribute to model development, MLOps pipelines, annotation workflows, and deployment support to help deliver scalable, production-ready systems. I’m passionate about technical innovation, leading the adoption of AI-driven solutions, and building tools that make work safer, easier, and more efficient.

During my time as a Graduate Researcher at the Multispectral Imaging Lab, I have leveraged cutting-edge deep learning techniques to advance defect detection and innovate data collection methodologies in multispectral imaging. Leading the strategic acquisition of a comprehensive Short-Wave Infrared dataset, I successfully amassed over 1,600 images. As the sole researcher on a $1.25 million DoD research contract, I have made significant contributions to the field. One of my notable achievements includes designing a Near Infrared Dorsal Hand Vein Authentication System, which utilized a unique knuckle alignment technique and ResNet-18, achieving a remarkable verification accuracy of 98.78% on the JLU dataset and 100% on the DHV dataset. Additionally, I developed an automated alignment method that processed 3,680 images in just 15 seconds, a significant improvement over the manual alignment time of approximately 5 hours and 38 minutes. My work also involved conducting an ablation study on global versus local hand vein blurring, demonstrating the system's resilience to variations in image quality. In another project, I spearheaded the development of a data synthesizer for crack detection using StyleGANv3 and BBDM. Through hyperparameter tuning, I achieved a 10% improvement in the DeepLabv3+ model's accuracy. I also applied blur augmentation techniques, which led to an additional 3% increase in model robustness and accuracy. By implementing ensemble modeling techniques on augmented datasets, I set a new benchmark in the field with a state-of-the-art Mean Intersection over Union accuracy of 65.62%. My pioneering research in diffusion models for synthetic image generation has significantly advanced structural safety applications in industrial settings. My experience at the Multispectral Imaging Lab has been incredibly rewarding, allowing me to contribute to impactful research and develop innovative solutions in the field of multispectral imaging.