Francisco Lourenço

Build end-to-end 3D perception pipelines for mapping and navigation clients. Developed LiDAR–camera fusion and colorization with filtering for stable RGB point clouds. Implemented orthostrip generation and georeferencing from aerial imagery for large-scale surveys. Optimized trajectory refinement, pose estimation, and QA tooling for long-range sensor logs.

Designed and shipped ML pipelines for 3D shape classification, canonical alignment, and part segmentation powering the Animate Anything tool and Anything World API. Reached 95.6% F1-score on the rotated ModelNet40 benchmark with a custom multi-view GCN focused on rotation invariance and efficient inference. Created a proprietary 3D rotation estimation system achieving 2.47 deg median angular error across 220+ diverse characters. Rebuilt the 3D part-segmentation stack with a GCN-based model that matched accuracy while cutting training and inference cost in production.

Sole 3D CV engineer designing a multi-camera RGB-D scanning rig for real-time part inspection in powder-coating lines. Implemented calibration, pose estimation, and point-cloud fusion software for precise per-part 3D reconstruction in production.