Francisco Lourenço

• • Delivered end‐to‐end ML/CV solutions for aerial imagery, LiDAR, street‐view, and geospatial applications, from data collection and annotation to model training, geometry extraction, measurement, and deployment.

• • Built 3D and sensor‐fusion pipelines for real‐world spatial data, including LiDAR‐camera colorization, georeferencing, orthophoto stitching, and reconstruction workflows.

• • Designed dataset and annotation systems, including taxonomy definition, QA processes, and leading annotation teams for segmentation and estimation tasks.

• • Designed and deployed ML pipelines for 3D shape classification, canonical alignment, and part segmentation used in the Animate Anything tool and Anything World API.

• • Achieved 95.6% F1‐score on the rotated ModelNet40 benchmark through a customized multi‐view GCN architecture optimized for rotation invariance and computational efficiency.

• • Developed a proprietary 3D rotation estimation system (Symmetry‐Aware and Base‐Plane‐Aided) achieving 2.47° median angular error across 220+ characters.

• • Built an optimized 3D shape segmentation pipeline with a GCN model reaching 88.8% mIoU on ShapeNet Part; reduced training and inference time significantly compared to the prior system.

• • Solely designed and developed a multi‐camera RGB‐D 3D scanning system for automated part inspection in powder coating lines.

• • Built and deployed a full pipeline for 3D reconstruction, camera pose estimation, and point cloud fusion from stereo and depth data.