Co-Founder & Technical Lead, Cruxnd Haze, Inc. — 3D reconstruction and segmentation pipeline (SMPL-based avatar reconstruction)
Built and improved an AI-based 3D reconstruction pipeline for human avatars from image/video inputs, including frame extraction and segmentation steps. Iterated segmentation quality by switching from MODNet to SCHP/CIHP human parsing systems to better extract body/part regions for downstream reconstruction. Supported avatar persistence to maintain consistent 3D identities across sessions. • Involved human parsing and segmentation workflows • Fed segmentation outputs into point cloud generation and mesh creation • Evaluated and deployed reconstruction model variants • Ensured consistency for avatar persistence