3D Point Cloud Instance Segmetation
This project focused on annotating 3D sensor data, specifically LiDAR-based point clouds, to support machine learning models for autonomous vehicles. The objective was to enable accurate perception of the driving environment for safe navigation and decision-making. The task involved point-level classification and instance segmentation within complex 3D road scenes.I labeled key classes including objects, soft vegetables, hard vegetables, drivable surfaces, and phantoms. I ensured each instance was accurately identified and separated with precise spatial boundaries. I handled challenging scenarios such as occlusions, sparse point distributions, and overlapping objects. I contributed to annotating thousands of point cloud frames, each containing millions of data point and I worked with data collected from diverse driving environments to enhance model robustness. I strictly followed detailed annotation guidelines to maintain consistency.A multi-stage quality assurance process was implemented throughout the project. This included peer reviews, consensus validation, and continuous feedback loops. High inter-annotator agreement and strict quality standards ensured reliable and high-precision labeled data for autonomous vehicle systems.