Vehicle Point Cloud Parts Annotation

I have a large set of LiDAR point-cloud scenes and I need every vehicle in those scenes precisely annotated, not just with a coarse 3D box but with labeled parts—roof, hood, trunk, wheels, lights, mirrors, and similar components. These annotations will feed an object-detection and classification model, so geometric and semantic accuracy are critical.

You may work in any professional 3D-labeling environment you’re comfortable with—CloudCompare, CVAT-3D, Scalabel, Supervisely, or a comparable tool—so long as the final export can be delivered in JSON or KITTI-style format with class and part tags intact. The raw data are supplied as .las and .pcd files and can be streamed directly from my S3 bucket; I will provide signed URLs once we start.

Deliverables
• For each scene: a single annotation file containing every vehicle and its constituent parts, aligned with the original point cloud
• A brief markdown report describing your labeling process, toolchain, and any edge-case conventions you adopted
• A quick demo video or screenshots showing one fully annotated scene (helps my QA team reproduce your steps)

Acceptance Criteria
• ≥ 95 % of vehicle points correctly assigned to the proper part label when sampled against my ground-truth subset
• No missing vehicles in any frame
• Files open error-free in CVAT-3D and CloudCompare and preserve original coordinate space

If you have a portfolio or sample project showing similar fine-grained 3D vehicle labeling, that will help me gauge fit right away. Looking forward to collaborating with someone who sweats the small details just as much as I do.