LiDAR Semantic Segmentation Annotation
Budget: $15 – $25 CAD
I need my raw LiDAR point-cloud maps turned into finely segmented training data. Every single point that belongs to a vehicle, building, or pedestrian along with several other classes must be tagged according to a predefined schema that I’ll provide at kickoff. The work calls for true fine-grained, per-point labeling—no loose bounding volumes—so that the dataset is immediately usable in a deep-learning pipeline.
I will supply the point-cloud files, the detailed class definitions, and a fully labeled reference scene. You may use any professional tool you like (CloudCompare, SemanticKITTI editor, Labelbox, etc.) as long as the finished files remain in the original format and coordinate system. If needed we can also provide our own custom desktop app to assist with labeling.
The point clouds have already been classified with an automated model, so the majority of the work is just correcting labels and extending them to more specific classes.
Deliverables
• Complete, per-point annotations for every supplied map
• A brief log of any edge cases or assumptions made
• Quick quality-control summary showing class accuracy and confirmation that no points are left unlabeled
Acceptance criteria
• 100 % of points carry one of the required classes from my schema
• Output loads flawlessly in standard .las viewers with all existing metadata intact
Once the sample scene passes review, you’re clear to move through the rest of the dataset. I’m ready to get started as soon as you are.
I will supply the point-cloud files, the detailed class definitions, and a fully labeled reference scene. You may use any professional tool you like (CloudCompare, SemanticKITTI editor, Labelbox, etc.) as long as the finished files remain in the original format and coordinate system. If needed we can also provide our own custom desktop app to assist with labeling.
The point clouds have already been classified with an automated model, so the majority of the work is just correcting labels and extending them to more specific classes.
Deliverables
• Complete, per-point annotations for every supplied map
• A brief log of any edge cases or assumptions made
• Quick quality-control summary showing class accuracy and confirmation that no points are left unlabeled
Acceptance criteria
• 100 % of points carry one of the required classes from my schema
• Output loads flawlessly in standard .las viewers with all existing metadata intact
Once the sample scene passes review, you’re clear to move through the rest of the dataset. I’m ready to get started as soon as you are.