103,450 3D keypoints annotated on 8,323 3D models across 16 object categories from the ShapeNet taxonomy. This dataset provides human-annotated keypoint correspondences to facilitate learning semantic 3D structural points for computer vision tasks.
Use Cases
- Train 3D keypoint detectors using the 3D coordinate labels and associated model meshes
- Benchmark semantic consistency in 3D vision tasks across the 16 object categories
- Perform 3D shape registration and alignment using the ground-truth keypoint pairs
Strengths
- 103,450 3D keypoints distributed across 8,323 unique 3D models
- Covers 16 distinct object categories including chairs, airplanes, and cars
- Aggregates multiple human annotations to ensure high semantic accuracy and consistency