research
The Point Cloud Learned From Surfaces the Sensor Never Saw
GhostPoint trains a predictor to hallucinate latent neighborhood features beyond measured LiDAR returns, improving sparse-scan 3D detection.
Summary
GhostPoint trains a predictor to hallucinate latent neighborhood features beyond measured LiDAR returns, improving sparse-scan 3D detection.
Most self-supervised LiDAR objectives supervise only visible returns, even though object detection must reason through occlusion and missing structure. GhostPoint dilates discovered instances into local neighborhoods and trains observed voxels against teacher-encoder targets while unobserved voxels follow teacher-predictor hallucinations. Reported nuScenes and Waymo tests improved downstream detection, especially with sparse scans and limited labels; the hallucinations are learned representations, not reconstructed ground truth.
Why it matters
GhostPoint trains a predictor to hallucinate latent neighborhood features beyond measured LiDAR returns, improving sparse-scan 3D detection.
Limits and context
- Most self-supervised LiDAR objectives supervise only visible returns, even though object detection must reason through occlusion and missing structure.
- Reported nuScenes and Waymo tests improved downstream detection, especially with sparse scans and limited labels; the hallucinations are learned representations, not reconstructed ground truth.
Key claims
GhostPoint trains a predictor to hallucinate latent neighborhood features beyond measured LiDAR returns, improving sparse-scan 3D detection.
Qualification: Most self-supervised LiDAR objectives supervise only visible returns, even though object detection must reason through occlusion and missing structure.
Evidence: source-2026-08-17-009
Sources
- arXiv preprint 2608.14428arXiv · primary research
Corrections
No corrections have been recorded for this story.