Octree-based Learned Point Cloud Geometry Compression: A Lossy Perspective

Dimensions overall score 5.0

GitHub Code Pulse

No public code linked for this paper yet.

• The proposed leaf nodes lossy compression method significantly outperforms previous octree-based methods on object point clouds.result 90%
• The proposed rate control method for LiDAR point clouds achieves about 1% bit error without finetuning.result 90%
• Traditional lossy compression using lossless octree representation with quantization step adjustment causes severe distortions due to massive missing points.limitation 85%
• For object point clouds, a new leaf nodes lossy compression method is proposed that uses bit-wise coding and binary prediction on leaf nodes.method 90%
• For LiDAR point clouds, a simple but effective rate control method is proposed as part of variable rate approaches.method 85%
• The proposed rate control method demonstrates generalization across different models as shown by well-controlled bit errors.result 80%
• The leaf nodes compression framework uses a backbone context model with different K settings for non-leaf coder, leaf bit-wise coder, and leaf bit predictor.technical 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card