Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improved Block Merging for 3D Point Cloud Instance Segmentation

Leon Denis, Remco Royen, Adrian Munteanu

TL;DR

The paper addresses inaccuracies in block-based 3D point cloud instance segmentation caused by non-overlapping blocks during merging. It introduces BlockMerging v2, a label-propagation strategy that stores per-instance label candidates and rotates labels toward a global consensus across blocks, controlled by a tunable threshold to balance recall and precision. The approach is architecture-agnostic and improves multiple metrics across several state-of-the-art models and voxel resolutions, with a modest runtime overhead. Results demonstrate substantial gains over the traditional BlockMerging v1, suggesting broad applicability to existing block-based methods in 3D scene understanding.

Abstract

This paper proposes a novel block merging algorithm suitable for any block-based 3D instance segmentation technique. The proposed work improves over the state-of-the-art by allowing wrongly labelled points of already processed blocks to be corrected through label propagation. By doing so, instance overlap between blocks is not anymore necessary to produce the desirable results, which is the main limitation of the current art. Our experiments show that the proposed block merging algorithm significantly and consistently improves the obtained accuracy for all evaluation metrics employed in literature, regardless of the underlying network architecture.

Improved Block Merging for 3D Point Cloud Instance Segmentation

TL;DR

The paper addresses inaccuracies in block-based 3D point cloud instance segmentation caused by non-overlapping blocks during merging. It introduces BlockMerging v2, a label-propagation strategy that stores per-instance label candidates and rotates labels toward a global consensus across blocks, controlled by a tunable threshold to balance recall and precision. The approach is architecture-agnostic and improves multiple metrics across several state-of-the-art models and voxel resolutions, with a modest runtime overhead. Results demonstrate substantial gains over the traditional BlockMerging v1, suggesting broad applicability to existing block-based methods in 3D scene understanding.

Abstract

This paper proposes a novel block merging algorithm suitable for any block-based 3D instance segmentation technique. The proposed work improves over the state-of-the-art by allowing wrongly labelled points of already processed blocks to be corrected through label propagation. By doing so, instance overlap between blocks is not anymore necessary to produce the desirable results, which is the main limitation of the current art. Our experiments show that the proposed block merging algorithm significantly and consistently improves the obtained accuracy for all evaluation metrics employed in literature, regardless of the underlying network architecture.
Paper Structure (6 sections, 3 equations, 4 figures, 2 tables, 1 algorithm)

This paper contains 6 sections, 3 equations, 4 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. BlockMerging v1
  3. Limitations
  4. Proposed method
  5. Experimental assessments
  6. Conclusion

Figures (4)

  • Figure 1: Top down view of a room divided in blocks. The portion of the sofa marked in blue will be wrongly labelled using current block merging techniques wang2018sgpn as no overlap is present.
  • Figure 2: The effect of the parameter $\tau_2$ with respect to the mean precision and mean recall for different voxel sizes.
  • Figure 3: Results for grid resolution $400^3$. From left to right: input, ground truth, BlockMerging v1 wang2018sgpn, proposed.
  • Figure 4: Visual results for grid size $500^3$.