Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

D-PLS: Decoupled Semantic Segmentation for 4D-Panoptic-LiDAR-Segmentation

Maik Steinhauser, Laurenz Reichardt, Nikolas Ebert, Oliver Wasenmüller

TL;DR

Addresses the challenge of 4D-Panoptic LiDAR segmentation by decoupling semantic and instance segmentation. D-PLS uses single-scan semantic predictions as priors, aggregates them temporally, and applies a KPConv-based instance branch with a refinement pipeline and DBScan clustering. The evaluation on SemanticKITTI uses the LSTQ metric, defined as $LSTQ = \sqrt{S_{cls} \cdot S_{assoc}}$, and shows substantial gains over the 4D-StOP baseline, with results indicating improvements in both classification and association. The method is modular and can plug into any single-scan semantic segmentation backbone without architectural changes or retraining, enabling rapid adoption of advances in semantic segmentation. Overall, the work demonstrates that semantic priors can meaningfully boost 4D panoptic segmentation performance and suggests avenues for extending to other networks and datasets.

Abstract

This paper introduces a novel approach to 4D Panoptic LiDAR Segmentation that decouples semantic and instance segmentation, leveraging single-scan semantic predictions as prior information for instance segmentation. Our method D-PLS first performs single-scan semantic segmentation and aggregates the results over time, using them to guide instance segmentation. The modular design of D-PLS allows for seamless integration on top of any semantic segmentation architecture, without requiring architectural changes or retraining. We evaluate our approach on the SemanticKITTI dataset, where it demonstrates significant improvements over the baseline in both classification and association tasks, as measured by the LiDAR Segmentation and Tracking Quality (LSTQ) metric. Furthermore, we show that our decoupled architecture not only enhances instance prediction but also surpasses the baseline due to advancements in single-scan semantic segmentation.

D-PLS: Decoupled Semantic Segmentation for 4D-Panoptic-LiDAR-Segmentation

TL;DR

Addresses the challenge of 4D-Panoptic LiDAR segmentation by decoupling semantic and instance segmentation. D-PLS uses single-scan semantic predictions as priors, aggregates them temporally, and applies a KPConv-based instance branch with a refinement pipeline and DBScan clustering. The evaluation on SemanticKITTI uses the LSTQ metric, defined as , and shows substantial gains over the 4D-StOP baseline, with results indicating improvements in both classification and association. The method is modular and can plug into any single-scan semantic segmentation backbone without architectural changes or retraining, enabling rapid adoption of advances in semantic segmentation. Overall, the work demonstrates that semantic priors can meaningfully boost 4D panoptic segmentation performance and suggests avenues for extending to other networks and datasets.

Abstract

This paper introduces a novel approach to 4D Panoptic LiDAR Segmentation that decouples semantic and instance segmentation, leveraging single-scan semantic predictions as prior information for instance segmentation. Our method D-PLS first performs single-scan semantic segmentation and aggregates the results over time, using them to guide instance segmentation. The modular design of D-PLS allows for seamless integration on top of any semantic segmentation architecture, without requiring architectural changes or retraining. We evaluate our approach on the SemanticKITTI dataset, where it demonstrates significant improvements over the baseline in both classification and association tasks, as measured by the LiDAR Segmentation and Tracking Quality (LSTQ) metric. Furthermore, we show that our decoupled architecture not only enhances instance prediction but also surpasses the baseline due to advancements in single-scan semantic segmentation.
Paper Structure (9 sections, 1 equation, 2 figures, 2 tables)

This paper contains 9 sections, 1 equation, 2 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. APPROACH
  4. SEMANTIC SEGMENTATION
  5. TEMPORAL INSTANCE SEGMENTATION
  6. EXPERIMENTS
  7. ABLATION STUDY
  8. COMPARISON WITH STATE OF THE ART
  9. CONCLUSION

Figures (2)

  • Figure 1: Qualitative comparison of our methods (D-PLS) results when compared to the ground-truth on the SemanticKITTI dataset.
  • Figure 2: Our method D-PLS decomposes the task of 4D-Panoptic-LiDAR-Segmentation into two modular branches: the single scan semantic segmentation branch which provides prior information to the temporally aggregated instance segmentation branch.