Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ParisLuco3D: A high-quality target dataset for domain generalization of LiDAR perception

Jules Sanchez, Louis Soum-Fontez, Jean-Emmanuel Deschaud, Francois Goulette

TL;DR

ParisLuco3D addresses the challenge of evaluating LiDAR perception models under cross-domain shifts by providing a Paris-centered dataset with annotations remapped to standard LSS/LOD label spaces, plus online benchmarks for segmentation, detection, and tracking. It analyzes domain generalization using multiple source datasets (SemanticKITTI, nuScenes, ONCE) across several architectures, revealing substantial generalization gaps and that current methods offer inconsistent gains. Key findings include that no single generalization strategy consistently boosts performance across domains, SRUNet often yields the strongest cross-domain segmentation performance, and intensity channel usage has dataset-dependent effects on generalization. The dataset and benchmark aim to standardize cross-domain evaluation, highlighting the need for architectures and training strategies explicitly targeting domain-invariant representations for robust real-world LiDAR perception.

Abstract

LiDAR is an essential sensor for autonomous driving by collecting precise geometric information regarding a scene. %Exploiting this information for perception is interesting as the amount of available data increases. As the performance of various LiDAR perception tasks has improved, generalizations to new environments and sensors has emerged to test these optimized models in real-world conditions. This paper provides a novel dataset, ParisLuco3D, specifically designed for cross-domain evaluation to make it easier to evaluate the performance utilizing various source datasets. Alongside the dataset, online benchmarks for LiDAR semantic segmentation, LiDAR object detection, and LiDAR tracking are provided to ensure a fair comparison across methods. The ParisLuco3D dataset, evaluation scripts, and links to benchmarks can be found at the following website:https://npm3d.fr/parisluco3d

ParisLuco3D: A high-quality target dataset for domain generalization of LiDAR perception

TL;DR

ParisLuco3D addresses the challenge of evaluating LiDAR perception models under cross-domain shifts by providing a Paris-centered dataset with annotations remapped to standard LSS/LOD label spaces, plus online benchmarks for segmentation, detection, and tracking. It analyzes domain generalization using multiple source datasets (SemanticKITTI, nuScenes, ONCE) across several architectures, revealing substantial generalization gaps and that current methods offer inconsistent gains. Key findings include that no single generalization strategy consistently boosts performance across domains, SRUNet often yields the strongest cross-domain segmentation performance, and intensity channel usage has dataset-dependent effects on generalization. The dataset and benchmark aim to standardize cross-domain evaluation, highlighting the need for architectures and training strategies explicitly targeting domain-invariant representations for robust real-world LiDAR perception.

Abstract

LiDAR is an essential sensor for autonomous driving by collecting precise geometric information regarding a scene. %Exploiting this information for perception is interesting as the amount of available data increases. As the performance of various LiDAR perception tasks has improved, generalizations to new environments and sensors has emerged to test these optimized models in real-world conditions. This paper provides a novel dataset, ParisLuco3D, specifically designed for cross-domain evaluation to make it easier to evaluate the performance utilizing various source datasets. Alongside the dataset, online benchmarks for LiDAR semantic segmentation, LiDAR object detection, and LiDAR tracking are provided to ensure a fair comparison across methods. The ParisLuco3D dataset, evaluation scripts, and links to benchmarks can be found at the following website:https://npm3d.fr/parisluco3d
Paper Structure (42 sections, 1 equation, 8 figures, 11 tables)

This paper contains 42 sections, 1 equation, 8 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LiDAR Datasets in autonomous driving
  4. Datasets for LiDAR semantic segmentation
  5. Datasets for LiDAR object detection
  6. Domain generalization for LiDAR perception
  7. Dataset presentation
  8. Acquisition and data generation
  9. Acquisition
  10. Data generation
  11. Label shift and annotation process
  12. Label shift
  13. Annotation process
  14. Labels for semantic segmentation
  15. Labels for object detection
  16. ...and 27 more sections

Figures (8)

  • Figure 1: A LiDAR scan of our ParisLuco3D dataset with ground truth annotation for semantic segmentation and object detection.
  • Figure 2: Trajectory of our dataset (2.1 km around the Luxembourg Garden in Paris) overlaid on a Google Satellite Image.
  • Figure 3: Illustrations depicting the diversity of scenes and quality of annotation in the ParisLuco3D dataset (left: point clouds by accumulating the scans of our dataset colorized with labels; right: images taken from Google Street View).
  • Figure 4: Distribution of labels in ParisLuco3D.
  • Figure 5: Labels for LSS.
  • ...and 3 more figures