Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GridNet-HD: A High-Resolution Multi-Modal Dataset for LiDAR-Image Fusion on Power Line Infrastructure

Antoine Carreaud, Shanci Li, Malo De Lacour, Digre Frinde, Jan Skaloud, Adrien Gressin

TL;DR

GridNet-HD introduces a public, high-resolution multimodal dataset for 3D semantic segmentation of overhead electrical infrastructure, pairing dense LiDAR ($200$–$800$ pts/m²) with oblique imagery (GSD ~1.5 cm). It provides $7{,}694$ images and $2.5$ billion co-registered LiDAR points, annotated into $11$ classes, with rigorous co-registration via direct georeferencing, aerotriangulation, and GCPs, plus depth-aware reprojection to 2D. Across baseline experiments, image–LiDAR fusion outperforms unimodal approaches, with Late Fusion achieving $74.22 ext{%}$ mIoU and DITR reaching $74.87 ext{%}$ mIoU in some configurations, while alignment quality drives gains more than fusion complexity (e.g., a +$5.55$ mIoU improvement over best unimodal). The dataset enables robust benchmarking and is accompanied by a public leaderboard, facilitating broader research in multimodal fusion for aerial electrical infrastructure inspection and beyond. $GridNet-HD$ thus offers a valuable resource for both domain-specific power-line monitoring and general LiDAR–image fusion research.

Abstract

This paper presents GridNet-HD, a multi-modal dataset for 3D semantic segmentation of overhead electrical infrastructures, pairing high-density LiDAR with high-resolution oblique imagery. The dataset comprises 7,694 images and 2.5 billion points annotated into 11 classes, with predefined splits and mIoU metrics. Unimodal (LiDAR-only, image-only) and multi-modal fusion baselines are provided. On GridNet-HD, fusion models outperform the best unimodal baseline by +5.55 mIoU, highlighting the complementarity of geometry and appearance. As reviewed in Sec. 2, no public dataset jointly provides high-density LiDAR and high-resolution oblique imagery with 3D semantic labels for power-line assets. Dataset, baselines, and codes are available: https://huggingface.co/collections/heig-vd-geo/gridnet-hd.

GridNet-HD: A High-Resolution Multi-Modal Dataset for LiDAR-Image Fusion on Power Line Infrastructure

TL;DR

GridNet-HD introduces a public, high-resolution multimodal dataset for 3D semantic segmentation of overhead electrical infrastructure, pairing dense LiDAR (– pts/m²) with oblique imagery (GSD ~1.5 cm). It provides images and billion co-registered LiDAR points, annotated into classes, with rigorous co-registration via direct georeferencing, aerotriangulation, and GCPs, plus depth-aware reprojection to 2D. Across baseline experiments, image–LiDAR fusion outperforms unimodal approaches, with Late Fusion achieving mIoU and DITR reaching mIoU in some configurations, while alignment quality drives gains more than fusion complexity (e.g., a + mIoU improvement over best unimodal). The dataset enables robust benchmarking and is accompanied by a public leaderboard, facilitating broader research in multimodal fusion for aerial electrical infrastructure inspection and beyond. thus offers a valuable resource for both domain-specific power-line monitoring and general LiDAR–image fusion research.

Abstract

This paper presents GridNet-HD, a multi-modal dataset for 3D semantic segmentation of overhead electrical infrastructures, pairing high-density LiDAR with high-resolution oblique imagery. The dataset comprises 7,694 images and 2.5 billion points annotated into 11 classes, with predefined splits and mIoU metrics. Unimodal (LiDAR-only, image-only) and multi-modal fusion baselines are provided. On GridNet-HD, fusion models outperform the best unimodal baseline by +5.55 mIoU, highlighting the complementarity of geometry and appearance. As reviewed in Sec. 2, no public dataset jointly provides high-density LiDAR and high-resolution oblique imagery with 3D semantic labels for power-line assets. Dataset, baselines, and codes are available: https://huggingface.co/collections/heig-vd-geo/gridnet-hd.
Paper Structure (39 sections, 10 equations, 13 figures, 3 tables)

This paper contains 39 sections, 10 equations, 13 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Summary of datasets
  4. Identified gaps
  5. Autonomous driving
  6. Indoor scene understanding
  7. Electrical infrastructure monitoring
  8. Addressing the gap
  9. Dataset description
  10. Dataset acquisition
  11. Dataset co-registration
  12. Dataset annotation
  13. Semantic class definition
  14. Manual labelling process
  15. Train/test split
  16. ...and 24 more sections

Figures (13)

  • Figure 1: Schematic view of a flight plan in comparison with the power line. The angle $\alpha$ (here of 50°) of the oblique view and the height H over the power line are input parameters of the calculated trajectory. (a) side view, (b) top view, (c) perspective view.
  • Figure 2: Manual GCP clicking for precise image–LiDAR co-registration. Left: before refinement, 3D points (filled dots) project away from the clicked image features (rings) across views. Right: after aerotriangulation using GCPs, projections align with image features, reducing residuals and enforcing consistent geometry.
  • Figure 3: Successful 3D to 2D label projection checks after final alignment refinement.
  • Figure 4: Qualitative comparison across three baselines on GridNet-HD. From left to right: (a) Ground Truth; (b) ImageVote shows small projection issue and occlusion artifacts near object boundaries; (c) PTv3 (XYZ+RGB) yields smoother regions but misses thin structures (cables/insulators); (d) DITR produces the cleanest boundaries.
  • Figure 5: Geographic distribution of the acquisition zones included in the GridNet-HD dataset. Each zone corresponds to a specific area captured by UAV.
  • ...and 8 more figures