README.md

metadata

dataset_info:
  features:
    - name: point_cloud
      dtype: binary
  splits:
    - name: train
      num_examples: 419
    - name: validation
      num_examples: 122
    - name: test
      num_examples: 120
tags:
  - point-cloud
  - lidar
  - semantic-segmentation
  - power-line
  - las
task_categories:
  - image-segmentation

TowerDataset

TowerDataset is a power-line corridor point cloud dataset for semantic segmentation. It contains annotated airborne/mobile LiDAR scenes of transmission towers, conductors, insulators, vegetation, ground objects, and other corridor elements.

Dataset Split

• Train: 419 scenes
• Validation: 122 scenes
• Test: 120 scenes
• Total: 661 scenes
• Total points: 2,466,076,987

File Structure

TowerDataset_v1/
  README.md
  manifest.json
  splits/
    train_files.txt
    val_files.txt
    test_files.txt
  train/
    *.las
  val/
    *.las
  test/
    *.las

Data Format

All point clouds are stored as .las files.

• LAS version: 1.2
• Point format: 2
• Coordinate scale: 0.001
• Point attributes: XYZ coordinates, RGB color, semantic label
• Semantic labels are stored in the LAS classification field

Semantic Classes

ID	Class
0	jumper
1	streetlight/sign
2	ground
3	building
4	road
5	greenhouse
6	railway
7	vehicle
8	low vegetation
9	high vegetation
10	conductor
11	tower
12	strain insulator
13	V-string insulator
14	spacer
15	distribution pole
16	distribution wire
17	water
18	suspension insulator
19	ground wire
20	OPGW
21	other