Dataset assetOpen Source CommunityInstance SegmentationUrban Mapping

NPM3D dataset with instance labels

The NPM3D urban mobile mapping dataset includes instance labels and is intended for precise instance segmentation research on large‑scale LiDAR point clouds.

Source

github

Created

Jul 5, 2023

Updated

May 23, 2024

Signals

161 views

Availability

Linked source ready

Overview

Dataset description and usage context

Dataset Overview

NPM3D Dataset

Download Link: NPM3D Dataset Download
Semantic Labels:
- 1: "ground"
- 2: "buildings"
- 3: "poles"
- 4: "bollards"
- 5: "trash cans"
- 6: "barriers"
- 7: "pedestrians"
- 8: "cars"
- 9: "natural"

Directory Structure:

data/
  npm3dfused/
    raw/
      *_train.ply          # training data
      *_val.ply            # validation data
      *_test.ply           # test data

FOR‑instance Dataset

Download Link: FOR‑instance Dataset Download
Forest Area Details:
Forest Area File Count Max Tree Height
CULS 3 25 m
NIBIO 20 30 m
NIBIO2 50 25 m
RMIT 2 15 m
SCION 5 30 m
TUWIEN 2 35 m
Splits:
- Training: 56 files
- Validation: 14 files
- Test: 26 files

Forest Area	File Count	Max Tree Height
CULS	3	25 m
NIBIO	20	30 m
NIBIO2	50	25 m
RMIT	2	15 m
SCION	5	30 m
TUWIEN	2	35 m

Directory Structure:

data/
  treeinsfused/
    raw/
      CULS/
        *_train.ply
        *_val.ply
        *_test.ply
      NIBIO/
        *_train.ply
        *_val.ply
        *_test.ply
      NIBIO2/
        *.ply
      RMIT/
        *.ply
      SCION/
        *.ply
      TUWIEN/
        *.ply

Need downstream help?

Pair the dataset with AI analysis and content workflows.

Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.

Explore AI studio