Datasets:

LouisGeist
/

MALiBU3D

license: etalab-2.0
pretty_name: MALiBU3D
task_categories:
  - image-segmentation
size_categories:
  - n>1T
tags:
  - point-cloud
  - lidar
  - remote-sensing
  - semantic-segmentation
  - france
  - ign
configs:
  - config_name: catalog
    data_files: tiles.parquet

MALiBU3D

Large-scale training-ready airborne LiDAR point clouds over France, with land-cover labels, natural-habitat axes, canopy-height (elevation), RGB, and road-network graphs. Built on IGNF/FLAIR-HUB 100 m tiles plus IGN LiDAR HD. Former working name: Flair3D.

This Hub repo ships the preprocessed arrays used for training (NumPy .npy inside one zip per ROI), not the raw GeoTIFFs / PLYs. Unzip onto local scratch for training (np.load / mmap). The zip is the distribution format.

A tile is a ~100×100 m square. An ROI groups neighbouring tiles (and holds the optional road graph).

Download

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="ORG/MALiBU3D",
    repo_type="dataset",
    allow_patterns=["data/train/D075-2021_LIDARHD/*", "labels.json", "tiles.csv"],
)

Do not load_dataset() on the point clouds (variable N ≈ 1e5–3e5). The Dataset Viewer reads tiles.parquet (one row per tile, no xyz). tiles.csv is the same table when parquet was not built. Tiles with missing LiDAR coordinates are omitted from the catalog.

A tiny extracted ROI lives under toy/ for inspection.

Layout

labels.json
palettes.json
scene_split_manifest.csv     # original split table (all FLAIR-HUB rows)
tiles.parquet                # Hub viewer catalog (released tiles only)
tiles.csv                    # same catalog (zip_path, forest georef, n_points, …)
SHA256SUMS
toy/                          # one extracted ROI
data/{train,val,test}/{dept}_LIDARHD/{roi}.zip

Each {roi}.zip (flat, no wrapping {roi}/ folder):

{tile_id}/coord.npy                 float32 (N, 3)   XYZ relative
{tile_id}/coord_translation.npy     float64 (3,)     Lambert-93 offset
{tile_id}/color.npy                 uint8   (N, 3)
{tile_id}/segment.npy               uint8   (N,)     land cover, Void=15
{tile_id}/strength.npy              float32 (N,)     LiDAR intensity ~[0, 1]
{tile_id}/elevation.npy             float32 (N,)     z − DTM (optional)
{tile_id}/natural_habitat.npy       uint8   (N, 4)   ecological axes (optional)
{tile_id}/forest_2d.npy             uint8   (1, H, W)
{deptcode}_{roi}_ROADS_graph.gpkg    optional, EPSG:2154

Absolute coordinates: xyz_abs = coord + coord_translation (EPSG:2154).

Not included: per-tile meta.json, network.npy, per-point forest.npy, land_use.npy, rail / transmission-line graphs. RAILROADS / TRANSMISSION_LINES columns in the catalog remain as FLAIR-HUB availability flags.

Manifest vs catalog

scene_split_manifest.csv is the original split table (identifiers only, column patch_id = the 100 m tile). It still lists FLAIR-HUB rows without LiDAR. Reconstruct on disk:

{data_root}/{split}/{dept_year}_LIDARHD/{roi}/{tile_id}

tiles.csv / tiles.parquet is the release catalog: only tiles shipped here (tile_id = that same identifier), plus a relative zip_path (data/train/D075-2021_LIDARHD/UU-S1-4.zip), n_points, and forest_2d georeferencing (forest_origin_x, forest_origin_y, forest_width, forest_height, forest_pixel_m). CRS, south-up axis, and class values are global (labels.json).

Labels

See labels.json. Land cover (segment.npy): 15 train classes + Void=15.

Natural habitat (natural_habitat.npy): (N, 4) uint8, already remapped from CarHab. Column order is labels.json → natural_habitat.columns:

col	key	classes	Void
0	`nathab_habitat_type`	Open, Forest, Mineral, Aquatic	4
1	`nathab_moisture_regime`	Humide, Mesique, Sec	3
2	`nathab_soil_chemistry`	Acidic, Alkaline	2
3	`nathab_bioclimatic_zone`	Temperate, Mediterranean, Alpine	3

import json, numpy as np
labels = json.load(open("labels.json"))
nh = np.load("natural_habitat.npy")          # (N, 4)
moisture = nh[:, 1]                          # Void = 3
# Pointcept: no LUT. Assign column i to task labels["natural_habitat"]["columns"][i].

`forest_2d` → points

South-up Lambert-93 grid. Always read forest_pixel_m / origin from the catalog (do not assume 0.5 m). Width/height also vary slightly per tile.

import numpy as np
import pandas as pd

row = pd.read_csv("tiles.csv").set_index("tile_id").loc[tile_id]
coord = np.load("coord.npy")
t = np.load("coord_translation.npy")
raster = np.load("forest_2d.npy")  # (1, H, W), 0 / 1 / 2=void
x = coord[:, 0] + t[0]
y = coord[:, 1] + t[1]
ix = np.floor((x - row.forest_origin_x) / row.forest_pixel_m).astype(int)
iy = np.floor((y - row.forest_origin_y) / row.forest_pixel_m).astype(int)
h, w = int(row.forest_height), int(row.forest_width)
inside = (ix >= 0) & (ix < w) & (iy >= 0) & (iy < h)
out = np.full(len(coord), 2, dtype=np.uint8)
out[inside] = raster[0, iy[inside], ix[inside]]

Road graphs: GeoPackage layers nodes, edges, metadata, coordinates absolute EPSG:2154.

Licence and attribution

Licence Ouverte 2.0 / Etalab (see LICENSE). Attribute IGN and FLAIR-HUB. Derived semantic labels and graphs are produced by this project.

Loader notes

On-disk segment is uint8 (N,). natural_habitat is uint8 (N, 4). Cast if a loader asserts another integer dtype. Per-point forest.npy, land_use.npy, and network.npy are absent.