README.md

---
license: etalab-2.0
task_categories:
  - image-classification
  - other
tags:
  - Aerial
  - Lidar
  - Point Cloud
  - Forest
  - Tree Species
  - Earth Observation
  - Multimodal
  - IGN
size_categories:
  - 100K<n<1M
---

# PureForest Clone

This is a clone of the original [IGNF/PureForest](https://huggingface.co/datasets/IGNF/PureForest) dataset,
re-uploaded as zip archives for easier downloading and use.

> **Original paper:** [PureForest: A Large-Scale Aerial Lidar and Aerial Imagery Dataset for Tree Species Classification in Monospecific Forests](https://arxiv.org/abs/2404.12064)
> Charles Gaydon, Floryne Roche — IGN, 2024

---

## Dataset Overview

PureForest is the largest publicly available dataset for tree species classification from:
- **ALS (Aerial Lidar Scanning)** point clouds — high density: ~40 pts/m²
- **VHR (Very High Resolution)** aerial images — 0.2 m resolution, 250×250 pixels, NIRGB channels

| Property | Value |
|----------|-------|
| Total patches | 135,569 |
| Coverage | 339 km² |
| Forests | 449 distinct monospecific forests |
| Departments | 40 French departments |
| Semantic classes | 13 (grouping 18 tree species) |
| Patch size | 50 m × 50 m |

---

## Dataset Structure

```
PureForest_clone/
├── train.zip        # 69,111 patches (~51.4 GB)
├── val.zip          # 13,523 patches (~10 GB)
├── test.zip         # 52,935 patches (~39.3 GB)
└── metadata/        # Metadata files
```

Each `.npz` patch file contains:
- `intensities` — Lidar intensity values
- `classes` — semantic class label (integer 0–12)
- Point cloud coordinates (x, y, z) colorized with aerial imagery

### File Naming Convention

```
{SPLIT}-{ClassName}-C{ClassID}-{PatchID}.npz

Example: TRAIN-Quercus_ilex-C1-177_13_242.npz
         TEST-Abies_alba-C9-001_02_015.npz
```

### Semantic Classes

| Class ID | Species | Common Name |
|----------|---------|-------------|
| 0 | Quercus (deciduous) | Deciduous oak |
| 1 | Quercus ilex | Evergreen oak |
| 2 | Fagus sylvatica | Beech |
| 3 | Castanea sativa | Chestnut |
| 4 | Robinia pseudoacacia | Black locust |
| 5 | Pinus pinaster | Maritime pine |
| 6 | Pinus sylvestris | Scotch pine |
| 7 | Pinus nigra | Black pine |
| 8 | Pinus halepensis | Aleppo pine |
| 9 | Abies alba | Fir |
| 10 | Picea abies | Spruce |
| 11 | Larix decidua | Larch |
| 12 | Pseudotsuga menziesii | Douglas |

### Train/Val/Test Split Distribution

| Class | Train | Val | Test |
|-------|-------|-----|------|
| (0) Deciduous oak | 22.92% | 32.35% | 52.59% |
| (1) Evergreen oak | 16.80% | 2.75% | 19.61% |
| (2) Beech | 10.14% | 12.03% | 7.62% |
| (3) Chestnut | 4.83% | 1.09% | 0.38% |
| (4) Black locust | 2.41% | 2.40% | 0.60% |
| (5) Maritime pine | 6.61% | 7.10% | 3.85% |
| (6) Scotch pine | 16.39% | 17.95% | 8.51% |
| (7) Black pine | 6.30% | 6.98% | 3.64% |
| (8) Aleppo pine | 5.83% | 1.72% | 0.83% |
| (9) Fir | 0.14% | 5.32% | 0.05% |
| (10) Spruce | 3.73% | 4.64% | 1.64% |
| (11) Larch | 3.67% | 3.73% | 0.48% |
| (12) Douglas | 0.23% | 1.95% | 0.20% |

---

## Download & Setup

### Step 1 — Download zip files

```python
from huggingface_hub import hf_hub_download

for split in ["train", "val", "test"]:
    hf_hub_download(
        repo_id="longdpkr/PureForest_clone",
        filename=f"{split}.zip",
        repo_type="dataset",
        local_dir="./PureForest"
    )
```

Or using the CLI:
```bash
huggingface-cli download longdpkr/PureForest_clone train.zip --repo-type=dataset
huggingface-cli download longdpkr/PureForest_clone val.zip --repo-type=dataset
huggingface-cli download longdpkr/PureForest_clone test.zip --repo-type=dataset
```

### Step 2 — Extract zip files

```python
import zipfile
from pathlib import Path

dataset_dir = Path("./PureForest")

for split in ["train", "val", "test"]:
    zip_path = dataset_dir / f"{split}.zip"
    print(f"Extracting {split}.zip ...")
    with zipfile.ZipFile(zip_path, "r") as zf:
        zf.extractall(dataset_dir)
    print(f"Done: {split}/")
```

Or manually using WinRAR / 7-Zip.

After extraction, the structure should be:
```
PureForest/
├── train/
│   ├── TRAIN-Quercus_ilex-C1-xxx.npz
│   └── ...
├── val/
│   ├── VAL-Fagus_sylvatica-C2-xxx.npz
│   └── ...
└── test/
    ├── TEST-Pinus_sylvestris-C6-xxx.npz
    └── ...
```

---

## Loading Data

### Load a single patch

```python
import numpy as np

data = np.load("train/TRAIN-Quercus_ilex-C1-177_13_242.npz")

print(list(data.keys()))     # show available keys
print(data["classes"])       # class label (0-12)
```

### PyTorch Dataset

```python
import numpy as np
import torch
from torch.utils.data import Dataset
from pathlib import Path


class PureForestDataset(Dataset):
    def __init__(self, split="train", data_dir="./PureForest"):
        self.files = sorted(Path(data_dir) / split).glob("*.npz"))
        
    def __len__(self):
        return len(self.files)
    
    def __getitem__(self, idx):
        data = np.load(self.files[idx])
        label = int(data["classes"])
        # add your preprocessing here
        return data, label


train_dataset = PureForestDataset(split="train")
val_dataset   = PureForestDataset(split="val")
test_dataset  = PureForestDataset(split="test")

print(f"Train: {len(train_dataset)} | Val: {len(val_dataset)} | Test: {len(test_dataset)}")
```

### With RandLA-Net

This dataset is compatible with [RandLA-Net](https://github.com/QingyongHu/RandLA-Net) for point cloud classification.
Place extracted folders under the `datasets/` directory of your RandLA-Net project:

```
RandLA-Net/
└── datasets/
    └── PureForest/
        ├── train/
        ├── val/
        └── test/
```

---

## License

This dataset is a clone of [IGNF/PureForest](https://huggingface.co/datasets/IGNF/PureForest)
and inherits its original license: **[Etalab Open License 2.0](https://www.etalab.gouv.fr/licence-ouverte-open-licence/)**.

---

## Citation

If you use this dataset in your research, please cite the original paper:

```bibtex
@misc{gaydon2024pureforest,
    title     = {PureForest: A Large-Scale Aerial Lidar and Aerial Imagery Dataset
                 for Tree Species Classification in Monospecific Forests},
    author    = {Charles Gaydon and Floryne Roche},
    year      = {2024},
    eprint    = {2404.12064},
    archivePrefix = {arXiv},
    url       = {https://arxiv.org/abs/2404.12064}
}
```

---

## Acknowledgements

Original dataset by [IGN — Institut national de l'information géographique et forestière](https://www.ign.fr/).
Lidar data from the [Lidar HD program](https://geoservices.ign.fr/lidarhd).
Aerial imagery from [ORTHO HR®](https://geoservices.ign.fr/bdortho).