Example Configuration Files

Version 2.3.0 includes four production-ready YAML configuration files for common workflows. These configs provide a simple way to get started without manually specifying all parameters.

📁 Location

All example configs are located in the examples/ directory:

examples/
├── README.md
├── config_gpu_processing.yaml
├── config_training_dataset.yaml
├── config_quick_enrich.yaml
└── config_complete.yaml

🚀 Usage

Basic Usage

ign-lidar-hd process \
  --config-file examples/config_training_dataset.yaml \
  input_dir=/path/to/raw \
  output_dir=/path/to/output

Preview Configuration

See what the config does before running:

ign-lidar-hd process \
  --config-file examples/config_training_dataset.yaml \
  --show-config \
  input_dir=/path/to/raw \
  output_dir=/path/to/output

Override Config Values

CLI arguments override config file values:

ign-lidar-hd process \
  --config-file examples/config_training_dataset.yaml \
  input_dir=/path/to/raw \
  output_dir=/path/to/output \
  processor.num_points=65536 \
  features.k_neighbors=50

📝 Available Configurations

1. `config_gpu_processing.yaml` - GPU-Accelerated LAZ Enrichment

Use Case: Fast feature enrichment with GPU acceleration for GIS analysis

Key Features:

✅ GPU acceleration enabled
✅ Enriched LAZ output only (no patches)
✅ Full feature computation (RGB, NIR, NDVI)
✅ Preprocessing and stitching enabled

Configuration:

defaults:
  - override /processor: gpu
  - override /features: full
  - override /preprocess: default
  - override /stitching: enhanced

processor:
  lod_level: LOD2
  architecture: default
  num_points: 16384
  use_gpu: true
  augment: false

features:
  k_neighbors: 20
  use_rgb: true
  compute_ndvi: true

output:
  format: npz
  processing_mode: enriched_only

Usage:

ign-lidar-hd process \
  --config-file examples/config_gpu_processing.yaml \
  input_dir=/path/to/raw \
  output_dir=/path/to/enriched

Output: tile_enriched.laz files with computed features

2. `config_training_dataset.yaml` - ML Training Dataset

Use Case: Create ML-ready patches with augmentation for training deep learning models

Key Features:

✅ Data augmentation (5x per patch)
✅ Preprocessing enabled (outlier removal)
✅ Tile stitching for boundary consistency
✅ Normalized features for deep learning
✅ Patches only (no enriched LAZ)

Configuration:

defaults:
  - override /processor: default
  - override /features: full
  - override /preprocess: default
  - override /stitching: enhanced

processor:
  lod_level: LOD2
  architecture: default
  num_points: 16384
  sampling_method: random
  augment: true
  num_augmentations: 5
  use_gpu: false

features:
  k_neighbors: 20
  use_rgb: true
  compute_ndvi: true
  normalize_xyz: true

output:
  format: npz
  processing_mode: patches_only

Usage:

ign-lidar-hd process \
  --config-file examples/config_training_dataset.yaml \
  input_dir=/path/to/raw \
  output_dir=/path/to/training

Output:

tile_patch_0001.npz, tile_patch_0001_aug_0.npz, ...
Each patch has 5 augmented versions

3. `config_quick_enrich.yaml` - Quick LAZ Enrichment

Use Case: Fastest option for adding features to LAZ files (GIS workflows)

Key Features:

✅ Minimal features (fastest processing)
✅ No preprocessing or stitching
✅ Enriched LAZ output only
✅ CPU-based (no GPU required)

Configuration:

defaults:
  - override /processor: cpu_fast
  - override /features: minimal

processor:
  lod_level: LOD2
  architecture: default
  num_points: 16384
  use_gpu: false
  augment: false

features:
  k_neighbors: 10
  use_rgb: false
  compute_ndvi: false

preprocess:
  enabled: false

stitching:
  enabled: false

output:
  format: npz
  processing_mode: enriched_only

Usage:

ign-lidar-hd process \
  --config-file examples/config_quick_enrich.yaml \
  input_dir=/path/to/raw \
  output_dir=/path/to/enriched

Output: tile_enriched.laz files with basic geometric features

4. `config_complete.yaml` - Complete Workflow

Use Case: Research projects needing both ML patches AND enriched LAZ files

Key Features:

✅ Both patches and enriched LAZ output
✅ Full preprocessing and stitching
✅ RGB, NIR, and NDVI computation
✅ Data augmentation enabled

Configuration:

defaults:
  - override /processor: default
  - override /features: full
  - override /preprocess: default
  - override /stitching: enhanced

processor:
  lod_level: LOD2
  architecture: default
  num_points: 16384
  sampling_method: random
  augment: true
  num_augmentations: 3
  use_gpu: false

features:
  k_neighbors: 20
  use_rgb: true
  compute_ndvi: true
  normalize_xyz: true

output:
  format: npz
  processing_mode: both

Usage:

# Option 1: Edit paths in the file, then:
ign-lidar-hd process --config-file examples/config_complete.yaml

# Option 2: Override paths from command line:
ign-lidar-hd process \
  --config-file examples/config_complete.yaml \
  input_dir=/path/to/raw \
  output_dir=/path/to/complete

Output:

ML patches: tile_patch_0001.npz, tile_patch_0001_aug_0.npz, ...
GIS files: tile_enriched.laz

🔧 Configuration Precedence

The system follows a clear precedence order for configuration values:

Package Defaults (lowest priority)
- Built-in defaults from ign_lidar/configs/
Custom Config File (medium priority)
- Values from your --config-file
CLI Overrides (highest priority)
- Parameters passed directly on command line

Example

# config_training_dataset.yaml has num_points=16384
# This command overrides it to 32768
ign-lidar-hd process \
  --config-file examples/config_training_dataset.yaml \
  input_dir=/path/to/raw \
  output_dir=/path/to/output \
  processor.num_points=32768  # <-- Overrides config file

📊 Comparison Table

Config	Processing Mode	GPU	Augment	RGB/NDVI	Preprocess	Stitching	Speed	Use Case
`config_gpu_processing`	enriched_only	✅	❌	✅	✅	✅	Very Fast	GIS analysis
`config_training_dataset`	patches_only	❌	✅ (5x)	✅	✅	✅	Medium	ML training
`config_quick_enrich`	enriched_only	❌	❌	❌	❌	❌	Fastest	Quick GIS enrichment
`config_complete`	both	❌	✅ (3x)	✅	✅	✅	Slow	Research/Full output

💡 Tips

When to Use Each Config

Use config_gpu_processing.yaml when:

You have NVIDIA GPU with CUDA support
You need enriched LAZ files for QGIS/CloudCompare
Speed is critical
You don't need ML patches

Use config_training_dataset.yaml when:

Training PointNet++, transformers, or other models
You need data augmentation
You want boundary-consistent features
ML is your primary goal

Use config_quick_enrich.yaml when:

You just need basic features added to LAZ
Speed is the top priority
You're on a laptop or limited hardware
No ML training needed

Use config_complete.yaml when:

You need both ML patches AND enriched LAZ
Running a research project
Want to compare ML vs GIS workflows
Disk space is not a concern

Customizing Configs

Create your own config based on the examples:

# my_custom_config.yaml
defaults:
  - override /processor: gpu
  - override /features: full

processor:
  lod_level: LOD3
  num_points: 32768
  use_gpu: true
  augment: true
  num_augmentations: 10

features:
  k_neighbors: 30
  use_rgb: true
  compute_ndvi: true

output:
  format: "npz,torch" # Multi-format output
  processing_mode: patches_only

input_dir: /mnt/data/lidar_tiles
output_dir: /mnt/data/training_dataset

Then use it:

ign-lidar-hd process --config-file my_custom_config.yaml

Processing Modes Guide - Deep dive into processing modes
Hydra CLI Guide - Learn about Hydra configuration system
Configuration System - Complete config reference
GPU Acceleration - GPU setup and optimization

📖 Full Example: Training Workflow

Here's a complete workflow using the example configs:

# Step 1: Download tiles
ign-lidar-hd download \
  --position 650000 6860000 \
  --radius 5000 \
  data/raw_tiles

# Step 2: Create training dataset with augmentation
ign-lidar-hd process \
  --config-file examples/config_training_dataset.yaml \
  input_dir=data/raw_tiles \
  output_dir=data/training

# Step 3: Verify dataset
ign-lidar-hd verify data/training

# Step 4: Preview statistics
python -c "
import numpy as np
from pathlib import Path

patches = list(Path('data/training').glob('*.npz'))
print(f'Total patches: {len(patches)}')

sample = np.load(patches[0])
print(f'Points per patch: {sample[\"points\"].shape[0]}')
print(f'Features: {sample[\"points\"].shape[1]}')
print(f'Available keys: {list(sample.keys())}')
"

This produces a production-ready ML training dataset with:

✅ Consistent patch sizes
✅ Boundary-aware features
✅ 5x augmentation per patch
✅ RGB and NDVI features
✅ Normalized coordinates

📁 Location​

🚀 Usage​

Basic Usage​

Preview Configuration​

Override Config Values​

📝 Available Configurations​

1. config_gpu_processing.yaml - GPU-Accelerated LAZ Enrichment​

2. config_training_dataset.yaml - ML Training Dataset​

3. config_quick_enrich.yaml - Quick LAZ Enrichment​

4. config_complete.yaml - Complete Workflow​

🔧 Configuration Precedence​

Example​

📊 Comparison Table​

💡 Tips​

When to Use Each Config​

Customizing Configs​

🔗 Related Documentation​

📖 Full Example: Training Workflow​

📁 Location

🚀 Usage

Basic Usage

Preview Configuration

Override Config Values

📝 Available Configurations

1. `config_gpu_processing.yaml` - GPU-Accelerated LAZ Enrichment

2. `config_training_dataset.yaml` - ML Training Dataset

3. `config_quick_enrich.yaml` - Quick LAZ Enrichment

4. `config_complete.yaml` - Complete Workflow

🔧 Configuration Precedence

Example

📊 Comparison Table

💡 Tips

When to Use Each Config

Customizing Configs

🔗 Related Documentation

📖 Full Example: Training Workflow