New Feature: RGB Augmentation from IGN Orthophotos

October 3, 2025 · 7 min read

Lead Developer & Researcher

We're excited to announce a major new feature for the IGN LiDAR HD Processing Library: RGB Augmentation from IGN Orthophotos! 🎨

This feature automatically enriches your LiDAR point clouds with RGB colors fetched directly from IGN's high-resolution orthophoto service, enabling multi-modal machine learning and enhanced visualization.

What's New?

🎨 RGB Augmentation

Your LiDAR patches can now include RGB color information automatically extracted from IGN BD ORTHO® orthophotos at 20cm resolution:

# Simple command to add RGB colors
ign-lidar-hd patch \
  --input enriched_tiles/ \
  --output patches/ \
  --include-rgb \
  --rgb-cache-dir cache/

📦 Renamed Command: `process` → `patch`

For better clarity, we've renamed the process command to patch. Don't worry - the old command still works for backwards compatibility!

# New recommended command
ign-lidar-hd patch --input tiles/ --output patches/

# Old command (still works, shows deprecation warning)
ign-lidar-hd process --input tiles/ --output patches/

Why RGB Augmentation?

Combine geometric features with photometric information for better classification:

Improved accuracy: Models can learn from both shape and color
Better generalization: Color helps disambiguate similar geometries
Enhanced features: 30+ geometric features + 3 color channels

Better Visualization

Color-coded point clouds make analysis and debugging much easier:

import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D

# Load patch with RGB
data = np.load('patch.npz')
points = data['points']
rgb = data['rgb']  # Normalized [0, 1]

# Visualize with true colors
fig = plt.figure(figsize=(12, 8))
ax = fig.add_subplot(111, projection='3d')
ax.scatter(points[:, 0], points[:, 1], points[:, 2], c=rgb, s=1)
plt.title('RGB-Augmented Point Cloud')
plt.show()

Zero Manual Work

No need to:

Download orthophotos manually
Align images with point clouds
Handle coordinate transformations
Manage multiple data sources

Everything is automatic! The library:

Fetches orthophotos from IGN Géoplateforme WMS
Maps 3D points to 2D pixels automatically
Extracts and normalizes RGB values
Caches downloads for performance

How It Works

The system is smart about caching - orthophotos are downloaded once per tile and reused for all patches, making the process fast and efficient.

Performance

Speed Benchmarks

Configuration	Time per Patch	Notes
Geometry only	0.5-2s	Baseline (no RGB)
RGB (cached)	0.6-2.5s	+0.1-0.5s (minimal impact)
RGB (first download)	2-7s	+2-5s (one-time cost)

With caching enabled (recommended!), the performance impact is minimal:

10-20x faster than downloading every time
~500KB-2MB cache size per tile
Automatic reuse across patches from same tile

Memory Impact

RGB augmentation adds only ~196KB per patch (16,384 points × 3 colors × 4 bytes), which is negligible compared to the geometric features.

Python API

Basic Usage

from pathlib import Path
from ign_lidar import LiDARProcessor

# Initialize with RGB support
processor = LiDARProcessor(
    lod_level="LOD2",
    include_rgb=True,
    rgb_cache_dir=Path("cache/")
)

# Process tiles
patches = processor.process_tile("enriched_tile.laz", "output/")

# Each patch now has RGB!
import numpy as np
data = np.load("output/patch_0001.npz")
print(data.keys())  # ['points', 'features', 'labels', 'rgb', 'metadata']
print(data['rgb'].shape)  # (N, 3) - normalized RGB [0, 1]

Advanced: Direct RGB Augmentation

from ign_lidar.rgb_augmentation import IGNOrthophotoFetcher

# Direct control over RGB fetching
fetcher = IGNOrthophotoFetcher(cache_dir=Path("cache/"))

# Fetch for specific bounding box
bbox = (x_min, y_min, x_max, y_max)  # Lambert 93
image = fetcher.fetch_orthophoto(bbox, tile_id="0123_4567")

# Augment points
import numpy as np
points = np.array([[x1, y1, z1], [x2, y2, z2]])
rgb = fetcher.augment_points_with_rgb(points, bbox, tile_id="0123_4567")

Data Specifications

IGN Service

Source: IGN Géoplateforme WMS
Layer: HR.ORTHOIMAGERY.ORTHOPHOTOS
Resolution: 20cm per pixel
CRS: EPSG:2154 (Lambert 93)
Coverage: Metropolitan France
Format: PNG (24-bit RGB)

Output Format

Each NPZ patch now includes:

{
    'points': np.ndarray,      # (N, 3) - X, Y, Z
    'features': np.ndarray,    # (N, 30+) - Geometric features
    'labels': np.ndarray,      # (N,) - Classification labels
    'rgb': np.ndarray,         # (N, 3) - RGB colors [0, 1]
    'metadata': dict           # Patch info
}

Getting Started

Installation

The RGB feature requires two additional packages:

pip install requests Pillow

# Or install with extras
pip install ign-lidar-hd[rgb]

Quick Start

# 1. Download tiles (as usual)
ign-lidar-hd download --bbox -2.0,47.0,-1.0,48.0 --output tiles/

# 2. Enrich with features (as usual)
ign-lidar-hd enrich --input-dir tiles/ --output enriched/

# 3. Create patches WITH RGB (new!)
ign-lidar-hd patch \
  --input-dir enriched/ \
  --output patches/ \
  --include-rgb \
  --rgb-cache-dir cache/ \
  --lod-level LOD2

Best Practices

1. Always Use Caching

# ✅ Good: With cache (fast)
ign-lidar-hd patch --include-rgb --rgb-cache-dir cache/

# ❌ Slow: No cache (downloads repeatedly)
ign-lidar-hd patch --include-rgb

Caching makes RGB augmentation 10-20x faster for subsequent patches.

2. Check RGB Statistics

import numpy as np

data = np.load('patch.npz')
rgb = data['rgb']

print(f"RGB mean: {rgb.mean(axis=0)}")
print(f"RGB std: {rgb.std(axis=0)}")
print(f"RGB range: [{rgb.min():.3f}, {rgb.max():.3f}]")

# Values should be in [0, 1]
assert rgb.min() >= 0 and rgb.max() <= 1, "RGB not normalized!"

3. Handle Missing Dependencies Gracefully

try:
    from ign_lidar.rgb_augmentation import IGNOrthophotoFetcher
    HAS_RGB = True
except ImportError:
    HAS_RGB = False
    print("Install 'requests' and 'Pillow' for RGB support")

processor = LiDARProcessor(
    include_rgb=HAS_RGB,
    rgb_cache_dir=Path("cache/") if HAS_RGB else None
)

Backwards Compatibility

Everything still works as before!

RGB augmentation is opt-in via --include-rgb flag
Default behavior unchanged (no RGB)
Old process command still works (with deprecation warning)
Existing scripts continue to function without modifications

Migration Guide

Command Line

# Old workflow (still works)
ign-lidar-hd process --input tiles/ --output patches/

# New workflow (recommended)
ign-lidar-hd patch --input tiles/ --output patches/

# New workflow with RGB
ign-lidar-hd patch --input tiles/ --output patches/ --include-rgb

Python API

# Old code (still works)
processor = LiDARProcessor(lod_level="LOD2")
patches = processor.process_tile("tile.laz", "output/")

# New with RGB (opt-in)
processor = LiDARProcessor(
    lod_level="LOD2",
    include_rgb=True,  # NEW!
    rgb_cache_dir=Path("cache/")  # NEW!
)
patches = processor.process_tile("tile.laz", "output/")

Use Cases

1. Building Material Classification

Use color to distinguish building materials:

Brick walls: Red/brown tones
Concrete: Gray tones
Glass windows: Reflective/variable colors
Vegetation: Green tones

2. Quality Control

Identify misalignments or processing errors by checking if colors match expected semantics:

data = np.load('patch.npz')
rgb = data['rgb']
labels = data['labels']

# Roofs should be dark (tiles) or light (metal)
roof_colors = rgb[labels == ROOF_CLASS]
print(f"Roof colors: R={roof_colors[:, 0].mean():.2f}")

Train models on both geometry and photometry:

import torch
import torch.nn as nn

class MultiModalNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.geom_branch = nn.Linear(30, 64)
        self.rgb_branch = nn.Linear(3, 16)
        self.classifier = nn.Linear(80, 15)

    def forward(self, geometry, rgb):
        g = self.geom_branch(geometry)
        r = self.rgb_branch(rgb)
        combined = torch.cat([g, r], dim=-1)
        return self.classifier(combined)

Documentation

Full documentation available:

RGB Augmentation Guide - Complete guide with examples
CLI Commands Reference - Updated command documentation
API Reference - Python API details

What's Next?

We're continuously improving the library. Upcoming features:

🔧 Support for custom orthophoto sources
📊 RGB-based feature engineering
🎯 Pre-trained multi-modal models
🌍 Support for other WMS services

Get Involved

Try It Now!

# Install/upgrade
pip install --upgrade ign-lidar-hd

# Install RGB dependencies
pip install requests Pillow

# Start using RGB augmentation
ign-lidar-hd patch \
  --input your_tiles/ \
  --output patches/ \
  --include-rgb \
  --rgb-cache-dir cache/

Happy processing! 🚀✨

What's New?​

🎨 RGB Augmentation​

📦 Renamed Command: process → patch​

Why RGB Augmentation?​

Multi-Modal Machine Learning​

Better Visualization​

Zero Manual Work​

How It Works​

Performance​

Speed Benchmarks​

Memory Impact​

Python API​

Basic Usage​

Advanced: Direct RGB Augmentation​

Data Specifications​

IGN Service​

Output Format​

Getting Started​

Installation​

Quick Start​

Best Practices​

1. Always Use Caching​

2. Check RGB Statistics​

3. Handle Missing Dependencies Gracefully​

Backwards Compatibility​

Migration Guide​

Command Line​

Python API​

Use Cases​

1. Building Material Classification​

2. Quality Control​

3. Multi-Modal Deep Learning​

Documentation​

What's Next?​

Get Involved​

Try It Now!​

What's New?

🎨 RGB Augmentation

📦 Renamed Command: `process` → `patch`

Why RGB Augmentation?

Multi-Modal Machine Learning

Better Visualization

Zero Manual Work

How It Works

Performance

Speed Benchmarks

Memory Impact

Python API

Basic Usage

Advanced: Direct RGB Augmentation

Data Specifications

IGN Service

Output Format

Getting Started

Installation

Quick Start

Best Practices

1. Always Use Caching

2. Check RGB Statistics

3. Handle Missing Dependencies Gracefully

Backwards Compatibility

Migration Guide

Command Line

Python API

Use Cases

1. Building Material Classification

2. Quality Control

3. Multi-Modal Deep Learning

Documentation

What's Next?

Get Involved

Try It Now!