Haha777/xtreme1-pointcloud-label-ready

Xtreme1 Point Cloud Dataset Processor

This repository contains a modular pipeline for processing raw point cloud datasets into xtreme1-ready format.

Overview

The pipeline transforms raw datasets containing RGB images, depth maps, and metadata into a structured format suitable for xtreme1 point cloud labeling tool.

👉 See QUICKSTART.md for step-by-step guide 👉 See CLI_REFERENCE.md for complete command reference

Raw Dataset Structure

raw/<dataset_name>/
├── depth_results/           # REQUIRED: Metadata JSON files with intrinsics and depth data
│   ├── 000000000139_metadata.json
│   ├── 000000000139_depth.npy
│   └── ...
├── camera_image_0/          # REQUIRED: RGB camera images for coloring point clouds
│   ├── 000000000139.jpg
│   └── ...
└── lidar_point_cloud_0/     # OPTIONAL: Original point clouds (new ones generated if absent)
    ├── 000000000139.pcd
    └── ...

Processed Dataset Structure

processed/<dataset_name>/
├── camera_config/           # Camera configuration files for xtreme1
│   ├── 000000000139.json
│   └── ...
├── camera_image_0/          # RGB camera images (copied)
│   ├── 000000000139.jpg
│   └── ...
└── lidar_point_cloud_0/     # Processed and rotated colored point clouds
    ├── 000000000139.pcd
    └── ...

Installation

1. Create Virtual Environment (Recommended)

cd /path/to/xtreme1-pointcloud-label-ready

# Create virtual environment
python3 -m venv venv

# Activate it (macOS/Linux)
source venv/bin/activate

# On Windows:
# venv\Scripts\activate

2. Install Dependencies

# Install from requirements.txt (recommended)
pip install -r requirements.txt

# Or install manually
pip install numpy pillow open3d

Required packages:

• numpy - Numerical operations
• pillow - Image processing
• open3d - Point cloud processing and visualization

3. Verify Installation

python -c "import open3d; import numpy; import PIL; print('✅ All dependencies installed!')"

Usage

Basic Command

python scripts/process_dataset.py raw/your_dataset

The pipeline will:

1. Generate colored point clouds from RGB and depth data
2. Rotate point clouds to top-down view
3. Create camera configuration files for xtreme1
4. Copy camera images
5. Create a zip archive

See QUICKSTART.md for detailed instructions and CLI_REFERENCE.md for all command options.

Modules

color_pcd.py

• Generates colored point clouds from RGB images and depth maps
• Can add RGB data to existing PCDs or create new ones from scratch
• Uses camera intrinsics from metadata JSON files

rotate_pcd.py

• Rotates point clouds by -90 degrees around the X-axis
• Converts upside-down point clouds to top-down view
• Essential for proper visualization in xtreme1

create_camera_config.py

• Generates xtreme1-compatible camera configuration files
• Extracts camera intrinsics and image dimensions from metadata
• Applies appropriate extrinsic transformations

process_dataset.py

• Main pipeline orchestrator
• Coordinates all processing steps
• Provides clean CLI interface with progress reporting
• Handles temporary files and cleanup

Dataset Naming Convention

When no custom name is provided, datasets are automatically named:

<raw_dataset_name>_processed_<timestamp>

Example: test_dataset_v0_processed_20251028_143022

Output Format

The processed dataset follows the xtreme1 expected structure and can be directly imported into the xtreme1 point cloud labeling tool as a zip file.

Additional Documentation

• QUICKSTART.md - Step-by-step usage guide with examples
• CLI_REFERENCE.md - Complete command-line reference

License

Apache 2.0