Depth Anything 3 ROS2 Wrapper

A ROS2 wrapper for Depth Anything 3, enabling monocular depth estimation in ROS2 environments.

Features

• Real-time monocular depth estimation from RGB images
• Support for multiple Depth Anything 3 model variants (Small, Base, Large, Giant)
• GPU acceleration with CUDA support
• Configurable ROS2 topics and parameters
• Easy integration with existing ROS2 camera pipelines

Requirements

• ROS2 (tested with Humble and later)
• Python >= 3.10
• PyTorch >= 2.0
• CUDA (optional, for GPU acceleration)

Installation

1. Install Dependencies

# Install PyTorch (adjust for your CUDA version)
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu118

# Install other Python dependencies
pip3 install -r requirements.txt

2. Install Depth Anything 3

pip3 install git+https://github.com/ByteDance-Seed/Depth-Anything-3.git

3. Build the ROS2 Package

# Navigate to your ROS2 workspace
cd ~/ros2_ws/src

# Clone this repository
git clone https://github.com/spectateursimon/Depth-Anything-3-ROS2.git

# Build the package
cd ~/ros2_ws
colcon build --packages-select depth_anything_3_ros2

# Source the workspace
source install/setup.bash

Usage

Basic Launch

Launch the node with default parameters:

ros2 launch depth_anything_3_ros2 default.launch.py

Custom Parameters

You can override parameters at launch:

ros2 launch depth_anything_3_ros2 default.launch.py \
  image_topic:=/your/camera/image_raw \
  depth_image_topic:=/your/depth/output \
  device:=cuda:0 \
  model_name:=depth-anything/DA3-Large

Run Node Directly

ros2 run depth_anything_3_ros2 depth_anything_node

Configuration

Parameters

Parameter	Type	Default	Description
image_topic	string	/camera/color/image_raw	Input RGB image topic
depth_image_topic	string	/depth	Output depth image topic
device	string	cuda:0	Device for inference (cuda:0, cuda:1, cpu)
model_name	string	depth-anything/DA3-Large	Model variant to use

Available Models

The following Depth Anything 3 models are available:

• depth-anything/DA3-Small (80M parameters) - Fastest
• depth-anything/DA3-Base (120M parameters) - Balanced
• depth-anything/DA3-Large (350M parameters) - High quality (default)
• depth-anything/DA3-Giant (1.15B parameters) - Best quality
• depth-anything/DA3Metric-Large - Metric depth estimation

Configuration File

Edit config/default.yaml to set default parameters:

depth_anything_3:
  ros__parameters:
    image_topic: "/camera/color/image_raw"
    depth_image_topic: "/depth"
    device: "cuda:0"
    model_name: "depth-anything/DA3-Large"

Topics

Subscribed Topics

• image_topic (sensor_msgs/Image): Input RGB image stream

Published Topics

• depth_image_topic (sensor_msgs/Image): Output depth image (16-bit grayscale)

Example with RealSense Camera

# Terminal 1: Start RealSense camera
ros2 launch realsense2_camera rs_launch.py

# Terminal 2: Start Depth Anything 3
ros2 launch depth_anything_3_ros2 default.launch.py \
  image_topic:=/camera/camera/color/image_raw

Visualization

View the depth output using RViz2:

rviz2

Add an Image display and set the topic to /depth (or your configured output topic).

Alternatively, use rqt_image_view:

ros2 run rqt_image_view rqt_image_view

Performance

Performance varies by model size and hardware:

Model	Parameters	GPU Memory	FPS (RTX 3090)
DA3-Small	80M	~2GB	~45 FPS
DA3-Base	120M	~3GB	~35 FPS
DA3-Large	350M	~5GB	~25 FPS
DA3-Giant	1.15B	~12GB	~10 FPS

Note: Performance may vary based on input resolution and hardware configuration.

Troubleshooting

CUDA Out of Memory

If you encounter CUDA out of memory errors, try:

• Using a smaller model (DA3-Small or DA3-Base)
• Reducing input image resolution
• Using CPU inference (slower): device:=cpu

Model Download Issues

Models are automatically downloaded from Hugging Face Hub on first run. If download fails:

• Check your internet connection
• Ensure you have sufficient disk space (~5GB per model)
• Try running with HF_HUB_OFFLINE=0 environment variable

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

The original Depth Anything 3 is also licensed under Apache 2.0.

Acknowledgments

• Depth Anything 3 by ByteDance
• ROS2 community

Citation

If you use this work in your research, please cite the original Depth Anything 3 paper:

@article{depthanything3,
  title={Depth Anything 3: Toward a Unified Framework for Monocular Depth Estimation},
  author={Yang, Lihe and others},
  journal={arXiv preprint},
  year={2024}
}

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Contact

For issues and questions, please open an issue on GitHub.