/depth_anything_ros

ROS1 wrapper package of Depth Anything V2

Primary LanguageC++

depth_anything_ros

ROS1 wrapper for Depth Anything V2.

Creating Pointcloud without depth

Alt text

Setup

Prerequisite

This package is build upon

  • ROS1 (Noetic)
  • cuda (11.8+)
  • TensorRT(8+)
  • (Optional) docker and nvidia-container-toolkit (for environment safety)

Build package

on your workspace

mkdir -p ~/ros/catkin_ws/src && cd ~/ros/catkin_ws/src
git clone https://github.com/ojh6404/depth_anything_ros.git
cd ~/ros/catkin_ws/src/depth_anything_ros && git submodule --init --recursive
cd ~/ros/catkin_ws && TENSORRT_DIR='/path/to/your/tensorrt_dir' catkin build
cd ~/ros/catkin_ws/src/depth_anything_ros && pip install -r requierments.txt
cd ~/ros/catkin_ws/src/depth_anything_ros && ./scripts/onnx2tensorrt.sh trained_data

using docker (Recommended)

git clone https://github.com/ojh6404/depth_anything_ros.git
cd depth_anything_ros && docker build -t depth_anything_ros .

How to use

roslaunch depth_anything_ros.launch depth_estimation.launch input_image:=/your/image/topic camera_info:=/your/camera_info/topic

Please refer depth_estimation.launch to see args.

In case you are using docker,

docker run --rm --net=host -it --gpus 1  depth_anything_ros:latest /bin/bash

then inside docker,

roscd depth_anything_ros
./scripts/onnx2tensorrt trained_data # convert onnx to tensorrt
roslaunch depth_anything_ros.launch depth_estimation.launch input_image:=/your/image/topic camera_info:=/your/camera_info/topic

For just test with provided rosbag, run

roslaunch depth_anything_ros test.launch

Acknowledgement