FastSAM ROS with Docker

This ROS package wraps the Official FastSAM repo. Further, nvidia-docker is used to build an image with all the necessary dependencies. Then, docker compose can be used to create a container to launch a ROS node. Note that the docker-compose.yml file shares the container network with the host network---thus, the node will be able to communicate with other ROS nodes on the host machine.

Tested on:

• Docker version 23.0.0, build e92dd87
• NVIDIA Container Runtime Hook version 1.12.0, commit: 62bd015475656ef795cb0d59cc4030a6bd4a9526
• NVIDIA GTX 1070 8GB
• Ubuntu 20.04 / ROS Noetic

Getting Started

Install docker, docker-compose, nvidia-docker (i.e., nvidia-container-toolkit). Note that you do not need CUDA installed on your host machine, but you do need an NVIDIA driver installed on the host for your particular NVIDIA GPU.

$ git clone --recurse-submodules https://github.com/mit-acl/fastsam_ros # get FastSAM as submodule
$ cd fastsam_ros
$ docker compose build
$ docker compose up

Running inside a namespace

You can use docker compose run to set the ROS_NAMESPACE environment variable:

docker compose run -e ROS_NAMESPACE=robot/camera fastsam_ros roslaunch fastsam_ros detect.launch

Webcam demo

1. Install your favorite ROS web camera package. We will use usb_cam: sudo apt install ros-noetic-usb-cam.
2. Start a roscore.
3. Start the USB camera (with remapping): rosrun usb_cam usb_cam_node usb_cam/image_raw:=image_raw
4. Start fastsam_ros via docker compose up

Credits

• FastSAM
• Meta's Segment Anything
• YOLOv7 ROS