This is a ROS package of Mask R-CNN algorithm for object detection and segmentation.

The package contains ROS node of Mask R-CNN with topic-based ROS interface.

Most of core algorithm code was based on Mask R-CNN implementation by Matterport, Inc.

This repository doesn't contain code for training Mask R-CNN network model. If you want to train the model on youer own class definition or dataset, try it on the upstream reposity and give the result weight to model_path parameter.

• ROS kinetic
• TensorFlow 1.3+
• Keras 2.0.8+
• Numpy, skimage, scipy, Pillow, cython, h5py

• ~model_path: string

  Path to the HDF5 model file. If the model_path is default value and the file doesn't exist, the node automatically downloads the file.

  Default: $ROS_HOME/mask_rcnn_coco.h5

• ~visualization: bool

  If true, the node publish visualized images to ~visualization topic. Default: true

• ~class_names: string[]

  Class names to be treated as detection targets. Default: All MS COCO classes.

• ~result: mask_rcnn_ros/Result

  Result of detection. See also Result.msg for detailed description.

• ~visualization: sensor_mgs/Image

  Visualized result over an input image.

• ~input: sensor_msgs/Image

  Input image to be proccessed

1. Clone this repository to your catkin workspace
2. Build workspace and source devel environment
3. Run mask_rcnn node

   $ rosrun mask_rcnn_ros mask_rcnn_node

There is a simple example launch file using RGB-D SLAM Dataset.

$ cd mask_rcnn_ros/examples
$ ./download_example_bag.sh
$ roslaunch example.launch

Then RViz window will appear and show result like following: