[Documentation] [TitleIndex] [WordIndex

Package Summary

The cnn_bridge package take freeze graphs and publishes them as ROS messages

  • Maintainer status: developed
  • Maintainer: Noam C. Golombek <ngolombek AT gmail DOT com>
  • Author: Noam C. Golombek <ngolombek AT gmail DOT com>, Alexander Beringolts <berengolts AT gmail DOT com>
  • License: BSD
  • Source: git https://github.com/wew84/cnn_bridge.git (branch: 0.8.4)

Package Summary

The cnn_bridge package take freeze graphs and publishes them as ROS messages

  • Maintainer status: developed
  • Maintainer: Noam C. Golombek <ngolombek AT gmail DOT com>
  • Author: Noam C. Golombek <ngolombek AT gmail DOT com>, Alexander Beringolts <berengolts AT gmail DOT com>
  • License: BSD
  • Source: git https://github.com/wew84/cnn_bridge.git (branch: 0.8.4)

Overview

This package provides support for parsing convolution neural networks (CNN), and publishing them as ROS messages. Currently the package supports both detection an segmentation networks.

Input can be either from camera topics, an OpenCV camera, or a video.

ROS Nodes

Launch File Examples

Start a cnn_bridge in segmentation mode:

$ roslaunch cnn_bridge segmentation_publisher.launch

Start a cnn_bridge in detection mode:

$ roslaunch cnn_bridge detection_publisher.launch

Hypes Example

{
    "frozen_graph_path": "<path_to_frozen_graph.pb>",
    "image_height": 361,
    "image_width": 641,
    "resize_image": true,
    //TODO Add additional fields
}

Metadata JSON

If segmentation mode:

{
    "classes": ["CLASS_NAME_1", "CLASS_NAME_2", ..., "CLASS_NAME_N"]
}

If detection mode:

{
    "classes": [{
        "color": (red, green, blue),
        "id": < The ID of the class as outputted from the network >,
        "name": < Name assigned to the class >
        "id_category": < The ID of a parent class (ie. If class dog parent Animal) >
        "category": < The name of a parent class (ie. If class dog parent Animal) >
    }]
}


2019-10-12 12:30