ros-node-object-recognition

ROS package for getting output from pre-trained caffe models on NVIDIA Digits.

To use the classification module, download your model from Digits server and extract it. Note that, some strings must be set properly before compilation.

Dependencies

• [OpenCV]
• [Robot Operating System]

Installation

Put it in your src file of your workspace and compile it with catkin_make.
The codes are successfully compiled in Ubuntu 12.04 + ROS Hydro by using Catkin.

Nodes

The package contains two nodes:

• classification node: Using python scripts written by NVIDIA to get the output trained with NVIDIA Digits
• clustering node: Clusters the images according to their 2D Hue-Saturation Histogram. If you want to use this, your dataset must include a “others” label meaning that module cannot label the image.

Topics

• “apes_robot/fixation_point”: std_msgs::String - The fullpath and name of image that will be labeled
• “object_recognition/label”: std_msgs::String - Top 1 label of current input image example: “/home/awesomeuser/input.jpg,table”

License

MIT