Detecting grasping positions with deep neural networks using RGB images
-
Use grasp_detection for training model. This is for test. grasp_detection Link : https://github.com/Juna2/grasp_detection
-
Most of the files in this repo is from his work except test.py
-
test.py loads pretrained model in
/home/<your_path>/catkin_ws/src/ros_grasp_detection/src/m4
-
you may change the path in test.py
-
I recommand you to train model with Juna2/grasp_detection(which is also modified version of robot-grasp-detection) first and copy the model to the path.
How to use
-
$ catkin_make
-
$ roslaunch ros_grasp_detection
-
It subscribes "/croppedRoI" message which is 224x224 object image and publishes "/objects" which contains center x, y and degree. So you have to use another package to publish 224x224 object image to this package.
-
The grasp detection result will be saved in
/home/<your_path>/catkin_ws/src/ros_grasp_detection/src/image/bbox.jpg
- One Thing you should know is that this test.py uses the model trained by modified robot-grasp-detection which is Juna2/grasp_detection.