ROS wrapper for PoseNET library in NodeJS PoseNet demo: Here
- Current implementation is for experiments only *
- ROS Kinetic
- NodeJS 8.x
- Cuda 9.0 + cuDNN 7.1 (for GPU acceleration only)
- Clone to ROS workspace and build
npm install
inside the package folder- if you use this program offline, you edit must their files.
- 1: '..node_modules/@tensorflow-models/posenet/dist/checkpoints.js'
- 2: '..node_modules/@tensorflow-models/posenet/dist/posenet.js'
Following ROS parameters should be set:
gpu: (true / false)
- Specifies if GPU acceleration should be usedtopic
- Uncompressed RGB8 encoded image topicout_topic
- specifies topic for result output. Output topic publishes JSON string that needs to be decoded asstd_msgs/String
message- Algorithm parameters to adjust performance. See launch file for full list> References could be found PoseNet Official Github
bash start_posenet.sh
- launch ros_kinect
- Only multiple pose detection implemented
- Requires internet to download the model weights
- Only ROS tpopics with RGB8 encoding are supported as inputs