CNN LiDAR Baidu Object Segmenter

Autoware package based on Baidu’s object segmenter.

Pre requisites

Caffe distributable installed in your home (~/caffe/distribute).

$ cd
$ git clone https://github.com/BVLC/caffe
$ cd caffe

Follow instructions from Installing Caffe from source.

• Use offical Make compilation procedure.
• Do not use thirdparty CMake setup.

Compile and create distributable:

$ make
$ make distribute

Recompile Autoware to build the node.

The Pretrained model

Use this link to download the pretrained model from Baidu:

https://github.com/ApolloAuto/apollo/tree/v5.5.0/modules/perception/production/data/perception/lidar/models/cnnseg

These two files are needed:

• deploy.prototxt
• deploy.caffemodel

How to launch

• From a sourced terminal:

Using rosrun: rosrun lidar_apollo_cnn_seg_detect lidar_apollo_cnn_seg_detect _network_definition_file:=/PATH/TO/FILE.prototxt _pretrained_model_file:=/PATH/TO/WEIGHTS.caffemodel _points_src:=/points_raw

Using launch file: roslaunch lidar_apollo_cnn_seg_detect lidar_apollo_cnn_seg_detect.launch network_definition_file:=/PATH/TO/FILE.prototxt pretrained_model_file:=/PATH/TO/WEIGHTS.caffemodel points_src:=/points_raw

• From Runtime Manager:

Computing Tab -> Detection/ lidar_detector -> lidar_cnn_baidu_detect. Configure parameters using the [app] button.

Parameters

Outputs

Notes

• To display the results in Rviz objects_visualizer is required. (Launch file launches automatically this node).

• Pre trained models can be downloaded from the Apollo project repository.

Changelog for package lidar_apollo_cnn_seg_detect

1.11.0 (2019-03-21)

• [feature] Baidu's CNN based LiDAR segmentation (#1800)
  • Add build caffe
  • add include apollo files
  • add apollo cnn
  • calculating time
  • • Cleaned node
  • Parametrized inputs
  • Works with custom caffe
  • • Parameterized
  • Works with custom Caffe
  • Removed hard coded params
  • Cleaned up dependencies
  • Added bboxes, labels
  • Minor fixes
  • Custom input topic
  • Added UI, launch file, readme
  • Added Compatibility for Perception Cleanup
  • • Added license messages
  • Updated readme
  • Added extra instructions
  • Fix markdown
• Contributors: Abraham Monrroy Cano