RangeNet++ Docker

Docker container with all dependencies to be able to test the inference of RangeNet++ with C++ and TensorRT. This container have the following code inside https://github.com/PRBonn/lidar-bonnetal and https://github.com/mgrova/rangenet_lib

Dependencies

• Docker

• Nvidia-Docker

Required files

• TensorRT 5.1 must be downloaded from the official NVIDIA website. The appropriate file is: nv-tensorrt-repo-ubuntu1804-cuda10.1-trt5.1.5.0-ga-20190427_1-1_amd64.deb. This file must be placed in this directory, along with the Dockerfile.

• To run the demo, you need a pre-trained model, which can be downloaded here, model. As the previous file, place the unzipped folder in the shared_folder directory.

Usage

---

For simplicity of use, the scripts/run.sh script is used to install the necessary dependencies, create the image and run the container. Note that this container will run in detached mode so that it can remain running in the background on a remote server. That is why, once the container is running, the following command will be used to open a new terminal:

docker ps -a #To ensute thar container still alive
docker exec -it rangenet /bin/bash

If you want to use the ROS version, use the following script: scripts/run_ros.sh.

Working inside the container

Test inference library (without ROS version)

Inside the docker container you must run the following command to test the inference:

./rangenet_lib/build/examples/simple_infer -p /home/user/shared_folder/darknet53 -s /home/user/rangenet_lib/examples/simple_infer/000000.bin --verbose

The result should look like the one shown in the following image:

Test training library

To test the training pipeline, it is necessary to download the Semantic KITTI dataset and labels. Once downloaded and placed in the working directory, the following commands must be executed inside the container:

cd /home/user/lidar-bonnetal/train/tasks/semantic 
./visualize.py -d /home/user/shared_folder/datasets/semantic_kitti/dataset/ -s 00

Training the model

To run the training, the use of tmux is recommended. By using a session, you can leave the training running in the background. To detach the session and leave it running in background, press the following key combination: ctrl b + d.

tmux attach || tmux new -s rangenet_train
cd /home/user/lidar-bonnetal/train/tasks/semantic
./train.py -d /home/user/shared_folder/datasets/semantic_kitti/dataset/ -ac /home/user/shared_folder/config_files/semantic_kitti_arch/darknet21.yaml -l /home/user/shared_folder/logs/

To re-attach the session, the following command shall be used:

tmux list-session # To ensure that session still alive
tmux attach-session rangenet_train

If you want to display the results in tensorboard, you must run the following command in a new terminal inside the docker container:

python3 -m tensorboard.main --logdir /home/user/shared_folder/logs --host 0.0.0.0