BagX

Bag Exchange for conversion between ROS Bag formats and mp4, jpg, png, csv, etc.

Requirements

• NVIDIA GPU
• x86_64 machine (recommended)

Docker

Grant permissions to all the scripts:

sudo chmod -R a+wrx ./scripts

To run (automatically pulls if missing) & attach:

./scripts/run.sh
docker attach BagX

To pull:

./scripts/pull.sh

To build:

./scripts/build.sh

Conda

If you have NVIDIA DeepStream and ROS setup locally, you can try the conda environment requirement.yaml

conda env create -f environment.yml
conda activate bagx

Usage

1. modify the DATASET_PATH in ./scripts/variables.sh according to your data directory on your host machine

   • DATASET_PATH will be mounted to ~/data inside the docker container
   • Remember to run ./scripts/run.sh again after modifying the DATASET_PATH to update the docker container mount points

2. modify the config according to your needs, e.g., ./config/example/mp4_to_imgdir_rosbag.yaml

   • Any invalid input file path will cause assertion failure and stop the program, so don't worry, it won't go crazy
   • Remember to specify all the pipelines you want to run in the pipeline list, otherwise it won't run

3. To run (inside the docker container):

./conversion.py -c < path to your config file >

How it works

• The library currently only supports conversion from mp4 to rosbag/directory of images, but the coding structure is general enough to add other input/output later, e.g., rosbag input, csv output(for IMU, point cloud, etc.)
• The library can concurrently convert multiple files, i.e. multiple pipelines, using Threading
• mp4 reading is using GStreamer, with NVIDIA DeepStream plugin to enable nvvideoconvert pipeline for Nvidia's hardware-accelerated decoding on NVDEC
• Please follow the example config for config file format, and follow the ./config folder structure if you need to push your config
• Should you run into any trouble, please try the Jupyter Notebook

Contacts

• Author: Mukai (Tom Notch) Yu: [email protected]

How to contribute

1. Setup your development environment

   sudo ./scripts/setup.sh

   This will install pre-commit hooks and its dependencies locally, so that each time before you commit, the code will be formatted and linted automatically. Remember to git add . after git commit failed since pre-commit will modify source code in-place.

2. Check the GitHub webpage after a few minutes to see if the CI passed. Passing CI will have a green check mark on the commit

   • If not, please fix the errors and push again

• Add more input/output classes that can connect to more formats
• append input/output types in macros.py and initializations in pipeline.py