events_h52bag

C++ code to convert event data from HDF5 to ROSbags. By default, data format of h5 input file is expected as in the DSEC dataset or TUM-VIE dataset. The exact data structure used can be found here.

Author: Suman Ghosh

Installation

• Install ROS. We need the rosbag and std_msgs packages.
• Install libhdf5:

sudo apt-get update
sudo apt-get install libhdf5-dev

• Install the blosc filter plugin for hdf5 for read-only access across all HDF5 applications in your system:

git clone https://github.com/Blosc/hdf5-blosc.git
cd hdf5-blosc
mkdir build && cd build
cmake ..
make

• Copy the libH5Zblosc.so shared library generated inside the build folder into your local hdf5 plugin path, which is usually /usr/local/hdf5/lib/plugin or /usr/lib/x86_64-linux-gnu/hdf5/plugins depending on your Linux distro. Installation location of hdf5 can be found using dpkg -L libhdf5-dev. Create the plugins directory if it does not exist already in your system.

Then, add the following line to your ~/.bashrc and source it

export HDF5_PLUGIN_PATH=<path to plugins>

• Optional: For M3ED dataset, lzf filter plugin needs to be built as a shared library. Then copy the library file liblzf_filter.so into the same HDF5 plugin path, as was done for the blosc filter above.

git clone git@github.com:h5py/h5py.git
cd h5py/lzf
h5fc -I"lzf" -O2 -fPIC -shared lzf/*.c lzf_filter.c -lhdf5 -o liblzf_filter.so
sudo cp liblzf_filter.so /usr/lib/x86_64-linux-gnu/hdf5/plugins/

• Add path to your installed dvs_msgs header in line 11 of CMakeLists.txt. If not already installed, install dvs_msgs from https://github.com/uzh-rpg/rpg_dvs_ros in your catkin workspace.
• Clone this repository: git clone https://github.com/tub-rip/events_h52bag.git
• Compile this repository:

cd events_h52bag
mkdir build && cd build
cmake ..
make

Execution

./events_h52bag <path/to/input/h5file> <path/to/output/bagfile(without extension)> <output_topic_name> <sensor_height> <sensor_width> [optional, default=500000000]<events_read_per_iteration> [optional, default=100000]<event_packet_size> [optional, default="events"]<dataset_prefix>

To overcome memory limitations for large data files, h5 files are read in multiple iterations (500M events per iteration max). The result is a single ROSBag.

Working example

• Download input data from DSEC dataset: https://download.ifi.uzh.ch/rpg/DSEC/train/thun_00_a/thun_00_a_events_left.zip
• Extract events.h5 from zip.
• Execute: ./events_h52bag events.h5 out /dvs/left/events 480 640
• The output ROSBag out.bag is generated.

Log

• 13.08.24: Changed behavior. A single ROSBag output is produced.
• 16.05.23: Now supports M3ED dataset. Dataset prefix can now be passed as an argument, which is prophesee/left and prophesee/right for M3ED h5 files.
• 14.09.21: Sensor size can now be passed an argument. Set to 0 when no t_offset data is present in h5 file. Tested on TUM-VIE dataset.
• 07.02.21: Optional argument events_per_bag can be used to generate multiple rosbags while converting from a single h5 file. Optional argument event_packet_size can be passed to modify ROS message rate.