An Efficient Spatial Convolution Implementation for Event Cameras

Installation

The Makefile file contains commands to download the necessary libraries and dependencies.

This project has been built on top of YARP and Event Driven libraries Robotology

Tested on Ubuntu 20.04

1. Download the libraries repositores
   • The makefile will download and install the libraries at ~/libraries folder.
   • You can set the path for the libraries' code and installation by modifying PROJECT_DIR and INSTALL_DIR at lines 1 and 2 of Makefile.

make sources

2. Compile the libraries

make

3. Attention: dependencies might occur depending on your system

   • Try installing one library at a time in order

   make install-YCM
   make install-yarp
   make install-event-driven

   • Some common dependencies can be installed with

     make dependencies

   • refer to YARP dependencies and Event Driven How to install for more info.

4. Install the convolutions project

make install-convolution

Running

0. Add the executables to your path (or in your .<shell>rc file)

export PATH=$PATH:/home/leandro/libraries/install/bin

1. Start a yarpserver

yarp server

1. Using yarpmanager is a convenient way to manage yarp modules. To start it:

yarpmanager

2. We coded an application with the convolution modules

   1. On yarp manager, go to file -> open file -> ~/libraries/install/share/event-driven/applications -> convolutions.xml
   2. You can launch the reference implementation refConv and our implementation liteConv
   3. Depending on your camera, you might want to use a spatio-temporal filter vPreProcess

3. If you are using a camera, set the connection directly on the connection panel on yarpmanager

4. If you are using recorded data, you can start a yarpdataplayer to stream the data to the camera:

yarpdataplayer

• Go to file->Open Directory, navigate ans select the folder generate And then set the connection on the connection panel on yarpmanager

• The datasets used in the paper are available in a raw text format at High Speed and HDR Datasets

• The conversion to yarp format can be done with Binvee Library

• An example script can be found in the src/python root folder of the sequence (should look like <sequence name>_converted.

Logging Results

1. On the Makefile, you can edit the LOG=<v> flag. <v=0,1,2> disables logging, logs computation time data, and logs accuracy data, respectively

Converting the HDR Dataset

1. Download and extract the HDR datasets

2. We use bimvee for converting the .txt HDR dataset file into yarp format

pip install bimvee
export DATASETS_PAT=<datasets location>
python src/python/gen3SequencesToYarp.py <sequence_name>

• <sequence_name> without the .txt extension
• <datasets location> is the folder in which you downloaded and extracted the datasets
• E.g.: python src/python/gen3SequencesToYarp.py hdr_sun