A Cython-based Dynamic Vision Sensor (DVS) emulator. It can be used within a Python environment as demonstrated in the various test files.
This project is under the GPLv3 License (https://www.gnu.org/copyleft/gpl.html)
A DVS is a camera-like sensor which emits per-pixel asynchronous Address-Event Representation (AER) packets. These are originated due to an intensity level change has reached a certain threshold.
This project emulates this behaviour using commodity hardware, that is a regular PC and camera. Modifications to the default behaviour where added, mainly: encoding of values using time, local inhibition and dynamic thresholds.
Fred Rotbart (https://github.com/fcr) has been kind enough to port to Python 3 and checked compatibility with MacOS Sierra, plus code clean-up!
Requirements:
- Cython
- Numpy
- OpenCV (with Python bindings)
Before using, compile the Cython modules by running
python setup.py build_ext --inplace
in the folder where pyDVS is installed.
Files:
-
pydvs/:generate_spikes.pyx- Cython backend library. Contains functions to generate spikes from a video feed.external_dvs_emulator_device*.py- sPyNNaker class wrapper, it's compatible with OpenCV VideoCapture devices and VirtualCam.virtual_cam.py- A camera emulator that feeds on images located in the users computer. It keeps images moving based on eye movements (e.g. micro saccades and saccades); there's a very simple attention model for the saccade mode.
-
spynnaker_test.py- Demonstration script for sPyNNaker compatible frontend. -
stand_alone*.py- Demonstration scripts for the Cython backend, they come in single- and multi-threaded versions. -
virtual_cam_test.py- Demonstration script for the VirtualCam class, it requires images in a directory ("./mnist" by default)
The multithreaded versions require either Python 3.4 and above or Windows