Update: The SNN architecture presented here is reimplemented in a cleaner and more efficient way in hybrid-stereo-matching. Although the latter repository is dedicated to a slightly different task, the SNN is essentially the same and all experiments from [1] can be reproduced using suitable configuration files. For more details on how to do this, see the subsection SNN only in Running custom experiments described in the README there.
This repository contains the original code and data needed to reproduce the experiments from my Bachelor thesis on spiking neural network (SNN) for real-time event-based stereo matching using dynamic vision sensors [4] and the SpiNNaker [3] neuromorphic hardware. The design is inspired from the guidelines for a cooperative network presented in [2] and is thoroughly described in [1]. The implementation uses the sPyNNaker tool chain which provides a PyNN-like API. The network has been tested both on a local SpiNNaker machine and on the HBP platform.
In order to reproduce an experiment just select the desired in main.py. If you are interested in running the network on
your own dataset, then see in examples one of the existing experiments and add your own custom_experiment.py
accordingly. The input and output data is normally stored under data/input and spikes respectively.
In addition to the evaluation presented in [1], you can find animations with the network output at https://figshare.com/s/0d9fb146149b832ed8ec (thanks to Christoph Richter).
Many thanks to Christoph Richter for the multiple fruitful discussions and invaluable support throughout my thesis.
[1] Dikov, G., Firouzi, M., Roehrbein, F., Conradt, J., Richter, C.: Spiking cooperative stereo-matching at 2 ms latency with neuromorphic hardware. Proc. Biomimetic and Biohybrid Systems, 119-137 (2017)
[2] Marr, D., Poggio, T.: Cooperative computation of stereo disparity. Science 194(4262), 283–287 (1976)
[3] Furber, S.B., Galluppi, F., Temple, S., Plana, L.A.: The SpiNNaker project. Proc. IEEE 102(5), 652–665 (2014)
[4] Lichtsteiner, P., Posch, C., Delbruck, T.: A 128 × 128 120 db 15 μs latency asynchronous temporal contrast vision sensor. IEEE J. Solid State Circ. 43(2), 566–576 (2008)