In order to familiarise myself with event-based cameras I implemented [1], in which "time surfaces" are generated in real-time from event camera feeds. Digits are classified by building time surfaces of each digit.
[1] Lagorce X, Orchard G, Galluppi F, Shi BE, Benosman RB. HOTS: A Hierarchy of Event-Based Time-Surfaces for Pattern Recognition. IEEE Trans Pattern Anal Mach Intell. 2017;39(7):1346‐1359. doi:10.1109/TPAMI.2016.2574707
Download the event-Python library from Github:
git clone https://github.com/gorchard/event-Python.git event_Python
touch event_Python/__init__.py
Install dependencies:
conda env create -f environment.yaml
Download the N-MNIST dataset from https://www.garrickorchard.com/datasets/n-mnist and place in ./datasets/mnist/
Run the Jupyter Notebook to train and visualise the digit classification:
jupyter notebook train_and_test_hots_model.ipynb
The training results for each of the layers will be shown. Here are screenshots from each of the layers: