Code for preprocessing and loading data from neuromorphic datasets. Part of this code has been used for the following works:
H. Jang, N. Skatchkovsky, and O. Simeone, VOWEL: A Local Online Learning Rule for Recurrent Networks of Probabilistic Spiking Winner-Take-All Circuits, to be presented at ICPR 2020 https://arxiv.org/abs/2004.09416
H. Jang, N. Skatchkovsky, and O. Simeone, BiSNN: Training Spiking Neural Networks with Binary Weights via Bayesian Learning, submitted for conference publication, https://arxiv.org/abs/2012.08300
This code can now be installed as a package and is meant to be eventually shared in pip. To clone and install locally the package, run
git clone https://github.com/kclip/neurodata
cd neurodata/
python -m pip install -e .
Scripts to preprocess the MNIST-DVS and DVSGestures dataset are given in the preprocessing module.
Make sure to first download and then preprocess the dataset using the script in preprocessing.
To add your own datasets, save them as an .hdf5 file respecting the current structure:
/ root (Group)
/stats (Group)
/stats/test_data (Array) [n_examples_test, n_pixels_per_dim]
/stats/test_label (Array) [n_examples_test, n_classes]
/stats/train_data (Array) [n_examples_train, n_pixels_per_dim]
/stats/train_label (Array) [n_examples_train, n_classes]
/train (Group)
/train/labels (Array) [n_examples_train, 1] # indicates labels for each train example
/train/1 (Array) [example_length, 4] # indicates event time, x axis position, y axis position, polarity
...
/train/n_examples_train (Array) [example_length, 4] # one array per train example
/test (Group)
/test/labels (Array) [n_examples_test, 1] # indicates labels for each test example
/test/1 (Array) [example_length, 4] # indicates event time, x axis position, y axis position, polarity
...
/test/n_examples_test (Array) [example_length, 4] # one array per test example
Author: Nicolas Skatchkovsky