This repository is a reimplementation of the paper STDP-based Spiking Deep Convolutional Neural Networks for Object Recognition by Kheradpisheh et al (2017) using PyTorch.
The code is based on the original implementation in SpykeTorch, but has been simplified and streamlined for ease of use and readability. The main goal of this reimplementation is to provide a more focused and accessible implementation of the STDP-based spiking deep convolutional neural network architecture described in the paper.
To get started, clone this repository and install the necessary dependencies:
git clone https://github.com/Fatma-Chaouech/STDP-based-DCNN.git
cd STDP-based-DCNN
pip install -r requirements.txt
Next, you can run the main training script to train the network on a dataset of your choice:
python3 main.py --phase train --dataset <path-to-train-dataset>
By default, the script will train the network on the MNIST dataset. You can specify a different dataset by providing the appropriate command line argument.
To test your model, you can run the following command in your terminal:
python3 --phase test --dataset <path-to-test-dataset> --weights_path <path-to-weights> --classifier_path <path-to-classifier>
You can customize the values of --dataset, --weights_path, and --classifier_path to suit your specific needs. Additionally, you can edit these parameters in the config.json file.
For a detailed summary of the paper, please refer to the file paper-summary.md.