Deep Learning pipeline for motor-imagery classification.

Last update: Oct 31, 2022

Overview

BCI-ToolBox

1. Introduction

BCI-ToolBox is deep learning pipeline for motor-imagery classification.
This repo contains five models: ShallowConvNet, DeepConvNet, EEGNet, FBCNet, BCI2021.
(BCI2021 is not an official name.)

2. Installation

Environment

Python == 3.7.10
PyTorch == 1.9.0
mne == 0.23.0
braindecode == 0.5.1
CUDA == 11.0

Create conda environment

conda install pytorch=1.9.0 cudatoolkit=11.1 -c pytorch -c nvidia
conda install numpy pandas matplotlib pyyaml ipywidgets
pip install torchinfo braindecode moabb mne

3. Directory structure

.
├── README.md
├── base
│   ├── constructor.py
│   └── layers.py
├── configs
│   ├── BCI2021
│   │   └── default.yaml
│   ├── DeepConvNet
│   │   └── default.yaml
│   ├── EEGNet
│   │   └── default.yaml
│   ├── FBCNet
│   │   └── default.yaml
│   ├── ShallowConvNet
│   │   └── default.yaml
│   └── demo
│       ├── arch.yaml
│       ├── bci2021.yaml
│       ├── test.yaml
│       ├── train.yaml
│       └── training_params.yaml
├── data_loader
│   ├── data_generator.py
│   ├── datasets
│   │   ├── __init__.py
│   │   ├── bnci2014.py
│   │   ├── cho2017.py
│   │   ├── folder_dataset.py
│   │   ├── openbmi.py
│   │   └── tmp_dataset.py
│   └── transforms.py
├── main.py
├── models
│   ├── BCI2021
│   │   ├── BCI2021.py
│   │   └── __init__.py
│   ├── DeepConvNet
│   │   ├── DeepConvNet.py
│   │   └── __init__.py
│   ├── EEGNet
│   │   ├── EEGNet.py
│   │   └── __init__.py
│   ├── FBCNet
│   │   ├── FBCNet.py
│   │   └── __init__.py
│   ├── ShallowConvNet
│   │   ├── ShallowConvNet.py
│   │   └── __init__.py
│   ├── __init__.py
│   └── model_builder.py
├── trainers
│   ├── __init__.py
│   ├── cls_trainer.py
│   └── trainer_maker.py
└── utils
    ├── calculator.py
    ├── painter.py
    └── utils.py

4. Dataset

Use braindecode

5. Get Started

We use W&B for experimental management tool.
Example for W&B Dashboard
Example for W&B Report

Create wandb_key.yaml file

Create wandb_key.yaml file in configs directory.
```
# wandb_key.yaml
key: WANDB API keys
```
WANDB API keys can be obtained from your W&B account settings.

train

Use W&B

python main.py --config_file=configs/demo/train.yaml

Not use W&B

python main.py --config_file=configs/demo/train.yaml --no_wandb

USE GPU

python main.py --config_file=configs/demo/train.yaml --device=0  # Use GPU 0
python main.py --config_file=configs/demo/train.yaml --device=1  # Use GPU 1
python main.py --config_file=configs/demo/train.yaml --device=2  # Use GPU 2

GPU numbers depend on your server.

USE Sweep

# W&B
sweep_file: configs/demo/training_params.yaml
project: Demo
tags: [train]

Add this block to config file for finding training parameters.

# W&B
sweep_file: configs/demo/arch.yaml
sweep_type: arch
project: Demo
tags: [train]

Add this block to config file for finding model architecture.

test

python main.py --config_file=configs/demo/test.yaml

5. References

ShallowConvNet [Paper] [Code]
DeepConvNet [Paper] [Code]
EEGNet [Paper] [Code]
FBCNet [Paper] [Code]
BCI2021 [Paper] [Code]

Deep Learning pipeline for motor-imagery classification.

Related tags

Overview

BCI-ToolBox

1. Introduction

2. Installation

Environment

Create conda environment

3. Directory structure

4. Dataset

5. Get Started

Create wandb_key.yaml file

train

test

5. References

Owner

DongHee

This is the official PyTorch implementation for "Mesa: A Memory-saving Training Framework for Transformers".

TensorFlow Implementation of "Show, Attend and Tell"

FPGA: Fast Patch-Free Global Learning Framework for Fully End-to-End Hyperspectral Image Classification

Contextual Attention Localization for Offline Handwritten Text Recognition

Python version of the amazing Reaction Mechanism Generator (RMG).

Curriculum Domain Adaptation for Semantic Segmentation of Urban Scenes, ICCV 2017

SANet: A Slice-Aware Network for Pulmonary Nodule Detection

The original implementation of TNDM used in the NeurIPS 2021 paper (no longer being updated)

Torch Implementation of "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network"

Implementation of a memory efficient multi-head attention as proposed in the paper, "Self-attention Does Not Need O(n²) Memory"

Unofficial PyTorch implementation of Attention Free Transformer (AFT) layers by Apple Inc.

This is the repository for the NeurIPS-21 paper [Contrastive Graph Poisson Networks: Semi-Supervised Learning with Extremely Limited Labels].

PyTorch implementation of saliency map-aided GAN for Auto-demosaic+denosing

a Lightweight library for sequential learning agents, including reinforcement learning

Manim is an engine for precise programmatic animations, designed for creating explanatory math videos

Density-aware Single Image De-raining using a Multi-stream Dense Network (CVPR 2018)

A simple Rock-Paper-Scissors game using CV in python

[CVPR 2020] GAN Compression: Efficient Architectures for Interactive Conditional GANs

Code for You Only Cut Once: Boosting Data Augmentation with a Single Cut

Official PyTorch implementation of Data-free Knowledge Distillation for Object Detection, WACV 2021.