Semi-Supervised Learning for Surface EMG-based Gesture Recognition

Prerequisite

• A CUDA compatible GPU
• Ubuntu 14.04 or any other Linux/Unix that can run Docker
• Docker
• Nvidia Docker
• Download the docker image:

  docker pull answeror/sigr:semi
  

  or build it by yourself:

  docker build -t answeror/sigr:semi -f docker/Dockerfile .
  

Steps to generate table 4

# Prepare data
mkdir -p .cache
# Download https://www.idiap.ch/project/ninapro
# Put NinaPro DB1 in .cache/ninapro-db1-raw
# Download http://zju-capg.org/myo/data
# Put CapgMyo DB-a in .cache/dba
# Put CapgMyo DB-b in .cache/dbb
# Put CapgMyo DB-c in .cache/dbc
# Download http://www.csl.uni-bremen.de/cms/forschung/bewegungserkennung
# Put csl-hdemg in .cache/csl

scripts/train_table_4.sh
scripts/app test_semimyo --cmd table_4

Training on NinaPro and CapgMyo will take 1 to 2 hours depending on your GPU. Training on csl-hdemg will take several days. You can accelerate traning and testing by distribute different folds on different GPUs with the gpu parameter.

You can also do train and test for each dataset on different machines or GPUs:

scripts/train_db1.sh
scripts/train_dba.sh
scripts/train_dbb.sh
scripts/train_dbc.sh
scripts/train_csl.sh
scripts/app test_semimyo --cmd table_4_db1
scripts/app test_semimyo --cmd table_4_dba
scripts/app test_semimyo --cmd table_4_dbb
scripts/app test_semimyo --cmd table_4_dbc
scripts/app test_semimyo --cmd table_4_csl

Steps to generate table 1

# Prepare data
mkdir -p .cache
# Download https://www.idiap.ch/project/ninapro
# Put NinaPro DB1 in .cache/ninapro-db1-raw
# Extract pre-calculated cluster (512 clusters) labels to data folder
unzip data/pose-512 -d .cache/ninapro-db1-raw

scripts/train_table_1.sh
scripts/app test_semimyo --cmd table_1

License

Licensed under an GPL v3.0 license.

Bibtex

@inproceedings{Du_IJCAI_2017,
  author    = {Yu Du, Yongkang Wong, Wenguang Jin, Wentao Wei, Yu Hu, Mohan Kankanhalli, Weidong Geng},
  title     = {Semi-Supervised Learning for Surface EMG-based Gesture Recognition},
  booktitle = {Proceedings of the Twenty-Sixth International Joint Conference on
               Artificial Intelligence, {IJCAI-17}},
  pages     = {1624--1630},
  year      = {2017},
  doi       = {10.24963/ijcai.2017/225},
  url       = {https://doi.org/10.24963/ijcai.2017/225},
}

Misc

Thanks DMLC team for their great MxNet!