Hand_Segmentation

We Present a Mask-RCNN based real-time(GPU) hand segmentation.

Preparation

Plz make sure you are have an python3.6 environment or folllow the steps below to create one.

Create Environment

1. Downloas Anaconda's Official Website [Here]
2. Install Anaconda

cd to/download/path
sudo chmod +x <Anaconda-installer-name>
./<Anaconda-installer-name>  ## make sure anser yes for every question

3. Create a python3.6 environment

source ~/.bashrc
conda create --name py36 python=3.6
conda activte py36
conda install jupyter

TsesorFlow-Gpu and Cuda

Make sure you follow the steps Here carefully

OpenCV

Make sure you follow the steps Here carefully. Also install opencv in your environment

conda activate py36
conda install -c conda-forge opencv 

Mask-RCNN

Download my files and mrcnn

git clone https://github.com/TsengMJ/Hand_Segmentation.git
pip install mrcnn

Preparing data

Try this tools to make your own training data, or Using EgoHands Dataset

We present a being processed dataset from egohands in datasets/hand

Training & Testing

Execute src/train_hand_shape.ipynb for training and testing on our data.

If you want to train on your own dataset, please change the Paths and beware the jason file's format

Result