Windows Only Pre-Download Steps
Open command prompt as administrator and run wsl --install to enable Windows subsytem for linux. After this completes windows will need to restart.
Visit the Microsoft Store and install Ubuntu 20.04 LTS. Once the store has downloaded the application launch Ubuntu 20.04 LTS from the start menu.
The first time you run the Ubuntu app it will need sometime to setup and ask you to create a username and password to use in the application.
Install the required pre-requisite packages by running the following commands:
sudo apt update
sudo apt install python3-pip
To ensure the python packages installed are accessible from the command line run:
echo -e "\nexport PATH=/home/$USER/.local/bin:\$PATH" >> ~/.bashrc
exec bash
To checkout the application to run locally on your machine run the commands:
git clone https://github.com/XDynames/SAI-app.git
cd SAI-app
SAI can either use a computer's CPU to process samples or its GPU. The GPU will be significantly faster but requires some extra installation steps:
CPU mode
MacOS
Ensure you have Homebrew setup and install the following packages
brew install geos gdal libjpeg
If you have a Macbook that uses an Apple Silicone based CPU run
pip3 install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu
Windows and Linux
Ensure you have libgeos installed:
sudo apt install libgeos-dev
Once the operation system specific steps are completed run:
bash setup.sh
GPU accelerated mode
Install the appropriate versions of Pytorch and Detectron2 to suit your GPU.
Run bash setup_gpu.sh to install the remaining dependencies.
Docker
If you have docker and nvidia container runtime configured you can start SAI by running:
cd docker
docker compose up
To make your data available to SAI inside the docker container modify the volume mount ../../stomata_data/:/code/data to suit your needs
Run ./SAI while inside the checked out folder. This will automatically open a webpage in your browser where you can view and use the application.
When using CPU only measurement time has been measured at 60-6 seconds per image, this is also dependant on image resolution. In GPU accelerated mode this time drops to 0.8-0.2 seconds per image.
If you want to test some samples the images we used for validation can be downloaded from here for Barley and Arabidopsis
Windows: 10 & 11
Ubuntu: 20.04 & 18.04
MacOS: BigSur & Monterey
If you use SAI as part of your work please cite us:
@article{https://doi.org/10.1111/nph.18765,
author = {Sai, Na and Bockman, James Paul and Chen, Hao and Watson-Haigh, Nathan and Xu, Bo and Feng, Xueying and Piechatzek, Adriane and Shen, Chunhua and Gilliham, Matthew},
title = {StomaAI: an efficient and user-friendly tool for measurement of stomatal pores and density using deep computer vision},
journal = {New Phytologist},
volume = {238},
number = {2},
pages = {904-915},
keywords = {applied deep learning, computer vision, convolutional neural network, phenotyping, stomata},
doi = {https://doi.org/10.1111/nph.18765},
url = {https://nph.onlinelibrary.wiley.com/doi/abs/10.1111/nph.18765},
eprint = {https://nph.onlinelibrary.wiley.com/doi/pdf/10.1111/nph.18765},
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent