NN-Illumination-Estimation-FPM

Accompanying code for the paper "Object Detection Neural Network Improves Fourier Ptychographic Reconstruction" (Paper).

Installation

Create a New Environment (optional)

To avoid messing up your exisiting setup, it is recommended to create a new virtual environment before you proceed. The following command is for Anaconda package manager, it is a useful package manager with a good repository system of its own that reduces some hassles. Anaconda can be downloaded from here. If you don't wish to use Anaconda, you can use venv to achieve the same effect.

conda create -n myenv python=3.7
conda activate myenv

Install PyTorch

Install PyTorch, Torchvision and CUDAToolkit (10.2 needed for compatibility with Detectron2, newer versions might give compatibility issues). If you don't wish to use Anaconda, you will have to install CUDAToolkit-10.2 separately and then install pytorch + torchvision using the pip command available on their website.

conda install pytorch torchvision cudatoolkit=10.2 -c pytorch

Detectron2 and Other Requirements

For Linux Users

The next step for Linux users is to simply install the necessary requirements.

pip install -r requirements.txt

For Windows Users

Detectron2 does not yet officially support Windows. The community has managed to get Detectron2 running on Windows by modifying some of the code.

1. Install Detectron2 by following the instructions given on this well-written blog.

2. Install rest of the requirements

pip install -r requirements-win.txt

Usage

TBD