/Astronomical-Images-Restoration

Primary LanguagePython

Astronomical-Images-Restoration

This project is about generating nearly original images using convoluted images by using Pix-2-Pix GAN.

As the dataset is private I cannot share the images here.

Steps to run the project

  1. We will create an environment to replicate the desired effects.
  2. Place the file "tf.yml" in a directory.
  3. In anaconda prompt go to the directory and run the command conda env create -v -f tf.yml
  4. After the environment is created run the command conda activate tf
  5. Run the command pip install opencv-python after step 4
  6. Now navigate inside TF_pix2pix and run preprocessing.py with appropriate arguments to generate the necessary files required for training
  7. Run pix2pix.py with appropriate arguments for training the model

Directory structure

Astronomical-Images-Restoration
|
|----real_images # original images 
|	|
|	|----train # original training images
|	|
|	|----test # only 1 test image
|
|----conv_images # convoluted images
|	|
|	|----train # convoluted training images
|	|
|	|----test # only 1 test image
|
|----pix2pix.py # main training code 
|
|----preprocessing.py # preprocessing code 
|
|----tf.yml # file used to create new environment in anaconda with required specifications