Extract the Bayesian Mask and Natural Color Images from SLSTR products

• folder_A is the location of SLSTR packages, they should have "S3A*" at the begining of the name.

• folder_B is the location you want to store Bayesian Masks. There is also a function for extract Summary Masks in the code. By default, we are extract Bayesian Masks.

• folder_C is the location you want to store the natural color satellite images.

  python3 slstr_preprocess.py --read_product --folder_A data --folder_B bayes_mask --folder_C natural_image
  

Resize the masks into the same size of satellite images

• folder_A is the original masks.

• folder_B is the location you want to store the resized masks.

  python3 slstr_preprocess.py --resize_mask --folder_A bayes_mask --folder_B bayes_mask/resized_mask
  

Crop the images into small pieces

• folder_A is the source folder.

• folder_B is the location you want to store the cropped images.

• window_size is the side of the square. e.g. 256*256

  python3 slstr_preprocess.py --crop_image --folder_A bayes_mask/resized_mask --folder_B bayes_mask/cropped_mask --window_size 256
  python3 slstr_preprocess.py --crop_image --folder_A natural_image --folder_B natural_image/cropped_natural --window_size 256
  

Remove pure white and black masks. And the corresponding natural color images.

• folder_A is the masks.

• folder_B is the natural color images.

  python3 slstr_preprocess.py --remove_bw --folder_A bayes_mask/cropped_mask --folder_B natural_image/cropped_natural

If you are using a Mac, you might need to remove all '.DC_Store' files before running the combination code.

  find . -name ".DS_Store" -delete

Then you can run the code from pix2pix to combine your images

  python3 datasets/combine_A_and_B.py --fold_A datasets/aligned/A --fold_B datasets/aligned/B --fold_AB datasets/AtoB_lambda_AB

search threshold. This command can find the best value for binnary transformation from 120-190. This could take a while. From my experiment, 180 usually achieves the best results.

• folder_A is the results of pix2pix

  python3 ImageClassify.py --search_threshold --folder_A test_results/BCELoss
  

binary transformation. This is default using 180 as threshold

• folder_A is the results of pix2pix. This will save the bianary results inside the folder_A. e.g. test_results/BCELoss/binary_fake

  python3 ImageClassify.py --binary_transformation --folder_A test_results/BCELoss
  

caculate accuracy, recall, precision and F1 Score. The output will be out print into a txt called acc_output.txt.

• folder_A is the results of pix2pix.

  python3 ImageClassify.py --caculate_acc --folder_A test_results/BCELoss
  

caculate PSNR and SSIM. This is used for the direction "mask to image"

• folder_A is the results of pix2pix.

  python3 ImageClassify.py --caculate_AtoB --folder_A test_results/sample