This is Pytorch code for edge connection with GAN,in which a autoencoder structure is utilized for generator

• data : Images data
• DefectDataset : Images of groundtruth and augmented dataset
  • gt
  • noise
  • background : Edges of clutter
• GAN_Image : Images during the training procedure
• Model
  • DIS
  • GAN
• Template : A set of well connected contour images
  • bin_contour
• Test_Image : The output of model
  • input
  • output
• src
  • AAE.py
  • AAEWithClassifier.py
  • datasetGenerate.py
  • testAAE.py
  • testAAEWithClassifier.py
  • region.py
• README.md
• script : Matlab scripts
• roi : Roi cropped by mask

pytorch
PIL
cv2

## Go to src folder
cd src/

## Generate  dataset

python3 datasetGenerate.py

## Train AEGAN
python3 AAE.py

## Test autoencoder
python3 testAAE.py

Using Matlab opening guidence_template.m

19.7.26: Update AAE.py, region.py, testAAE.py; The problem that the holes inside mask expanded is resolved.

19.8.02: Update AAEWithClassifier.py, testAAEWithClassifier.py; Add illumination normalization procedure for origin images, then choose SQI technique results as sources.