Encoder-Decoder network to convert Color images to Black&White

Instructions to run:

• Make sure you have all required setup for tensorflow GPU(step available on tensorflow website)
• Install the requirements using pip(I recommend a virtual environment)
• Run: python3 testing.py
• Enter name of image to be converted
• The converted image will be saved as res.jpg folder

Retraining the model

• Copy any color images dataset to train_data/originalImages folder(I had used animals dataset from kaggle - not uploaded to github)
• Run: python3 data/createData.py
  • This will resize the images to 128x128 and save both color and grayscale images in respective folders
• Repeat the same for test images (Did not write that in script 😅 )
• Change the number of training and testing image variables as per your dataset in training.py file
• Run: python3 training.py (You can modify the epochs and other parameters if you like)
• Model will be saved in SavedModel folder

Note

I wrote this code because i wanted to try out encoder-decoder type of models. There must be a lot of better ways/models to do this.