color2grayscale

CNN based model to convert any RGB image to Gray-scale

Environment

• Ubuntu 18.04
• Python 3.6.*
• Tensorflow 1.11.0

Directories

• model/ keep saved model files.
• images/intermediate/ any intermediate files/images.
• images/predicted/ predicted images or output gray-scale images from trained model file.
• images/source/ RGB source images.
• images/target/ target gray-scale images of source color images for training.
• images/validate/ fresh color images for validation, not a subset of source images.

Use color2gray to genarate gray-scale images from source images for training. Configure directory paths and parameters in config.ini file.

How to use

from color2gray import COLOR2GRAY as C2G
from predict import PREDICT as PRED
## Train ##

# Adopt type1 or type2 for training
# Type 1, w.r.t to config file

c2g_obj = C2G()
c2g_obj.read_dataset()
c2g_obj.train()

# Type 2
read_args = {
    'color_img_dir': color_image_path,
    'target_gray_img_dir': target_grayscale_image_path,
    'model_dir': model_directory_path
}
c2g_obj = C2G(**read_args)
c2g_obj.read_dataset(img_type='png')
train_args = {
    'batch_size': 32,
    'epoch_num': 50
}
c2g_obj.train(**train_args)

## Predict
# Type 1
pred_obj = PRED()
pred_obj.predict()

# Type 2
pred_obj = PRED()
pred_args = {
    'color_img_dir': testing_input_image_directory_path,
    'predict_img_dir': result_image_directory_path,
    'img_type': png/jpg
}
pred_obj.predict(**pred_args)