jatiink/Image-Colorization-Pytorch

Image-Colorization-Pytorch

In this project, I used the Imagenet Dataset and a U-net model. This model works with the L channel from LAB COLOR SPACE and can predict the A and B channels.

I also used Tensorboard for live results of training and validation.

INSTRUCTIONS

This project requires the following libraries :

• io(Scikit-image)
• Torch(Pytorch)
• Numpy
• Cv2(OpenCV)
• Matplotlib
• torch.utils.tensorboard

Please ensure you have installed the following libraries mentioned above before continuing.

To install the following libraries.

Activate your virtual environment and type:

pip install -r requirement.txt

HOW TO Get results

To get results on new Image.
Extract all files in one folder.

Run CMD in folder directory and type:

To view only

python get_result.py --img "Image path"

To save output Image

python get_result.py --img "Image path" --save-output

To save concatenated Images

python get_result.py --img "Image path" --save-conc

Example:

python get_result.py --img_root "C:/user/data/img.jpg" --show --save-output

Output will be saved in same directory.

Note: There is a limitations of colors in trained model.

Training Results