CNN_model_ColorConstancy

hienpham15 / Client-side-Watermarking

A C++ implementation of client-side watermarking using Paillier Cryptosystem.

0 stars 0 forks source link

CNN_model_ColorConstancy #1

Open ywj55555 opened 2 years ago

ywj55555 commented 2 years ago

Hello, first of all, thank you very much for being able to open source your code. In your project: CNN_model_ColorConstancy, I have a question, are the images in the Train_set generated in preapare_dataset.py? In this case, tthe images in the training set have passed White balance correction. From my understanding, shouldn’t the training image be the original image with color cast? I will be very happy if you can give me an answer.

hienpham15 commented 2 years ago

Hi,

The images for train/test are generated by the generate_data.py script.
The prepare_dataset.py performs white-balancing on the color-casted images to obtain ground truth images (the illuminants are provided with the dataset). You need these images to compare (calculate the angular error, etc.) with the images corrected by the CNN.

ywj55555 commented 2 years ago

Thank you for your recovery, I see