Image Inpainting

Run

python dataset_prep.py

This script downloads, extracts and stores a subset of imagenet into the data directory within the current working directory.

Folder structure

-data
    -inpaint
        -original
            -train
            -train_mask
            -val
            -val_mask
        -line_mask
            -train
            -val

original folder contains the uncorrupted images, line_mask contains corrupted images which have random noise in the form lines with varying thickness, and random_mask contains random noise across all pixels.

train_mask contains .npy files which are 2D numpy arrays where the value at (i, j)th location is 1 if that value is known in the image, 0 otherwise.

train - 9296 images val - 3856 images