abnvanand/imageinpainting

Implementation of the paper Image denoising and inpainting with DNN - J Xie

imageinpainting's Issues

Fix memory error when patching

See Patching.ipynb
Current implementation of patchify_v() function given memory error when run on our training set even after downsampling the images from 512x512 to 128x128.
Our training set contains 62 images of dimension 512x512

def patchify_v(images, shape):
    import numpy as np
    # By default it creats numpy array of dtype=float64 which will lead to Memory Error
    # float32 is a temporary solution and works sometimes
    # FIXME: find a way to prevent memory error
    res=np.array([patchify(img, shape) for img in images], dtype="float32")
    return res

def patchify(img, shape):
    from sklearn.feature_extraction.image import extract_patches_2d

    patches = extract_patches_2d(img, shape)
    return patches