rnc.py

import os
import numpy as np


def pad(img, crop_size):
    h, w, c = img.shape
    n_h = int(h/crop_size)
    n_w = int(w/crop_size)
    w_toadd = (n_w+1) * crop_size - w
    h_toadd = (n_h+1) * crop_size - h
    img_pad = np.pad(img, [(0, h_toadd), (0, w_toadd), (0,0)], mode='constant')
    return img_pad

def crop(img, crop_size, stride):
    cropped_images = []
    h, w, c = img.shape  
    n_h = int(h/stride)
    n_w = int(w/stride)

    for i in range(n_h):
        for j in range(n_w):
            crop_img = img[(i * stride):((i * stride) + crop_size), (j * stride):((j * stride) + crop_size), :]
            if (crop_img.shape) == (crop_size, crop_size, c):
                cropped_images.append(crop_img)
    return cropped_images

def uncrop(shape, crops, crop_size, stride):
    img = np.zeros(shape)
    h, w, c = shape  
    n_h = int(h/stride)
    n_w = int(w/stride)

    for i in range(n_h):
        for j in range(n_w):
            img[(i * stride):((i * stride) + crop_size), (j * stride):((j * stride) + crop_size), :] = crops[i * n_w + j]
    return img

def unpad(shape, img):
    h, w, c = shape
    return img[:h, :w, :]