File size: 10,384 Bytes
a3a12c6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 |
"""
Original code: https://github.com/swz30/Restormer/blob/main/Defocus_Deblurring/generate_patches_dpdd.py
by Syed Waqas Zamir
"""
##### Data preparation file for training Restormer on the DPDD Dataset ########
import cv2
import numpy as np
from glob import glob
from natsort import natsorted
import os
from tqdm import tqdm
from copy import deepcopy
from joblib import Parallel, delayed
def shapness_measure(img_temp,kernel_size):
conv_x = cv2.Sobel(img_temp,cv2.CV_64F,1,0,ksize=kernel_size)
conv_y = cv2.Sobel(img_temp,cv2.CV_64F,0,1,ksize=kernel_size)
temp_arr_x=deepcopy(conv_x*conv_x)
temp_arr_y=deepcopy(conv_y*conv_y)
temp_sum_x_y=temp_arr_x+temp_arr_y
temp_sum_x_y=np.sqrt(temp_sum_x_y)
return np.sum(temp_sum_x_y)
def filter_patch_sharpness(patches_src_c_temp, patches_trg_c_temp, patches_src_l_temp, patches_src_r_temp):
patches_src_c, patches_trg_c, patches_src_l, patches_src_r = [], [], [], []
fitnessVal_3=[]
fitnessVal_7=[]
fitnessVal_11=[]
fitnessVal_15=[]
num_of_img_patches=len(patches_trg_c_temp)
for i in range(num_of_img_patches):
fitnessVal_3.append(shapness_measure(cv2.cvtColor(patches_trg_c_temp[i], cv2.COLOR_BGR2GRAY),3))
fitnessVal_7.append(shapness_measure(cv2.cvtColor(patches_trg_c_temp[i], cv2.COLOR_BGR2GRAY),7))
fitnessVal_11.append(shapness_measure(cv2.cvtColor(patches_trg_c_temp[i], cv2.COLOR_BGR2GRAY),11))
fitnessVal_15.append(shapness_measure(cv2.cvtColor(patches_trg_c_temp[i], cv2.COLOR_BGR2GRAY),15))
fitnessVal_3=np.asarray(fitnessVal_3)
fitnessVal_7=np.asarray(fitnessVal_7)
fitnessVal_11=np.asarray(fitnessVal_11)
fitnessVal_15=np.asarray(fitnessVal_15)
fitnessVal_3=(fitnessVal_3-np.min(fitnessVal_3))/np.max((fitnessVal_3-np.min(fitnessVal_3)))
fitnessVal_7=(fitnessVal_7-np.min(fitnessVal_7))/np.max((fitnessVal_7-np.min(fitnessVal_7)))
fitnessVal_11=(fitnessVal_11-np.min(fitnessVal_11))/np.max((fitnessVal_11-np.min(fitnessVal_11)))
fitnessVal_15=(fitnessVal_15-np.min(fitnessVal_15))/np.max((fitnessVal_15-np.min(fitnessVal_15)))
fitnessVal_all=fitnessVal_3*fitnessVal_7*fitnessVal_11*fitnessVal_15
to_remove_patches_number=int(to_remove_ratio*num_of_img_patches)
for itr in range(to_remove_patches_number):
minArrInd=np.argmin(fitnessVal_all)
fitnessVal_all[minArrInd]=2
for itr in range(num_of_img_patches):
if fitnessVal_all[itr]!=2:
patches_src_c.append(patches_src_c_temp[itr])
patches_trg_c.append(patches_trg_c_temp[itr])
patches_src_l.append(patches_src_l_temp[itr])
patches_src_r.append(patches_src_r_temp[itr])
return patches_src_c, patches_trg_c, patches_src_l, patches_src_r
def slice_stride(_img_src_c, _img_trg_c, _img_src_l, _img_src_r):
coordinates_list=[]
coordinates_list.append([0,0,0,0])
patches_src_c_temp, patches_trg_c_temp, patches_src_l_temp, patches_src_r_temp = [], [], [], []
for r in range(0,_img_src_c.shape[0],stride[0]):
for c in range(0,_img_src_c.shape[1],stride[1]):
if (r+patch_size[0]) <= _img_src_c.shape[0] and (c+patch_size[1]) <= _img_src_c.shape[1]:
patches_src_c_temp.append(_img_src_c[r:r+patch_size[0],c:c+patch_size[1]])
patches_trg_c_temp.append(_img_trg_c[r:r+patch_size[0],c:c+patch_size[1]])
patches_src_l_temp.append(_img_src_l[r:r+patch_size[0],c:c+patch_size[1]])
patches_src_r_temp.append(_img_src_r[r:r+patch_size[0],c:c+patch_size[1]])
elif (r+patch_size[0]) <= _img_src_c.shape[0] and not ([r,r+patch_size[0],_img_src_c.shape[1]-patch_size[1],_img_src_c.shape[1]] in coordinates_list):
patches_src_c_temp.append(_img_src_c[r:r+patch_size[0],_img_src_c.shape[1]-patch_size[1]:_img_src_c.shape[1]])
patches_trg_c_temp.append(_img_trg_c[r:r+patch_size[0],_img_trg_c.shape[1]-patch_size[1]:_img_trg_c.shape[1]])
patches_src_l_temp.append(_img_src_l[r:r+patch_size[0],_img_src_l.shape[1]-patch_size[1]:_img_src_l.shape[1]])
patches_src_r_temp.append(_img_src_r[r:r+patch_size[0],_img_src_r.shape[1]-patch_size[1]:_img_src_r.shape[1]])
coordinates_list.append([r,r+patch_size[0],_img_src_c.shape[1]-patch_size[1],_img_src_c.shape[1]])
elif (c+patch_size[1]) <= _img_src_c.shape[1] and not ([_img_src_c.shape[0]-patch_size[0],_img_src_c.shape[0],c,c+patch_size[1]] in coordinates_list):
patches_src_c_temp.append(_img_src_c[_img_src_c.shape[0]-patch_size[0]:_img_src_c.shape[0],c:c+patch_size[1]])
patches_trg_c_temp.append(_img_trg_c[_img_trg_c.shape[0]-patch_size[0]:_img_trg_c.shape[0],c:c+patch_size[1]])
patches_src_l_temp.append(_img_src_l[_img_src_l.shape[0]-patch_size[0]:_img_src_l.shape[0],c:c+patch_size[1]])
patches_src_r_temp.append(_img_src_r[_img_src_r.shape[0]-patch_size[0]:_img_src_r.shape[0],c:c+patch_size[1]])
coordinates_list.append([_img_src_c.shape[0]-patch_size[0],_img_src_c.shape[0],c,c+patch_size[1]])
elif not ([_img_src_c.shape[0]-patch_size[0],_img_src_c.shape[0],_img_src_c.shape[1]-patch_size[1],_img_src_c.shape[1]] in coordinates_list):
patches_src_c_temp.append(_img_src_c[_img_src_c.shape[0]-patch_size[0]:_img_src_c.shape[0],_img_src_c.shape[1]-patch_size[1]:_img_src_c.shape[1]])
patches_trg_c_temp.append(_img_trg_c[_img_trg_c.shape[0]-patch_size[0]:_img_trg_c.shape[0],_img_trg_c.shape[1]-patch_size[1]:_img_trg_c.shape[1]])
patches_src_l_temp.append(_img_src_l[_img_src_l.shape[0]-patch_size[0]:_img_src_l.shape[0],_img_src_l.shape[1]-patch_size[1]:_img_src_l.shape[1]])
patches_src_r_temp.append(_img_src_r[_img_src_r.shape[0]-patch_size[0]:_img_src_r.shape[0],_img_src_r.shape[1]-patch_size[1]:_img_src_r.shape[1]])
coordinates_list.append([_img_src_c.shape[0]-patch_size[0],_img_src_c.shape[0],_img_src_c.shape[1]-patch_size[1],_img_src_c.shape[1]])
return patches_src_c_temp, patches_trg_c_temp, patches_src_l_temp, patches_src_r_temp
def train_files(file_):
lrL_file, lrR_file, lrC_file, hrC_file = file_
filename = os.path.splitext(os.path.split(lrC_file)[-1])[0]
lrL_img = cv2.imread(lrL_file, -1)
lrR_img = cv2.imread(lrR_file, -1)
lrC_img = cv2.imread(lrC_file, -1)
hrC_img = cv2.imread(hrC_file, -1)
lrC_patches, hrC_patches, lrL_patches, lrR_patches = slice_stride(lrC_img, hrC_img, lrL_img, lrR_img)
lrC_patches, hrC_patches, lrL_patches, lrR_patches = filter_patch_sharpness(lrC_patches, hrC_patches, lrL_patches, lrR_patches)
num_patch = 0
for lrC_patch, hrC_patch, lrL_patch, lrR_patch in zip(lrC_patches, hrC_patches, lrL_patches, lrR_patches):
num_patch += 1
lrL_savename = os.path.join(lrL_tar, filename + '-' + str(num_patch) + '.png')
lrR_savename = os.path.join(lrR_tar, filename + '-' + str(num_patch) + '.png')
lrC_savename = os.path.join(lrC_tar, filename + '-' + str(num_patch) + '.png')
hrC_savename = os.path.join(hrC_tar, filename + '-' + str(num_patch) + '.png')
cv2.imwrite(lrL_savename, lrL_patch)
cv2.imwrite(lrR_savename, lrR_patch)
cv2.imwrite(lrC_savename, lrC_patch)
cv2.imwrite(hrC_savename, hrC_patch)
def val_files(file_):
lrL_file, lrR_file, lrC_file, hrC_file = file_
filename = os.path.splitext(os.path.split(lrC_file)[-1])[0]
lrL_savename = os.path.join(lrL_tar, filename + '.png')
lrR_savename = os.path.join(lrR_tar, filename + '.png')
lrC_savename = os.path.join(lrC_tar, filename + '.png')
hrC_savename = os.path.join(hrC_tar, filename + '.png')
lrL_img = cv2.imread(lrL_file, -1)
lrR_img = cv2.imread(lrR_file, -1)
lrC_img = cv2.imread(lrC_file, -1)
hrC_img = cv2.imread(hrC_file, -1)
w, h = lrC_img.shape[:2]
i = (w-val_patch_size)//2
j = (h-val_patch_size)//2
lrL_patch = lrL_img[i:i+val_patch_size, j:j+val_patch_size,:]
lrR_patch = lrR_img[i:i+val_patch_size, j:j+val_patch_size,:]
lrC_patch = lrC_img[i:i+val_patch_size, j:j+val_patch_size,:]
hrC_patch = hrC_img[i:i+val_patch_size, j:j+val_patch_size,:]
cv2.imwrite(lrL_savename, lrL_patch)
cv2.imwrite(lrR_savename, lrR_patch)
cv2.imwrite(lrC_savename, lrC_patch)
cv2.imwrite(hrC_savename, hrC_patch)
############ Prepare Training data ####################
num_cores = 10
src = 'DPDD/train/'
tar = 'train-dpdd'
lrL_tar = os.path.join(tar, 'inputL_crops')
lrR_tar = os.path.join(tar, 'inputR_crops')
lrC_tar = os.path.join(tar, 'inputC_crops')
hrC_tar = os.path.join(tar, 'target_crops')
os.makedirs(lrL_tar, exist_ok=True)
os.makedirs(lrR_tar, exist_ok=True)
os.makedirs(lrC_tar, exist_ok=True)
os.makedirs(hrC_tar, exist_ok=True)
lrL_files = natsorted(glob(os.path.join(src, 'inputL', '*.png')))
lrR_files = natsorted(glob(os.path.join(src, 'inputR', '*.png')))
lrC_files = natsorted(glob(os.path.join(src, 'inputC', '*.png')))
hrC_files = natsorted(glob(os.path.join(src, 'target', '*.png')))
files = [(i, j, k, l) for i, j, k, l in zip(lrL_files, lrR_files, lrC_files, hrC_files)]
patch_size = [512, 512]
stride = [204, 204]
p_max = 0
to_remove_ratio = 0.3
Parallel(n_jobs=num_cores)(delayed(train_files)(file_) for file_ in tqdm(files))
############ Prepare validation data ####################
val_patch_size = 256
src = 'DPDD/test'
tar = 'test-dpdd'
lrL_tar = os.path.join(tar, 'inputL_crops')
lrR_tar = os.path.join(tar, 'inputR_crops')
lrC_tar = os.path.join(tar, 'inputC_crops')
hrC_tar = os.path.join(tar, 'target_crops')
os.makedirs(lrL_tar, exist_ok=True)
os.makedirs(lrR_tar, exist_ok=True)
os.makedirs(lrC_tar, exist_ok=True)
os.makedirs(hrC_tar, exist_ok=True)
lrL_files = natsorted(glob(os.path.join(src, 'inputL', '*.png')))
lrR_files = natsorted(glob(os.path.join(src, 'inputR', '*.png')))
lrC_files = natsorted(glob(os.path.join(src, 'inputC', '*.png')))
hrC_files = natsorted(glob(os.path.join(src, 'target', '*.png')))
files = [(i, j, k, l) for i, j, k, l in zip(lrL_files, lrR_files, lrC_files, hrC_files)]
Parallel(n_jobs=num_cores)(delayed(val_files)(file_) for file_ in tqdm(files)) |