-
Notifications
You must be signed in to change notification settings - Fork 1
/
Comparison.py
440 lines (326 loc) · 12 KB
/
Comparison.py
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
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
# coding: utf-8
# In[1]:
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image, ImageFilter
import math
from os import listdir
import random
import cv2
import pywt
import keras
from keras import backend as K
from keras.models import model_from_json
from keras.models import Model, Sequential
from keras.layers import Input, Convolution2D, Conv2D, MaxPooling2D, Dense, Dropout, Flatten, Conv2D, BatchNormalization, GlobalAveragePooling2D
from keras.callbacks import LearningRateScheduler
from keras.optimizers import SGD, Adam
from keras import regularizers, optimizers, losses, initializers
import tensorflow as tf
from tensorflow.python.client import device_lib
import skimage
from skimage import measure
from PIL import Image
# In[2]:
from srcnn import *
from utility import *
# In[3]:
#Division by 4
window_size= 64
scale = 4
stride = 16
# In[4]:
def calcPSNR(original,prediction):
return measure.compare_psnr(original, prediction,data_range=256)
# In[5]:
def calcSSIM(orig,pred):
return measure.compare_ssim(orig,pred,data_range=256, multichannel=True)
# In[6]:
def calcMSE(orig,pred):
return measure.compare_mse(orig,pred)
# In[7]:
#Returns original, bicubic and model prediction
def getImages(filename):
print("hellO?")
im = Image.open("set14/"+filename)
print("LLLLLLL")
img = np.asarray(im.convert('YCbCr'))
img = img[:,:,0]
org, bic = get_input_images(img, 4)
h, w = retSize(org)
sub_img = getSubImages(bic)
low_w = []
for i in range(len(sub_img)):
print(i,len(sub_img))
dwt_w = pywt.dwt2(sub_img[i], 'haar')
dwt_w = np.asfarray([dwt_w[0], dwt_w[1][0], dwt_w[1][1], dwt_w[1][2]])
low_w.append(dwt_w)
low_wavelet = np.asarray(low_w)
pred_w = model.predict(low_wavelet)
res = low_wavelet + pred_w
sub_imgs = []
for i in range(len(res)) :
patch = res[i]
dwt = (patch[0], (patch[1], patch[2], patch[3]))
wavelet = pywt.idwt2(dwt, 'haar')
sub_imgs.append(wavelet)
im = patch_to_image(sub_imgs, h, w)
return org,bic,im
# In[8]:
def convertToYCbCr(x):
return [image.convert('YCbCr') for image in x]
#Display images
def print_luminance(img, ch=1) :
if (ch == 1) :
plt.imshow(img, cmap=plt.get_cmap('gray'), interpolation='nearest')
else :
plt.imshow(img[:,:,0], cmap=plt.get_cmap('gray'), interpolation='nearest')
def getSubImage(img, height, width, centerX, centerY, len_interval, scale = 1) :
window = window_size // scale
return img[ height : height//2 + window + centerX*len_interval,
width // 2 + centerY*len_interval : width//2 + window + centerY*len_interval : ]
def retSize(x):
return (x.shape[0], x.shape[1])
def getSubImages(image, stride = stride) :
sub = []
height, width = retSize(image)
for h in range(0, height - window_size, stride) :
for w in range (0, width - window_size, stride) :
sub.append(image[h : h + window_size, w : w + window_size])
return sub
def appendSubbands(l1,l2,l3,l4, dwt):
l1 = np.append(l1 , dwt[0])
l2 = np.append(l2, dwt[1][0])
l3 = np.append(l3, dwt[1][1])
l4 = np.append(l4, dwt[1][2])
return (l1,l2,l3,l4)
def reshape(matrix, dwt_shape, channel='channels_last', ch=1) :
if (channel == 'channels_last') :
return matrix.reshape(-1, dwt_shape[0], dwt_shape[1], ch)
else :
return matrix.reshape(ch, -1, dwt_shape[0], dwt_shape[1])
def equals(a, b) :
t = a == b
count = 0
for a in range(0, len(t)) :
for b in range(len(t[0])) :
for c in range(len(t[1])) :
if ( t[a, b, c] == False):
count = count + 1
return count / (32*32*4) < 0.20
def getSubImages(image, stride = stride) :
sub = []
height, width = retSize(image)
for h in range(0, height - window_size, stride) :
for w in range (0, width - window_size, stride) :
sub.append(image[h : h + window_size, w : w + window_size])
return sub
def plot4images(im1, im2, im3, im4, ch=1) :
size = (10, 10)
fig = plt.figure(figsize=size)
plt.subplot(221)
plt.title("Approximation")
print_luminance(im1, ch )
plt.subplot(222)
plt.title("Horizontal")
print_luminance(im2, ch)
plt.subplot(223)
plt.title("Vertical")
print_luminance(im3, ch)
plt.subplot(224)
plt.title("Diagonal")
print_luminance(im4, ch)
def plot_waveletTrans(wt, ch=1) :
plot4images(wt[0], wt[1][0], wt[1][1], wt[1][2], ch)
# In[9]:
PATH = ""
TRAIN_FILE = PATH + "set14/"
obj_files = sorted(listdir(TRAIN_FILE )) #the filenames the call to sorted is actually necessary to ensure independence to filesystems
# In[10]:
high_res=obj_files[0::2]
low_res=obj_files[1::2]
# In[11]:
model_grey = load_model("SRCNN_deep")
#model.summary()
# In[12]:
model_r = load_model("SRCNN_red_channel")
model_g = load_model("SRCNN_green_channel")
model_b = load_model("SRCNN_blue_channel")
model_space= load_model("sptial_deep_model")
print("Models loaded")
# In[14]:
#This method assumes it gets a picture with mirror padding.
#It basically recreates said picture from patches and then crops accordingly to get back the good picture
def patch_to_image(patches, height, width, strides=stride, window_size=64) :
count = 0
if(len(patches[0].shape) == 3):
img = np.zeros((int(np.ceil(height/stride))*window_size, int(np.ceil(width/stride))*window_size,3))
else:
img = np.zeros((int(np.ceil(height/stride))*window_size, int(np.ceil(width/stride))*window_size))
for h in range(0, height, stride) :
for w in range (0, width, stride) :
if(len(patches[count].shape) == 3):
img[h : h + window_size, w : w + window_size,:] = patches[count]
else:
img[h : h + window_size, w : w + window_size,] = patches[count]
count = count + 1
return img[:height, :width]
#This method basically takes a picture and applies mirror-folding boundary conditions to it.
#This way, we actually don't have to put "black" pixels when near borders, which is better for our transforms
def getSubImages(image, stride = stride, window_size=64) :
sub = []
height, width = retSize(image)
moar = False
if(len(image.shape) == 3):
moar= True
for h in range(0, height, stride) :
for w in range (0, width, stride) :
x = h+window_size
y = w+window_size
i = h
j = w
concatX = False
concatY = False
if(x >= height):
x = x-height
if(y>= width):
y = y-width
i1 = image[h:height, w:width, :] if moar else image[h:height, w:width]
i2 = np.flip(image,axis=1)[h:height,0:y,:] if moar else np.flip(image,axis=1)[h:height,0:y]
i3 = np.flip(np.flip(image, axis=0), axis=1)[0:x,0:y,:] if moar else np.flip(np.flip(image, axis=0), axis=1)[0:x,0:y]
i4 = np.flip(image, axis=0)[0:x, w:width,:] if moar else np.flip(image, axis=0)[0:x, w:width]
e = np.concatenate((np.concatenate((i1,i2), axis=1),np.concatenate((i3,i4), axis=1)), axis=0)
else:
a = image[h:height, w:y, :] if moar else image[h:height, w:y]
b = np.flip(image, axis=0)[0:x, w:y,:] if moar else np.flip(image, axis=0)[0:x, w:y]
e = np.concatenate((a,b), axis=0)
elif(y >= width):
y = y-width
a = image[h:x, w:width,:] if moar else image[h:x, w:width]
b = np.flip(image,axis=1)[h:x, 0:y,:] if moar else np.flip(image,axis=1)[h:x, 0:y]
e = np.concatenate((a,b), axis=1)
else:
e = image[h:x, w:y,:] if moar else image[h:x, w:y]
sub.append(e)
return sub
#Returns original, bicubic and model prediction
def getSpacePred(org, bic, modelspace):
print("Predicting")
h, w = retSize(bic)
if(len(bic.shape) != 3): #meaning we've a greyscale image
bic = np.stack((bic,bic,bic), axis=2)
sub_img = getSubImages(bic,window_size=32)
print(np.asarray(sub_img).shape)
pred_s = modelspace.predict(np.asarray(sub_img),64)
im_sp = patch_to_image(pred_s, h, w, window_size=32)
return im_sp#np.uint8((im-np.min(im))/(np.max(im)-np.min(im))*255)
def getImages(img, modelfreq):
org, bic = get_input_images(img, 4)
h, w = retSize(org)
sub_img = getSubImages(bic)
low_w = []
for i in range(len(sub_img)):
dwt_w = pywt.dwt2(sub_img[i], 'haar')
dwt_w = np.asfarray([dwt_w[0], dwt_w[1][0], dwt_w[1][1], dwt_w[1][2]])
low_w.append(dwt_w)
low_wavelet = np.asarray(low_w)
print("predicting")
pred_w = modelfreq.predict(low_wavelet)
res = low_wavelet + pred_w
sub_imgs = []
for i in range(len(res)) :
patch = res[i]
dwt = (patch[0], (patch[1], patch[2], patch[3]))
wavelet = pywt.idwt2(dwt, 'haar')
sub_imgs.append(wavelet)
im = patch_to_image(sub_imgs, h, w)
return org,bic, im#np.uint8((im-np.min(im))/(np.max(im)-np.min(im))*255)
def getColourImages(filename):
img = Image.open("set14/"+filename)
img = np.asarray(img)
if(len(img.shape) == 3):
im_r = img[:,:,0]
im_g = img[:,:,1]
im_b = img[:,:,2]
org_r, bic_r, im_r = getImages(im_r,model_r)
org_g, bic_g, im_g = getImages(im_g,model_g)
org_b, bic_b, im_b = getImages(im_b,model_b)
org = np.stack((org_r, org_g, org_b), axis=2)
bic = np.stack((bic_r, bic_g, bic_b), axis=2)
im = np.stack((im_r, im_g, im_b), axis=2)
else:
im = Image.open("set14/"+filename)
img = np.asarray(im.convert('YCbCr'))
org, bic,im= getImages(img[:,:,0],model_grey)
space_pred = getSpacePred(org, bic, model_space)
return org, bic, im, space_pred
# In[ ]:
results=[]
for f in high_res:
print(f)
org,bic,pred, space_pred=getColourImages(f)
print("Got colout img")
if(len(org.shape) != 3):
space_pred = space_pred[:,:,0]
results.append((org,bic,pred, space_pred))
print("All results done")
# In[15]:
def convertToPicture(img):
if(len(img.shape) == 3):
img = Image.fromarray(img, 'RGB')
else:
img = Image.fromarray(img)
if img.mode != 'RGB':
img = img.convert('RGB')
return img
# In[16]:
i=0
ims = []
for res in results:
im_w = convertToPicture(np.uint8(np.clip(res[2],0,255)))
im_s = convertToPicture(np.uint8(np.clip(res[3],0,255)))
im_w.save("srcnn_results/"+str(i)+".png")
i=i+1
ims.append((convertToPicture(res[0]), convertToPicture(res[1]),im_w, im_s))
# In[17]:
def plot_results(i):
result=results[i]
org=result[0].astype(float)
bic=result[1].astype(float)
sr=result[2].astype(float)
spred=result[3].astype(float)
psnr_bic=round(calcPSNR(org,bic),4)
psnr_sr=round(calcPSNR(org,sr),4)
psnr_sd=round(calcPSNR(org, spred),4)
org,bic,sr, spred = ims[i]
plot_images([org,bic,sr, spred], ["Original", "Bic (PSNR: "+str(psnr_bic)+")", "SR (PSNR: "+str(psnr_sr)+")", "Spatial (PSNR: "+str(psnr_sd)+")"], size= (20,20), ch=1)
# In[18]:
def calcMeasures(i):
result=results[i]
org=result[0].astype(float)
bic=result[1].astype(float)
sr=result[2].astype(float)
psnr_bic=round(calcPSNR(org,bic),4)
psnr_sr=round(calcPSNR(org,sr),4)
ssim_bic=round(calcSSIM(org,bic),4)
ssim_sr=round(calcSSIM(org,sr),4)
mse_bic=round(calcMSE(org,bic),4)
mse_sr=round(calcMSE(org,sr),4)
print("Test image: ",i)
print("")
print("Bicubic")
print("PSNR=",psnr_bic)
print("SSIM=",ssim_bic)
print("MSE=",mse_bic)
print("")
print("SR")
print("PSNR=",psnr_sr)
print("SSIM=",ssim_sr)
print("MSE=",mse_sr)
print("-------------")
for i in range(14):
plot_results(i)
# In[19]:
for i in range(14):
calcMeasures(i)
# In[ ]: