-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.py
425 lines (355 loc) · 20.6 KB
/
test.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
import gc
import pathlib
import pickle
import time
# from PIL import Image
import numpy as np
import tensorflow as tf
import tensorflow_io as tfio
from comet_ml import Experiment
from matplotlib import pyplot as plt
from skimage.color import deltaE_cie76, deltaE_ciede94
from tabulate import tabulate
# Import SpecSeg network
from SpecSeg import SpecSeg
from utils import (
rescale_01,
)
"""
# ------------------------------------------------
#
# ████████ ███████ ███████ ████████
# ██ ██ ██ ██
# ██ █████ ███████ ██
# ██ ██ ██ ██
# ██ ███████ ███████ ██
#
# TEST FUNCTION FOR SHMGAN
# The test function has the following features:
# - Load RGB image from test folder as I0 (or Itot)
# - Set all other layers to zero
# - Set target image label as ED
# - Average CbCr is replaced with CbCr of the image
# - Generate images. Both G1 and G_cyclic
# - No need for losses
# ------------------------------------------------
"""
def test( self, args ):
self.comet_experiment = Experiment(
api_key = "insert your comet api key here",
project_name = "insert your project name here",
workspace = "insert your workspace here",
auto_param_logging = True,
auto_metric_logging = True,
log_env_details = True,
log_code = True, # code logging
log_graph = True,
log_env_gpu = True,
log_env_host = True,
log_env_cpu = True,
auto_histogram_tensorboard_logging = True,
auto_histogram_weight_logging = True,
auto_histogram_gradient_logging = True,
auto_histogram_activation_logging = True,
# auto_histogram_epoch_rate=1,
)
self.comet_experiment.add_tag("TEST RUN")
# Do not flip the image
self.random_flip = 0.0
# do not randomize target label values
self.TARGET_LABELS = 1.0
# Disable deleting by mistake
self.delete_old_checkpoints = False
# Step1: Load the test images
rootfolder = args.test_dir
testpath = pathlib.Path( rootfolder )
# NOTE: While loading the images, only difference is that the images are not flipped. Otherwise it is the same function as
# the dataset loading images
# Intialize array for saving each image's values
self.stddev_arr = []
self.mean_arr = []
self.variance_arr = []
test_images = tf.keras.preprocessing.image_dataset_from_directory(
str( testpath ),
labels = None,
# label_mode = 'categorical',
color_mode = 'rgb',
validation_split = None,
shuffle = False,
seed = 1337,
image_size = (self.image_size, self.image_size),
batch_size = 1
) \
.cache() \
.map(lambda x: (x / 255.0), num_parallel_calls=tf.data.AUTOTUNE ) \
.prefetch(25)
# .map(lambda x: tf.image.per_image_standardization( x ) ) \
# .map(lambda x: x if self.random_flip else tf.image.flip_up_down( x ), num_parallel_calls=tf.data.AUTOTUNE) \
# .map(lambda x: ((x / 127.5) - 1 ), num_parallel_calls=tf.data.AUTOTUNE ) \
test_images.class_names = 'TEST'
# Only load diffuse images if the flag is True
if args.calc_metrics is True:
diffusefolder = args.diffuse_dir
diffusepath = pathlib.Path( diffusefolder )
# NOTE: While loading the images, only difference is that the images are not flipped. Otherwise it is the same function as
# the dataset loading images
diffuse_images = tf.keras.preprocessing.image_dataset_from_directory(
str( diffusepath ),
labels = None,
# label_mode = 'categorical',
color_mode = 'rgb',
validation_split = None,
shuffle = False,
seed = 1337,
image_size = (self.image_size, self.image_size),
batch_size = 1
) \
.cache() \
.map(lambda x: (x / 255.0), num_parallel_calls=tf.data.AUTOTUNE ) \
.prefetch(25)
# .map(lambda x: tf.image.per_image_standardization( x ) ) \
# .map(lambda x: x if self.random_flip else tf.image.flip_up_down( x ), num_parallel_calls=tf.data.AUTOTUNE) \
# .map(lambda x: ((x / 127.5) - 1 ), num_parallel_calls=tf.data.AUTOTUNE ) \
test_images.class_names = 'TEST'
# return the number of files loaded
self.number_of_test_images = len(np.concatenate([i for i in test_images], axis=0))
# ZIP the datasets into one dataset
if args.calc_metrics is True:
loadedDataset = tf.data.Dataset.zip ( ( test_images, diffuse_images ) )
else:
loadedDataset = tf.data.Dataset.zip ( test_images )
options = tf.data.Options()
options.threading.max_intra_op_parallelism = 1
loadedDataset = loadedDataset.with_options(options)
loadedDataset = loadedDataset.cache().prefetch( buffer_size =25)
# Load the G and D
self.G = self.build_generator( )
self.D = self.build_discriminator( )
# Print Model summary to console and file
self.G.summary()
self.D.summary()
with open('Generator_summary.txt', 'w') as f:
self.G.summary(print_fn=lambda x: f.write(x + '\n'))
with open('Discriminator_summary.txt', 'w') as f:
self.D.summary(print_fn=lambda x: f.write(x + '\n'))
"""
----------------------------------------------
Adding SpecSeg to Test
----------------------------------------------
"""
self.SpecSeg = SpecSeg(self.image_size, self.image_size, 1)
self.SpecSeg = tf.keras.models.load_model('specsegv3_chkpt.h5',compile=False)
self.SpecSeg.summary()
with open('SpecSeg_summary.txt', 'w') as f:
self.SpecSeg.summary(print_fn=lambda x: f.write(x + '\n'))
# ------------------------------------------------
# STEP2: Load checkpoints
checkpoint_dir = self.checkpoint_save_dir
ckpt = tf.train.Checkpoint( generator = self.G,
discriminator = self.D,
optimizer_D = self.optimizer_D,
optimizer_G = self.optimizer_G )
ckpt_manager = tf.train.CheckpointManager(ckpt, checkpoint_dir, max_to_keep=3)
ckpt.restore(ckpt_manager.latest_checkpoint).expect_partial()
print ('Latest checkpoint restored!!')
plt.close("all")
gc.collect()
# STEP3: Iterate over the loaded test images
test_iterator = iter(loadedDataset)
# create zeros and ones for labels
tmp_zeros = tf.zeros( [1, self.image_size, self.image_size, 1], dtype=tf.float32 )
trg_ones = tf.ones( [1, self.image_size, self.image_size, 1], dtype=tf.float32 )
# initialize lists for printing data
MSE = []
SSIM = []
PSNR = []
index = []
table = []
delE76 = []
delE94 = []
processing_time_taken = []
print('\n\n\n->> "I\'m sorry, Dave. You will have to wait a little while I process... Regards, HAL 9000 ◍ <<- \n\n\n')
# for all images in the test folder
for i in range(self.number_of_test_images):
self.start_time = time.time()
# Randomly generate target labels for more robustness instead of a hard value of 1
self.TARGET_LABELS = tf.random.uniform( [], minval=0.8, maxval=1.2, dtype=tf.float32 )
# Get the image
element = test_iterator.get_next()
if args.calc_metrics is True:
self.rgb_testImage = element[0]
self.rgb_diffuseImage = element[1]
else:
self.rgb_testImage = element
# Setting target labels for Cyclic generation
self.target_label_ED = tf.Variable( [0,0,0,0,self.TARGET_LABELS], dtype=tf.float32 )
# setting both G and D as non-trainable
self.G.trainable = False
self.D.trainable = False
# setting ED as input image and other channels as zero
RGBInput = self.custom_per_image_standardization( tf.image.rgb_to_yuv( self.rgb_testImage[:, :, :, :] ) )
# Generating the specular mask from the input RGB image
self.specular_candidate = (self.SpecSeg.predict(RGBInput[:, :, :, 0, tf.newaxis], verbose=0))
# setting the CbCr same as the input image
averageCbCr = RGBInput[:, :, :, 1:]
# Y channel input are set to zero and the input is 0 degree
ych_inp1 = RGBInput[:, :, :, 0, tf.newaxis]
ych_inp2 = tmp_zeros
ych_inp3 = tmp_zeros
ych_inp4 = tmp_zeros
ych_inp5 = tmp_zeros
# generate the inputs
rand_input_Ych = tf.concat( [ych_inp1, ych_inp2, ych_inp3, ych_inp4, ych_inp5], axis = 3 )
self.gen_input = tf.concat( [rand_input_Ych, tmp_zeros, tmp_zeros, tmp_zeros, tmp_zeros, trg_ones], axis = 3 )
self.target_img = self.rgb_testImage
self.Target_angle_label = self.target_label_ED
# test plot the input
# debug_plot( self.gen_input )
"""--------------------G(1)-------------------"""
self.gen_Y = self.G ( self.gen_input, training=False )
self.gen_YCbCr = tf.concat( [self.gen_Y, averageCbCr], axis = 3 )
avg_stddev_arr = tf.reduce_mean(self.stddev_arr)
tf.reduce_mean(self.mean_arr)
# avg_variance_arr = tf.reduce_mean(self.variance_arr)
self.gen_rgb_output = tf.image.yuv_to_rgb( ( tf.multiply(self.gen_YCbCr, avg_stddev_arr) ) * 255.0 )
self.gen_rgb = tf.image.yuv_to_rgb( self.gen_YCbCr )
orig_Ych = self.gen_rgb[:, :, :, 0, tf.newaxis]
# plot_single_image ( self.gen_Y )
# plot_single_image ( self.gen_rgb, title="Generated RGB" )
"""--------------------G(2)-------------------"""
# -----------------CONFIG B -------------------
# NOTE: This gives **slightly** better results than CONFIG-A above
cyc_Y1 = tf.concat( [tmp_zeros, orig_Ych, orig_Ych, orig_Ych, orig_Ych], axis = 3 )
cyc_Y2 = tf.concat( [orig_Ych, tmp_zeros, orig_Ych, orig_Ych, orig_Ych], axis = 3 )
cyc_Y3 = tf.concat( [orig_Ych, orig_Ych, tmp_zeros, orig_Ych, orig_Ych], axis = 3 )
cyc_Y4 = tf.concat( [orig_Ych, orig_Ych, orig_Ych, tmp_zeros, orig_Ych], axis = 3 )
cyc_Y5 = tf.concat( [orig_Ych, orig_Ych, orig_Ych, orig_Ych, tmp_zeros], axis = 3 )
# ####################################################################################
# Append one-hot tensors as labels
# NOTE: Use ZEROS as input; same as training.
# -----------------CONFIG A -------------------
cyclic_input1 = tf.concat( [cyc_Y1, trg_ones, tmp_zeros, tmp_zeros, tmp_zeros, tmp_zeros], axis = 3 )
cyclic_input2 = tf.concat( [cyc_Y2, tmp_zeros, trg_ones, tmp_zeros, tmp_zeros, tmp_zeros], axis = 3 )
cyclic_input3 = tf.concat( [cyc_Y3, tmp_zeros, tmp_zeros, trg_ones, tmp_zeros, tmp_zeros], axis = 3 )
cyclic_input4 = tf.concat( [cyc_Y4, tmp_zeros, tmp_zeros, tmp_zeros, trg_ones, tmp_zeros], axis = 3 )
cyclic_input5 = tf.concat( [cyc_Y5, tmp_zeros, tmp_zeros, tmp_zeros, tmp_zeros, trg_ones], axis = 3 )
# debug_plot( cyclic_input5 )
# Generate 5 Y channel CYCLIC images, to reconstruct the original images.
cyc_0_Y = self.G( cyclic_input1, training=False )
cyc_45_Y = self.G( cyclic_input2, training=False )
cyc_90_Y = self.G( cyclic_input3, training=False )
cyc_135_Y = self.G( cyclic_input4, training=False )
cyc_ED_Y = self.G( cyclic_input5, training=False )
cyc_gen0 = tf.concat( [ cyc_0_Y, averageCbCr], axis = 3 )
cyc_gen45 = tf.concat( [ cyc_45_Y, averageCbCr], axis = 3 )
cyc_gen90 = tf.concat( [ cyc_90_Y, averageCbCr], axis = 3 )
cyc_gen135 = tf.concat( [ cyc_135_Y, averageCbCr], axis = 3 )
cyc_genED = tf.concat( [ cyc_ED_Y, averageCbCr], axis = 3 )
# convert to RGB
self.cyc_gen0_rgb = tf.image.yuv_to_rgb( cyc_gen0 )
self.cyc_gen45_rgb = tf.image.yuv_to_rgb( cyc_gen45 )
self.cyc_gen90_rgb = tf.image.yuv_to_rgb( cyc_gen90 )
self.cyc_gen135_rgb = tf.image.yuv_to_rgb( cyc_gen135 )
self.cyc_genED_rgb = tf.image.yuv_to_rgb( cyc_genED )
processing_time_taken.append( (time.time() - self.start_time) )
# self.test_plot()
# ---------------------- COMET Logging ------------------------
# Plotting output of G1
self.comet_experiment.log_image( tf.squeeze( (self.gen_Y) ), name="G1 Y-ch", step=i)
self.comet_experiment.log_image( tf.squeeze((self.gen_rgb)), name="G1 RGB", step=i)
self.comet_experiment.log_image( tf.squeeze(self.target_img), name="1. Input Image ", step=i)
self.comet_experiment.log_image( tf.squeeze( (self.cyc_gen0_rgb) ), name="G2 0°", step=i)
self.comet_experiment.log_image( tf.squeeze( (self.cyc_gen45_rgb) ), name="G2 45°", step=i)
self.comet_experiment.log_image( tf.squeeze( (self.cyc_gen90_rgb) ), name="G2 90°", step=i)
self.comet_experiment.log_image( tf.squeeze( (self.cyc_gen135_rgb) ), name="G2 135°", step=i)
self.comet_experiment.log_image( tf.squeeze( (self.cyc_genED_rgb) ), name="G2 ED", step=i)
self.comet_experiment.log_image( tf.squeeze(self.specular_candidate), name="Specular Mask ", step=i)
if args.calc_metrics is True:
self.comet_experiment.log_image( tf.squeeze(self.rgb_diffuseImage), name="2. Target Diffuse ", step=i)
# ---------------------- PyPlt printing ------------------------
# image_grid( self.cyc_gen0_rgb, self.cyc_gen45_rgb, self.cyc_gen90_rgb, self.cyc_gen135_rgb, self.cyc_genED_rgb )
# plot the generated images :fingerscrossed:
# plot_single_image ( self.cyc_gen0_rgb )
# plot_single_image ( self.cyc_gen45_rgb )
# plot_single_image ( self.cyc_gen90_rgb )
# plot_single_image ( self.cyc_gen135_rgb )
# plot_single_image ( self.cyc_genED_rgb )
# plt.close("all")
# gc.collect()
# ----------------- calculating Metrics -------------------
# calculate only if the flag is true
if args.calc_metrics is True:
index.append(i+1)
# FID_score = self.calculate_FID( self.cyc_genED_rgb , self.target_img )
SSIM.append( (tf.image.ssim ( rescale_01( self.gen_rgb ), rescale_01( self.rgb_diffuseImage ), 5 )).numpy() )
im1 = tf.image.convert_image_dtype(self.gen_rgb, tf.float32)
im2 = tf.image.convert_image_dtype(self.rgb_diffuseImage, tf.float32)
# psnr2 = tf.image.psnr(im1, im2, max_val=1.0)
# PSNR.append( (tf.image.psnr ( tf.clip_by_value(self.gen_rgb, 0, 255), tf.clip_by_value(self.rgb_diffuseImage, 0, 255), max_val=255 )).numpy() )
PSNR.append( (tf.image.psnr(im1 , im2 , max_val=1.0)).numpy() )
# Calculate L1 loss to original image?
# Or use builtin functions to evaluate the Generator?
L2_loss = tf.keras.losses.MeanSquaredError()
MSE.append( L2_loss(self.gen_rgb, self.rgb_diffuseImage ).numpy() )
# print ( 'Processing Image# {}: {:.3f} secs, MSE:{:.4f}, SSIM:{:.4f}, PSNR:{:.4f} \n' .format( i, processing_time_taken[i], MSE[i], SSIM[i], PSNR[i]) )
img1 = tfio.experimental.color.rgb_to_lab( tf.image.convert_image_dtype(self.gen_rgb, dtype=tf.float32) )
img2 = tfio.experimental.color.rgb_to_lab( tf.image.convert_image_dtype(self.rgb_diffuseImage, dtype=tf.float32) )
delE76.append ( tf.reduce_mean(tf.py_function(deltaE_cie76, inp=[img1,img2], Tout=tf.float32)) )
delE94.append ( tf.reduce_mean(tf.py_function(deltaE_ciede94, inp=[img1,img2], Tout=tf.float32)) )
# populate table
column = [index[i], processing_time_taken[i], MSE[i], SSIM[i], PSNR[i], delE76[i], delE94[i]]
table.append(column)
# # print table inline
# print( tabulate( table, tablefmt="plain" ))
self.comet_experiment.log_metric ( "Processing Time", processing_time_taken[i], step=i )
self.comet_experiment.log_metric ( "MSE", MSE[i], step=i )
self.comet_experiment.log_metric ( "SSIM", SSIM[i], step=i )
self.comet_experiment.log_metric ( "PSNR", PSNR[i], step=i )
# Print metrics only if flag is true
if args.calc_metrics is True:
print('\n\n --- PRINTING ALL CALCUATED METRICS --- ')
print(tabulate(table, headers=['Image#', 'Time', 'MSE', 'SSIM', 'PSNR', 'delE76', 'delE94']))
# Calculating mean values
mean_mse = sum(MSE) / len(MSE)
mean_ssim = sum(SSIM) / len(SSIM)
mean_psnr = sum(PSNR) / len(PSNR)
mean_delE76 = sum(delE76) / len(delE76)
mean_delE94 = sum(delE94) / len(delE94)
print('\n\n --- PRINTING MEAN METRICS --- ')
mean_metrics = [mean_mse, mean_ssim, mean_psnr, mean_delE76, mean_delE94]
print(tabulate([mean_metrics], headers=['Mean MSE', 'Mean SSIM', 'Mean PSNR', 'Mean dleE76', 'Mean delE94']))
print('\n\n' )
# saving all the calculated metrics as txt
with open("SSIM.txt", 'wb+') as file1:
pickle.dump(SSIM, file1)
with open("MSE.txt", 'wb+') as file2:
pickle.dump(MSE, file2)
with open("PSNR.txt", 'wb+') as file3:
pickle.dump(PSNR, file3)
# logging means to Comet also before closing experiment
# self.comet_experiment.log_other( value = MSE, key="All MSE")
# self.comet_experiment.log_other( value = SSIM, key="All SSIM")
# self.comet_experiment.log_other( value = PSNR, key="All PSNR")
# self.comet_experiment.log_other( value = mean_mse, key="Mean MSE")
# self.comet_experiment.log_other( value = mean_ssim, key="Mean SSIM")
# self.comet_experiment.log_other( value = mean_psnr, key="Mean PSNR")
self.comet_experiment.end()
print('\n\n\n->> "Thank you for a very enjoyable game - HAL 9000 ◍ <<- \n\n\n')
return
# ------------------------------------------
# PLOTTING TEST IMAGES POST TRAINING
def test_plot( self ):
figure = plt.figure( figsize=(10,15) )
figure.add_subplot( 2, 1, 1, title="Orig")
plt.imshow ( tf.squeeze( rescale_01( self.rgb_testImage ) ).numpy().astype("float32") )
figure.add_subplot( 2, 1, 2, title="Generated G1")
plt.imshow ( tf.squeeze( rescale_01( self.gen_rgb ) ).numpy().astype("float32") )
figure = plt.figure( figsize=(10,15) )
figure.add_subplot( 2, 2, 1, title="Cyc0")
plt.imshow ( tf.squeeze( rescale_01( self.cyc_gen0_rgb ) ).numpy().astype("float32") )
figure.add_subplot( 2, 2, 2, title="Cyc45")
plt.imshow ( tf.squeeze( rescale_01( self.cyc_gen45_rgb ) ).numpy().astype("float32") )
figure.add_subplot( 2, 2, 3, title="Cyc90")
plt.imshow ( tf.squeeze( rescale_01( self.cyc_gen90_rgb ) ).numpy().astype("float32") )
figure.add_subplot( 2, 2, 4, title="Cyc135")
plt.imshow ( tf.squeeze( rescale_01( self.cyc_gen135_rgb ) ).numpy().astype("float32") )