-
Notifications
You must be signed in to change notification settings - Fork 0
/
preprocessing.py
395 lines (330 loc) · 12.1 KB
/
preprocessing.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
###############################################################################
#Copyright (C) 2017 Michael O. Vertolli michaelvertolli@gmail.com
#
#This program is free software: you can redistribute it and/or modify
#it under the terms of the GNU General Public License as published by
#the Free Software Foundation, either version 3 of the License, or
#(at your option) any later version.
#
#This program is distributed in the hope that it will be useful,
#but WITHOUT ANY WARRANTY; without even the implied warranty of
#MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
#GNU General Public License for more details.
#
#You should have received a copy of the GNU General Public License
#along with this program. If not, see http://www.gnu.org/licenses/
###############################################################################
from evaluations import flatten2color
from glob import glob
import json
from nltk.corpus import wordnet as wn
import numpy as np
import os
from PIL import Image
from random import randint, shuffle
import tensorflow as tf
import xml.etree.ElementTree as ET
def get_hyponyms(syn):
hypo = set()
for h in syn.hyponyms():
hypo |= set(get_hyponyms(h))
return hypo | set(syn.hyponyms())
def get_syn_imgs(syn, dir_):
dirs = os.listdir(dir_)
hypos = get_hyponyms(syn)
wnids = dict([('n{:08}'.format(s.offset()), True) for s in hypos])
out = []
for d in dirs:
try:
temp = wnids[d.split('_')[0]]
except KeyError:
continue
else:
out.append(d)
return out
def get_bbox(file_, from_dir):
wnid = file_.split('_')[0]
xml = file_[:-4]+'xml'
path = os.path.join(from_dir, wnid, xml)
try:
root = ET.parse(path).getroot()
except IOError:
out = None
else:
box = root[5][4]
min_ = [int(box[0].text), int(box[1].text)]
max_ = [int(box[2].text), int(box[3].text)]
size = root[3]
img_size = [int(size[0].text), int(size[1].text)]
out = min_, max_, img_size
return out
def square_bbox(min_, max_, size):
bsize = [max_[0]-min_[0], max_[1]-min_[1]]
if bsize[0] < bsize[1]:
i = 0
diff = bsize[1] - bsize[0]
else:
i = 1
diff = bsize[0] - bsize[1]
if diff+bsize[i] > size[i]:
out = None
else:
if diff%2 == 0:
ltdiff = diff/2
rbdiff = ltdiff
else:
ltdiff = diff/2
rbdiff = ltdiff+1
mn = [x for x in min_]
mx = [x for x in max_]
mn[i] -= ltdiff
mx[i] += rbdiff
if mn[i] < 0:
mx[i] -= mn[i]
mn[i] = 0
if mx[i] > size[i]:
mn[i] -= mx[i] - size[i]
mx[i] = size[i]
out = [mn[0], mn[1], mx[0], mx[1]]
return out
def save_bbox_crop_imgs(syn, imgs_dir, annt_dir, out_dir):
files = get_syn_imgs(syn, imgs_dir)
for f in files:
bbox = get_bbox(f, annt_dir)
if bbox is None:
continue
min_, max_, size = bbox
crop_pos = square_bbox(min_, max_, size)
if crop_pos is None:
continue
im = Image.open(os.path.join(imgs_dir, f))
crop = im.crop(crop_pos)
resized = crop.resize((256, 256), Image.BICUBIC)
resized.save(os.path.join(out_dir, f))
im.close()
def symlink_imgs(files, from_dir, to_dir):
for f in files:
os.symlink(os.path.join(from_dir, f), os.path.join(to_dir, f))
def imgnetmain():
item = wn.synsets('animal')[0]
imgs_dir = './train/'
annt_dir = '/home/olias/data/imgnet/annotations/'
out_dir = '/home/olias/data/imgnet_animal/splits/train/'
save_bbox_crop_imgs(item, imgs_dir, annt_dir, out_dir)
def crop_resize(imgs_dir, out_dir):
files = os.listdir(imgs_dir)
for f in files:
im = Image.open(os.path.join(imgs_dir, f))
h, w = im.size
if h > w:
crop = im.crop((0, 0, w, w))
else:
crop = im.crop((0, 0, h, h))
resized = crop.resize((128, 128), Image.BICUBIC)
resized.save(os.path.join(out_dir, f))
im.close()
def crop_resize2(imgs_dir, out_dir, crop_box, resize):
if not os.path.exists(out_dir):
os.makedirs(out_dir)
files = os.listdir(imgs_dir)
for f in files:
im = Image.open(os.path.join(imgs_dir, f))
h, w = im.size
if h > w:
crop = im.crop(crop_box)
else:
crop = im.crop(crop_box)
if resize is not None:
resized = crop.resize(resize, Image.BICUBIC)
else:
resized = crop
resized.save(os.path.join(out_dir, f))
im.close()
def mscelebmain():
crop_resize('/home/olias/data/msceleb/MsCeleb', '/home/olias/data/msceleb/splits/train')
def dataset_from_img(img_file, dataset_folder, dataset_size, out_img_size):
im = Image.open(img_file)
w, h = im.size
w -= (out_img_size + 1)
h -= (out_img_size + 1)
for i in range(dataset_size):
w_ = randint(0, w)
h_ = randint(0, h)
crop = im.crop((w_, h_, w_+out_img_size, h_+out_img_size))
crop.save(os.path.join(dataset_folder, '{:07}.jpg'.format(i)))
im.close()
def dataset_resize(imgs_dir, new_dir, new_size):
if not os.path.exists(new_dir):
os.makedirs(new_dir)
files = os.listdir(imgs_dir)
for f in files:
im = Image.open(os.path.join(imgs_dir, f))
o_im = im.resize([new_size, new_size], Image.NEAREST)
o_im.save(os.path.join(new_dir, f))
im.close()
def celeb_to_imgs(new_dir, imgs_dir='/home/olias/data/img_align_celeba/', train=True, all_dir=None):
if all_dir is None:
dirs = [new_dir]
else:
dirs = [new_dir, all_dir]
for dr in dirs:
if not os.path.exists(dr):
os.makedirs(dr)
TRAIN_STOP = 162770
NUM_EXAMPLES = 202599
CROP_BOX = [25, 50, 128+25, 50+128]
if train:
start, stop = 0, TRAIN_STOP
else:
start, stop = TRAIN_STOP, NUM_EXAMPLES # collapsedd validation to test
files = ['{:06}.jpg'.format(i+1) for i in range(start, stop)]
for f in files:
im = Image.open(os.path.join(imgs_dir, f))
crop = im.crop(CROP_BOX)
for dr in dirs:
crop.save(os.path.join(dr, f))
im.close()
def mnist_to_imgs(new_dir, base_size=32, train=True, all_dir=None):
if all_dir is None:
dirs = [new_dir]
else:
dirs = [new_dir, all_dir]
for dr in dirs:
if not os.path.exists(dr):
os.makedirs(dr)
mnist = tf.contrib.learn.datasets.load_dataset('mnist')
if train:
images = mnist.train.images
labels = mnist.train.labels
base_index = 0
else:
images = mnist.test.images
labels = mnist.test.labels
base_index = 55000
labels = np.asarray(labels, dtype=np.int32)
for i in range(images.shape[0]):
arr = np.reshape(images[i, :], [28, 28])
im = Image.fromarray(arr*255, 'I').convert('RGB')
o_im = im.resize([base_size, base_size], Image.NEAREST)
label = labels[i]
for dr in dirs:
o_im.save(os.path.join(dr, '{}_{:05}.jpg'.format(label, base_index+i)))
def cifar10_to_imgs(new_dir, train=True, all_dir=None):
if all_dir is None:
dirs = [new_dir]
else:
dirs = [new_dir, all_dir]
for dr in dirs:
if not os.path.exists(dr):
os.makedirs(dr)
if train:
files = glob('/home/olias/data/cifar-10-batches-bin/data_batch*.bin')
count = 0
else:
files = ['/home/olias/data/cifar-10-batches-bin/test_batch.bin']
count = 50000
label_bytes = 1
h, w, c = 32, 32, 3
image_bytes = h * w * c
# convert = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']
# convert = dict([(i, s) for i, s in enumerate(convert)])
for f in files:
with open(f, 'rb') as stream:
label_i = stream.read(label_bytes)
while label_i != '':
label_i = np.frombuffer(label_i, dtype=np.uint8)[0]
# label = convert[label_i]
img = np.frombuffer(stream.read(image_bytes), dtype=np.int8)
img = np.transpose(np.reshape(img, [c, w, h]), [1, 2, 0]).astype(np.uint8)
img = Image.fromarray(img)
for dr in dirs:
img.save(os.path.join(dr, '{}_{:05}.jpg'.format(label_i, count)))
count += 1
label_i = stream.read(label_bytes)
def _bytes_feature(value):
return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))
def _int64_feature(value):
return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))
def to_tfrecord_shard(fname, new_dir, img_dir, shape, shards=25): # assumes shape is [h, w, c]
if not os.path.exists(new_dir):
os.makedirs(new_dir)
with open(os.path.join(new_dir, 'img_shape.json'), 'w') as f:
f.write(json.dumps(shape))
imgs = glob(os.path.join(img_dir, '*.jpg'))
shuffle(imgs)
nimgs_pershard = len(imgs) / shards
imgs = imgs[:nimgs_pershard*shards] # cuts off remainder < num shards
print '{} images kept after filtering.'.format(len(imgs))
shardstart_i = 0
shard_i = 0
writer = tf.python_io.TFRecordWriter(os.path.join(new_dir, fname+'_{:02}.tfrecords'.format(shard_i)))
for i, img in enumerate(imgs):
if (i - shardstart_i) == nimgs_pershard:
writer.close()
shardstart_i = i
shard_i += 1
writer = tf.python_io.TFRecordWriter(os.path.join(new_dir, fname+'_{:02}.tfrecords'.format(shard_i)))
with open(img, 'rb') as f:
im_raw = f.read()
example = tf.train.Example(features=tf.train.Features(feature={
'index': _int64_feature(i),
'image_raw': _bytes_feature(im_raw)}))
writer.write(example.SerializeToString())
writer.close()
def to_tfrecord(fname, new_dir, img_dir, shape, shuffle=False): # assumes shape is [h, w, c]
if not os.path.exists(new_dir):
os.makedirs(new_dir)
with open(os.path.join(new_dir, 'img_shape.json'), 'w') as f:
f.write(json.dumps(shape))
imgs = glob(os.path.join(img_dir, '*.jpg'))
if shuffle:
shuffle(imgs)
writer = tf.python_io.TFRecordWriter(os.path.join(new_dir, fname+'.tfrecords'))
for i, img in enumerate(imgs):
with open(img, 'rb') as f:
im_raw = f.read()
example = tf.train.Example(features=tf.train.Features(feature={
'image_raw': _bytes_feature(im_raw),
'image_name': _bytes_feature(os.path.split(img)[1])}))
writer.write(example.SerializeToString())
writer.close()
#Param search
def strip_extra_headers(ar):
new = []
for i in range(ar.shape[0]):
if not np.isnan(ar[i][0]):
new.append(ar[i])
return np.stack(new)
def to_distance(ar, start_index, end_index):
new = np.copy(ar)
temp = new[:, start_index:end_index]
new[:, start_index:end_index] = np.ones(temp.shape) - temp
return new
def nan_to_val(ar, nan_val):
new = np.copy(ar)
return new
def prep_param_csv(fname, start_index, end_index=14, nan_val=-.1, log=True):
path, tail = os.path.split(fname)
with open(fname, 'r') as f:
header = f.readline()
ar = np.genfromtxt(f, np.float32, delimiter=',')
ar = strip_extra_headers(ar)
# ar = to_distance(ar, start_index, end_index)
if log:
ar = -np.log10(ar+1e-10)
if start_index == 3:
ar[ar < 0.] = 10**(-ar[ar < 0.])
ar[ar > 9.] = 0.
ar[np.isnan(ar)] = nan_val
np.savetxt(os.path.join(path, 'prepped_'+tail), ar, fmt='%.4f', delimiter=',', header=header[:-1], comments='')
def get_all_colors(log_dir, size=[32, 32], quantize=4, all_colors=set([])):
files = os.listdir(log_dir)
for f in files:
im = Image.open(os.path.join(log_dir, f))
if size is not None:
im = im.resize(size)
im = np.array(im)
if quantize is not None:
im = (im / quantize) * quantize
all_colors |= set([tuple(c) for c in flatten2color(im).tolist()])
return all_colors