-
Notifications
You must be signed in to change notification settings - Fork 0
/
data.py
303 lines (236 loc) · 10.1 KB
/
data.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
from PIL import Image
from matplotlib.animation import FuncAnimation
import matplotlib.colors as mcolors
import matplotlib.pyplot as plt
import numpy as np
import os
import subprocess
import sys
import torch
from torch.utils.data import DataLoader
import torchvision
from data_opts import DATA_OPTS
DATA_NAMES = list(DATA_OPTS.keys())
class Data:
def __init__(self, name, batch_trn=256, batch_tst=32,
norm_m=None, norm_v=None, root='result'):
if not name in DATA_NAMES:
raise ValueError(f'Dataset name "{name}" is not supported')
self.name = name
self.opts = DATA_OPTS[self.name]
self.root = root
self.batch_trn = batch_trn
self.batch_tst = batch_tst
self.norm_m = norm_m
if self.norm_m is None:
self.norm_m = self.opts.get('norm_m')
self.norm_v = norm_v
if self.norm_v is None:
self.norm_v = self.opts.get('norm_v')
self.labels = self.opts.get('labels', {})
self.sz = self.opts['sz']
self.ch = self.opts['ch']
self._set_transform()
self._load()
def animate(self, X, titles, size=3, interval=6, fps=0.7, fpath=None):
if X is None or len(X) == 0 or len(X) != len(titles):
print('WRN: invalid data for animation')
return
def prep(x):
x = self.tr_norm_inv(x).detach().to('cpu').squeeze().numpy()
x = np.clip(x, 0, 1) if np.mean(x) < 2 else np.clip(x, 0, 255)
x = x.transpose(1, 2, 0) if len(x.shape) == 3 else x
return x
fig = plt.figure(figsize=(size, size))
ax = fig.add_subplot(111)
ax.axis('off')
img = ax.imshow(prep(X[0]))
def update(k, *args):
ax.set_title(titles[k], fontsize=9)
img.set_data(prep(X[k]))
return (img,)
anim = FuncAnimation(fig, update, interval=interval,
frames=len(X), blit=True, repeat=False)
if fpath:
anim.save(fpath, writer='pillow', fps=fps)
plt.close(fig)
def get(self, i=None, tst=False):
data = self.data_tst if tst else self.data_trn
if i is None:
i = torch.randint(len(data), size=(1,)).item()
x, c = data[i]
l = self.labels.get(c)
return x, c, l
def img_load(self, fpath, device='cpu', resize=False, norm=True):
x = torchvision.io.read_image(fpath)
x = torchvision.transforms.ConvertImageDtype(torch.float32)(x)
x = x.expand(self.ch, *x.shape[1:]) if x.shape[0] < self.ch else x
x = self.tr_size(x) if resize else x
x = self.tr_norm(x) if norm else x
return x.to(device)
def img_rand(self, device='cpu', norm=True):
pix = np.random.rand(self.sz, self.sz, self.ch) * 255
img = Image.fromarray(pix.astype('uint8')).convert('RGB')
x = self.tr_tens(x)
x = self.tr_norm(x) if norm else x
return x.to(device)
def info(self):
text = ''
text += f'Dataset : {self.name}\n'
text += f'Number of classes : {len(self.labels):-10d}\n'
if self.data_trn is not None:
text += f'Size of trn dataset : {len(self.data_trn):-10d}\n'
if self.data_tst is not None:
text += f'Size of tst dataset : {len(self.data_tst):-10d}\n'
return text
def plot(self, x, title='', fpath=None, is_new=True, do_tr=True):
x = self.tr_norm_inv(x) if do_tr else x
size = self.opts['plot_size']
cmap = self.opts['plot_cmap']
return self.plot_base(x, title, size, cmap, fpath, is_new)
def plot_attr(self, x, size=12, fpath=None):
x = torch.tensor(x) if not torch.is_tensor(x) else x
x = x.detach().to('cpu').squeeze().numpy()
x = np.clip(x, 0, 1) if np.mean(x) < 2 else np.clip(x, 0, 255)
fig = plt.figure(figsize=(size, size))
plt.imshow(x)
plt.axis('off')
if fpath:
plt.savefig(fpath, bbox_inches='tight')
plt.close(fig)
else:
plt.show()
def plot_base(self, x, title, size=3, cmap='hot', fpath=None, is_new=True):
if torch.is_tensor(x):
x = x.detach().to('cpu').squeeze().numpy()
if len(x.shape) == 3:
x = x.transpose(1, 2, 0)
x = np.clip(x, 0, 1) if np.mean(x) < 2 else np.clip(x, 0, 255)
if is_new:
fig = plt.figure(figsize=(size, size))
plt.imshow(x, cmap=cmap)
if title:
plt.title(title, fontsize=9)
plt.axis('off')
if fpath:
plt.savefig(fpath, bbox_inches='tight')
elif is_new:
plt.show()
if is_new:
plt.close(fig)
def plot_changes(self, x, fpath=None):
fig = plt.figure(figsize=(12, 12))
plt.imshow(x, cmap=mcolors.ListedColormap(['white', 'black', 'red']))
plt.axis('off')
plt.gca().set_facecolor('black')
if fpath:
plt.savefig(fpath, bbox_inches='tight')
plt.close(fig)
else:
plt.show()
def plot_many(self, X=None, titles=None, cols=4, rows=4, size=3,
fpath=None, wspace=0.3, hspace=0.3):
fig = plt.figure(figsize=(size*cols, size*rows))
plt.subplots_adjust(wspace=wspace, hspace=hspace)
for j in range(1, cols * rows + 1):
if X is None:
i = torch.randint(len(self.data_tst), size=(1,)).item()
x, c, l = self.get(i, tst=True)
title = l[:17] + '...' if len(l) > 20 else l
else:
x = X[j-1].detach().to('cpu')
title = titles[j-1] if titles else ''
fig.add_subplot(rows, cols, j)
self.plot(x, title, is_new=False)
plt.savefig(fpath, bbox_inches='tight') if fpath else plt.show()
plt.close(fig)
def _load(self):
self.data_trn = None
self.data_tst = None
self.dataloader_trn = None
self.dataloader_tst = None
fpath = os.path.join(self.root, '_data', self.name)
load = not os.path.isdir(fpath)
os.makedirs(fpath, exist_ok=True)
if self.opts.get('dataset'):
func = eval(f'torchvision.datasets.{self.opts["dataset"]}')
self.data_trn = func(root=fpath, train=True,
download=load, transform=self.tr)
self.data_tst = func(root=fpath, train=False,
download=load, transform=self.tr)
self.dataloader_trn = DataLoader(self.data_trn,
batch_size=self.batch_trn, shuffle=True)
self.dataloader_tst = DataLoader(self.data_tst,
batch_size=self.batch_tst, shuffle=True)
if self.opts.get('repo'):
# TODO: add support for trn/tst repo
if load:
load_repo(self.opts['repo'], fpath)
repo = self.opts['repo'].split('.git')[0].split('/')[-1]
fpath = os.path.join(fpath, repo)
# Below is the code to parse IMAGENET images from the repo:
# https://github.com/EliSchwartz/imagenet-sample-images.git
l_rep = {}
for f in os.listdir(fpath):
if not f.endswith('JPEG'):
continue
l = ' '.join(f.split('.JPEG')[0].split('_')[1:])
l = l.lower().replace("'", '`')
c = None
is_found = False
for c_real, l_real in self.labels.items():
if l == l_real.split(',')[0]:
if is_found:
l_rep[l] = 0
else:
if not l in l_rep or l_rep[l] == 1:
c = c_real
is_found = True
else:
l_rep[l] = 1
if c is None:
raise ValueError
path_old = os.path.join(fpath, f)
path_new = os.path.join(fpath, f'{c}.jpg')
os.rename(path_old, path_new)
class Dataset(torch.utils.data.Dataset):
def __init__(self, fpath, labels, transform):
self.fpath = fpath
self.labels = labels
self.transform = transform
def __len__(self):
return len(self.labels)
def __getitem__(self, i):
img_path = os.path.join(fpath, f'{i}.jpg')
x = torchvision.io.read_image(img_path)
x = torchvision.transforms.ConvertImageDtype(
torch.float32)(x)
if x.shape[0] == 1:
x = x.expand(3, *x.shape[1:])
x = self.transform(x)
return x, i
tr = torchvision.transforms.Compose([self.tr_size, self.tr_norm])
self.data_tst = Dataset(fpath, self.labels, transform=tr)
self.dataloader_tst = DataLoader(self.data_tst,
batch_size=self.batch_tst, shuffle=True)
def _set_transform(self):
self.tr_tens = torchvision.transforms.ToTensor()
self.tr_size = torchvision.transforms.Compose([
torchvision.transforms.Resize(self.sz),
torchvision.transforms.CenterCrop(self.sz)])
self.tr_norm = lambda x: x
self.tr_norm_inv = lambda x: x
if self.norm_m is not None and self.norm_v is not None:
self.tr_norm = torchvision.transforms.Normalize(
self.norm_m, self.norm_v)
self.tr_norm_inv = torchvision.transforms.Compose([
torchvision.transforms.Normalize(
[0., 0., 0.], 1./np.array(self.norm_v)),
torchvision.transforms.Normalize(
-np.array(self.norm_m), [1., 1., 1.]),
])
# TODO: see https://github.com/pytorch/vision/blob/f69eee6108cd047ac8b62a2992244e9ab3c105e1/torchvision/transforms/_presets.py#L38
self.tr = torchvision.transforms.Compose([self.tr_tens, self.tr_norm])
def load_repo(url, fpath):
prc = subprocess.getoutput(f'cd {fpath} && git clone {url}')
print(prc)