forked from assafshocher/InGAN
-
Notifications
You must be signed in to change notification settings - Fork 1
/
util.py
334 lines (265 loc) · 15 KB
/
util.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
import numpy as np
from PIL import Image
from matplotlib import pyplot as plt, gridspec
import os
import glob
from time import strftime, localtime
from shutil import copy
from scipy.misc import imresize
import torch
def read_data(conf):
input_images = [read_shave_tensorize(path, conf.must_divide) for path in conf.input_image_path]
return input_images
def read_shave_tensorize(path, must_divide):
input_np = (np.array(Image.open(path).convert('RGB')) / 255.0)
input_np_shaved = input_np[:(input_np.shape[0] // must_divide) * must_divide,
:(input_np.shape[1] // must_divide) * must_divide,
:]
input_tensor = im2tensor(input_np_shaved)
return input_tensor
def tensor2im(image_tensors, imtype=np.uint8):
if not isinstance(image_tensors, list):
image_tensors = [image_tensors]
image_numpys = []
for image_tensor in image_tensors:
# Note that tensors are shifted to be in [-1,1]
image_numpy = image_tensor.detach().cpu().float().numpy()
if np.ndim(image_numpy) == 4:
image_numpy = image_numpy.transpose((0, 2, 3, 1))
image_numpy = np.round((image_numpy.squeeze(0) + 1) / 2.0 * 255.0)
image_numpys.append(image_numpy.astype(imtype))
if len(image_numpys) == 1:
image_numpys = image_numpys[0]
return image_numpys
def im2tensor(image_numpy, int_flag=False):
# the int flag indicates whether the input image is integer (and [0,255]) or float ([0,1])
if int_flag:
image_numpy /= 255.0
# Undo the tensor shifting (see tensor2im function)
transformed_image = np.transpose(image_numpy, (2, 0, 1)) * 2.0 - 1.0
return torch.FloatTensor(transformed_image).unsqueeze(0).cuda()
def random_size(orig_size, curriculum=True, i=None, iter_for_max_range=None, must_divide=8.0,
min_scale=0.25, max_scale=2.0, max_transform_magniutude=0.3):
cur_max_scale = 1.0 + (max_scale - 1.0) * np.clip(1.0 * i / iter_for_max_range, 0, 1) if curriculum else max_scale
cur_min_scale = 1.0 + (min_scale - 1.0) * np.clip(1.0 * i / iter_for_max_range, 0, 1) if curriculum else min_scale
cur_max_transform_magnitude = (max_transform_magniutude * np.clip(1.0 * i / iter_for_max_range, 0, 1)
if curriculum else max_transform_magniutude)
# set random transformation magnitude. scalar = affine, pair = homography.
random_affine = -cur_max_transform_magnitude + 2 * cur_max_transform_magnitude * np.random.rand(2)
# set new size for the output image
new_size = np.array(orig_size) * (cur_min_scale + (cur_max_scale - cur_min_scale) * np.random.rand(2))
return tuple(np.uint32(np.ceil(new_size * 1.0 / must_divide) * must_divide)), random_affine
def image_concat(g_preds, d_preds=None, size=None):
hsize = g_preds[0].shape[0] + 6 if size is None else size[0]
results = []
if d_preds is None:
d_preds = [None] * len(g_preds)
for g_pred, d_pred in zip(g_preds, d_preds):
# noinspection PyUnresolvedReferences
dsize = g_pred.shape[1] if size is None or size[1] is None else size[1]
result = np.ones([(1 + (d_pred is not None)) * hsize, dsize, 3]) * 255
if d_pred is not None:
d_pred_new = imresize((np.concatenate([d_pred] * 3, 2) - 128) * 2, g_pred.shape[0:2], interp='nearest')
result[hsize-g_pred.shape[0]:hsize+g_pred.shape[0], :g_pred.shape[1], :] = np.concatenate([g_pred,
d_pred_new], 0)
else:
result[hsize - g_pred.shape[0]:, :, :] = g_pred
results.append(np.uint8(np.round(result)))
return np.concatenate(results, 1)
def save_image(image_tensor, image_path):
image_pil = Image.fromarray(tensor2im(image_tensor), 'RGB')
image_pil.save(image_path)
def get_scale_weights(i, max_i, start_factor, input_shape, min_size, num_scales_limit, scale_factor):
num_scales = np.min([np.int(np.ceil(np.log(np.min(input_shape) * 1.0 / min_size)
/ np.log(scale_factor))), num_scales_limit])
# if i > max_i * 2:
# i = max_i * 2
factor = start_factor ** ((max_i - i) * 1.0 / max_i)
un_normed_weights = factor ** np.arange(num_scales)
weights = un_normed_weights / np.sum(un_normed_weights)
#
# np.clip(i, 0, max_i)
#
# un_normed_weights = np.exp(-((np.arange(num_scales) - (max_i - i) * num_scales * 1.0 / max_i) ** 2) / (2 * sigma ** 2))
# weights = un_normed_weights / np.sum(un_normed_weights)
return weights
class Visualizer:
def __init__(self, gan, conf, test_inputs):
self.gan = gan
self.conf = conf
self.G_loss = [None] * conf.max_iters
self.D_loss_real = [None] * conf.max_iters
self.D_loss_fake = [None] * conf.max_iters
self.test_inputs = test_inputs
self.test_input_sizes = [test_input.shape[2:] for test_input in test_inputs]
if conf.reconstruct_loss_stop_iter > 0:
self.Rec_loss = [None] * conf.max_iters
def recreate_fig(self):
self.fig = plt.figure(figsize=(18, 9))
gs = gridspec.GridSpec(8, 8)
self.result = self.fig.add_subplot(gs[0:8, 0:4])
self.gan_loss = self.fig.add_subplot(gs[0:2, 5:8])
self.reconstruct_loss = self.fig.add_subplot(gs[3:5, 5:8])
self.reconstruction = self.fig.add_subplot(gs[6:8, 5:6])
self.real_example = self.fig.add_subplot(gs[7, 6])
self.d_map_real = self.fig.add_subplot(gs[7, 7])
# First plot data
self.plot_gan_loss = self.gan_loss.plot([], [], 'b-',
[], [], 'c--',
[], [], 'r--')
self.gan_loss.legend(('Generator loss', 'Discriminator loss (real image)', 'Discriminator loss (fake image)'))
self.gan_loss.set_ylim(0, 1)
if self.conf.reconstruct_loss_stop_iter > 0:
self.plot_reconstruct_loss = self.reconstruct_loss.semilogy([], [])
# Set titles
self.gan_loss.set_title('Gan Losses')
self.reconstruct_loss.set_title('Reconstruction Loss')
self.reconstruction.set_title('Reconstruction')
self.d_map_real.set_xlabel('Current Discriminator \n map for real example')
self.real_example.set_xlabel('Real example')
self.result.set_title('Current result')
self.result.axes.get_xaxis().set_visible(False)
self.result.axes.get_yaxis().set_visible(False)
self.reconstruction.axes.get_xaxis().set_visible(False)
self.reconstruction.axes.get_yaxis().set_visible(False)
self.d_map_real.axes.get_yaxis().set_visible(False)
self.real_example.axes.get_yaxis().set_visible(False)
self.result.axes.get_yaxis().set_visible(False)
def test_and_display(self, i):
if not i % self.conf.print_freq and i > 0:
self.G_loss[i-self.conf.print_freq:i] = self.gan.losses_G_gan.detach().cpu().float().numpy().tolist()
self.D_loss_real[i-self.conf.print_freq:i] = self.gan.losses_D_real.detach().cpu().float().numpy().tolist()
self.D_loss_fake[i-self.conf.print_freq:i] = self.gan.losses_D_fake.detach().cpu().float().numpy().tolist()
if self.conf.reconstruct_loss_stop_iter > i:
self.Rec_loss[i-self.conf.print_freq:i] = self.gan.losses_G_reconstruct.detach().cpu().float().numpy().tolist()
if self.conf.reconstruct_loss_stop_iter < i:
print('iter: %d, G_loss: %f, D_loss_real: %f, D_loss_fake: %f, LR: %f' %
(i, self.G_loss[i-1], self.D_loss_real[i-1], self.D_loss_fake[i-1],
self.gan.lr_scheduler_G.get_lr()[0]))
else:
print('iter: %d, G_loss: %f, D_loss_real: %f, D_loss_fake: %f, Rec_loss: %f, LR: %f' %
(i, self.G_loss[i-1], self.D_loss_real[i-1], self.D_loss_fake[i-1], self.Rec_loss[i-1],
self.gan.lr_scheduler_G.get_lr()[0]))
if not i % self.conf.display_freq and i > 0:
plt.gcf().clear()
plt.close()
self.recreate_fig()
# choice = np.random.randint(0, len(self.test_inputs))
# test_input, test_input_size = self.test_inputs[choice], self.test_input_sizes[choice]
# # Determine output size of G (dynamic change)
# output_size, rand_h = random_size(orig_size=test_input_size,
# curriculum=self.conf.curriculum,
# i=i,
# iter_for_max_range=self.conf.iter_for_max_range,
# must_divide=self.conf.must_divide,
# min_scale=self.conf.min_scale,
# max_scale=self.conf.max_scale)
#
# g_preds, d_preds, reconstructs = self.gan.test(test_input, output_size, rand_h, test_input_size)
g_preds = [self.gan.input_tensor_noised, self.gan.G_pred]
d_preds = [self.gan.D.forward(self.gan.input_tensor_noised.detach(), self.gan.scale_weights),
self.gan.d_pred_fake]
reconstructs = self.gan.reconstruct
input_size = self.gan.input_tensor_noised.shape[2:]
result = image_concat(tensor2im(g_preds), tensor2im(d_preds), (input_size[0]*2, input_size[1]*2))
self.plot_gan_loss[0].set_data(range(i), self.G_loss[:i])
self.plot_gan_loss[1].set_data(range(i), self.D_loss_real[:i])
self.plot_gan_loss[2].set_data(range(i), self.D_loss_fake[:i])
self.gan_loss.set_xlim(0, i)
if self.conf.reconstruct_loss_stop_iter > i:
self.plot_reconstruct_loss[0].set_data(range(i), self.Rec_loss[:i])
self.reconstruct_loss.set_ylim(np.min(self.Rec_loss[:i]), np.max(self.Rec_loss[:i]))
self.reconstruct_loss.set_xlim(0, i)
self.result.imshow(np.clip(result, 0, 255), vmin=0, vmax=255)
self.real_example.imshow(np.clip(tensor2im(self.gan.real_example[0:1, :, :, :]), 0, 255), vmin=0, vmax=255)
self.d_map_real.imshow(self.gan.d_pred_real[0:1, :, :, :].detach().cpu().float().numpy().squeeze(),
cmap='gray', vmin=0, vmax=1)
if self.conf.reconstruct_loss_stop_iter > i:
self.reconstruction.imshow(np.clip(image_concat([tensor2im(reconstructs)]), 0, 255), vmin=0, vmax=255)
plt.savefig(self.conf.output_dir_path + '/monitor_%d' % i)
save_image(self.gan.G_pred, self.conf.output_dir_path + '/result_iter_%d.png' % i)
def prepare_result_dir(conf):
# Create results directory
conf.output_dir_path += '/' + conf.name + strftime('_%b_%d_%H_%M_%S', localtime())
os.makedirs(conf.output_dir_path)
# Put a copy of all *.py files in results path, to be able to reproduce experimental results
if conf.create_code_copy:
local_dir = os.path.dirname(__file__)
for py_file in glob.glob(local_dir + '/*.py'):
copy(py_file, conf.output_dir_path)
if conf.resume:
copy(conf.resume, os.path.join(conf.output_dir_path, 'starting_checkpoint.pth.tar'))
return conf.output_dir_path
def homography_based_on_top_corners_x_shift(rand_h):
p = np.array([[1., 1., -1, 0, 0, 0, -(-1. + rand_h[0]), -(-1. + rand_h[0]), -1. + rand_h[0]],
[0, 0, 0, 1., 1., -1., 1., 1., -1.],
[-1., -1., -1, 0, 0, 0, 1 + rand_h[1], 1 + rand_h[1], 1 + rand_h[1]],
[0, 0, 0, -1, -1, -1, 1, 1, 1],
[1, 0, -1, 0, 0, 0, 1, 0, -1],
[0, 0, 0, 1, 0, -1, 0, 0, 0],
[-1, 0, -1, 0, 0, 0, 1, 0, 1],
[0, 0, 0, -1, 0, -1, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 1]], dtype=np.float32)
b = np.zeros((9, 1), dtype=np.float32)
b[8, 0] = 1.
h = np.dot(np.linalg.inv(p), b)
return torch.from_numpy(h).view(3, 3).cuda()
def homography_grid(theta, size):
r"""Generates a 2d flow field, given a batch of homography matrices :attr:`theta`
Generally used in conjunction with :func:`grid_sample` to
implement Spatial Transformer Networks.
Args:
theta (Tensor): input batch of homography matrices (:math:`N \times 3 \times 3`)
size (torch.Size): the target output image size (:math:`N \times C \times H \times W`)
Example: torch.Size((32, 3, 24, 24))
Returns:
output (Tensor): output Tensor of size (:math:`N \times H \times W \times 2`)
"""
a = 1
b = 1
y, x = torch.meshgrid((torch.linspace(-b, b, np.int(size[-2]*a)), torch.linspace(-b, b, np.int(size[-1]*a))))
n = np.int(size[-2] * a) * np.int(size[-1] * a)
hxy = torch.ones(n, 3, dtype=torch.float)
hxy[:, 0] = x.contiguous().view(-1)
hxy[:, 1] = y.contiguous().view(-1)
out = hxy[None, ...].cuda().matmul(theta.transpose(1, 2))
# normalize
out = out[:, :, :2] / out[:, :, 2:]
return out.view(theta.shape[0], np.int(size[-2]*a), np.int(size[-1]*a), 2)
def hist_match(source, template, mask_3ch):
"""
Adjust the pixel values of a grayscale image such that its histogram
matches that of a target image
Arguments:
-----------
source: np.ndarray
Image to transform; the histogram is computed over the flattened
array
template: np.ndarray
Template image; can have different dimensions to source
Returns:
-----------
matched: np.ndarray
The transformed output image
"""
oldshape = source.shape
source_masked = source.ravel()[mask_3ch.ravel() > 128]
template = template.ravel()
# get the set of unique pixel values and their corresponding indices and
# counts
s_values, bin_idx, s_counts = np.unique(source_masked, return_inverse=True,
return_counts=True)
t_values, t_counts = np.unique(template, return_counts=True)
# take the cumsum of the counts and normalize by the number of pixels to
# get the empirical cumulative distribution functions for the source and
# template images (maps pixel value --> quantile)
s_quantiles = np.cumsum(s_counts).astype(np.float64)
s_quantiles /= s_quantiles[-1]
t_quantiles = np.cumsum(t_counts).astype(np.float64)
t_quantiles /= t_quantiles[-1]
# interpolate linearly to find the pixel values in the template image
# that correspond most closely to the quantiles in the source image
interp_t_values = np.interp(s_quantiles, t_quantiles, t_values)
out = source.copy().ravel()
out[mask_3ch.ravel() > 128] = interp_t_values[bin_idx]
return out.reshape(oldshape)