-
Notifications
You must be signed in to change notification settings - Fork 5
/
visualization.py
197 lines (160 loc) · 7.28 KB
/
visualization.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
import torch
import PIL.Image
import numpy as np
import io
from torchvision.utils import save_image
from utils.util import torch_to_numpy, manipulation
import time
def fig_to_image(fig):
buf = io.BytesIO()
fig.savefig(buf)
buf.seek(0)
return Image.open(buf)
def convert_array_to_images(np_array):
"""Converts numpy array to images with data type `uint8`.
This function assumes the input numpy array is with range [-1, 1], as well as
with shape [batch_size, channel, height, width]. Here, `channel = 3` for color
image and `channel = 1` for gray image.
The return images are with data type `uint8`, lying in range [0, 255]. In
addition, the return images are with shape [batch_size, height, width,
channel]. NOTE: the channel order will be the same as input.
Inputs:
np_array: The numpy array to convert.
Returns:
The converted images.
Raises:
ValueError: If this input is with wrong shape.
"""
input_shape = np_array.shape
if len(input_shape) != 4 or input_shape[1] not in [1, 3]:
raise ValueError('Input `np_array` should be with shape [batch_size, '
'channel, height, width], where channel equals to 1 or 3. '
'But {} is received!'.format(input_shape))
images = (np_array + 1.0) * 127.5
images = np.clip(images + 0.5, 0, 255).astype(np.uint8)
images = images.transpose(0, 2, 3, 1)
return images
@torch.no_grad()
def make_interpolation_chart(G, z=None, direction=None, shifts_r=10, shifts_count=5, resolution=512, canvas=None ,y_axis=0 ,boundary=False, name="",**kwargs):
'''
get z+direction image.
Arguments:
G {[type]} -- [fixed]
kwargs** :dpi
Keyword Arguments:
z {[tensor]} -- [noise input] (n, 512)
direction {[tensor]} -- [trained] dir (n, m, 512) + alpha(n, m, 1) --m:direction number
shifts_r {number} -- [shift range] (default: {10})
shifts_count {number} -- [shift number] (default: {5})
'''
assert z.shape[0] == 1, 'just for batch size is one'
#z = z.expand([shifts_count, z.shape[1]])
#solo_image_dir = "/home/beryl/Documents/github/multi-direction-code/paper_test_image"
if isinstance(shifts_r, int):
shifts_r = [shifts_r]
l = torch.linspace(0, 2, shifts_count).reshape((shifts_count,1)).expand(shifts_count, len(shifts_r))
shifts_r = torch.tensor(shifts_r, dtype=torch.float32)
shifts = ((l-1)* shifts_r).cuda()
fake_imgs = []
for index, shift in enumerate(shifts):
fake_img = G(z, direction, shift, is_edit=True)
#solo_canvas = PIL.Image.new('RGB',(resolution, resolution), 'white')
#numpy_img = convert_array_to_images(fake_img.detach().cpu().numpy()).squeeze()
#solo_canvas.paste(PIL.Image.fromarray(numpy_img, mode='RGB'), (0, 0))
#solo_canvas.save(solo_image_dir+ "/{}_{:.2}.png".format(name,(shift.detach().cpu().numpy()[0])))
fake_imgs.append(fake_img)
fake_imgs = torch.stack(fake_imgs, dim=1).squeeze()
fake_img = fake_imgs.cpu().detach().numpy()
#fake_img = fake_img[::-1]
fake_img = convert_array_to_images(fake_img)
#绘画
if canvas == None:
canvas = PIL.Image.new('RGB', (resolution*shifts_count, resolution*(y_axis+1)), 'white')
for col, image in enumerate(list(fake_img)):
canvas.paste(PIL.Image.fromarray(image, mode='RGB'), (col*resolution, y_axis*resolution))
return canvas
def adjust_pixel_range(images, min_val=-1.0, max_val=1.0, channel_order='NCHW'):
"""Adjusts the pixel range of the input images.
This function assumes the input array (image batch) is with shape [batch_size,
channel, height, width] if `channel_order = NCHW`, or with shape [batch_size,
height, width] if `channel_order = NHWC`. The returned images are with shape
[batch_size, height, width, channel] and pixel range [0, 255].
NOTE: The channel order of output images will remain the same as the input.
Args:
images: Input images to adjust pixel range.
min_val: Min value of the input images. (default: -1.0)
max_val: Max value of the input images. (default: 1.0)
channel_order: Channel order of the input array. (default: NCHW)
Returns:
The postprocessed images with dtype `numpy.uint8` and range [0, 255].
Raises:
ValueError: If the input `images` are not with type `numpy.ndarray` or the
shape is invalid according to `channel_order`.
"""
if not isinstance(images, np.ndarray):
raise ValueError(f'Images should be with type `numpy.ndarray`!')
channel_order = channel_order.upper()
if channel_order not in ['NCHW', 'NHWC']:
raise ValueError(f'Invalid channel order `{channel_order}`!')
if images.ndim != 4:
raise ValueError(f'Input images are expected to be with shape `NCHW` or '
f'`NHWC`, but `{images.shape}` is received!')
if channel_order == 'NCHW' and images.shape[1] not in [1, 3]:
raise ValueError(f'Input images should have 1 or 3 channels under `NCHW` '
f'channel order!')
if channel_order == 'NHWC' and images.shape[3] not in [1, 3]:
raise ValueError(f'Input images should have 1 or 3 channels under `NHWC` '
f'channel order!')
images = images.astype(np.float32)
images = (images - min_val) * 255 / (max_val - min_val)
images = np.clip(images + 0.5, 0, 255).astype(np.uint8)
if channel_order == 'NCHW':
images = images.transpose(0, 2, 3, 1)
return images
@torch.no_grad()
def output_interpolation_chart(G, ws=None, direction=None, shifts_r=10, mani_layers=None, shifts_vec=[], attr=1, shifts_count=5, resolution=512, canvas=None ,y_axis=0 ,**kwargs):
'''
get z+direction image.
Arguments:
G {[type]} -- [fixed]
kwargs** :dpi
Keyword Arguments:
z {[tensor]} -- [noise input] (n, 512)
direction {[tensor]} -- [trained] dir (n, m, 512) + alpha(n, m, 1) --m:direction number
shifts_r {number} -- [shift range] (default: {10})
shifts_count {number} -- [shift number] (default: {5})
'''
assert ws.shape[0] == 1, 'just for batch size is one'
#z = z.expand([shifts_count, z.shape[1]])
if isinstance(shifts_r, int):
shifts_r = [shifts_r]
l1 = np.array(shifts_vec, dtype=np.float32) / 100.0
l1 = np.expand_dims(l1,0)
l1 = l1.repeat(shifts_count, axis=0)
l2 = np.linspace(0, 1, shifts_count).reshape((shifts_count,1))
l1[:,attr-1:attr] = l2
l = torch.from_numpy(l1)
shifts_r = torch.tensor(shifts_r, dtype=torch.float32)
shifts = (l * shifts_r).cuda()
fake_imgs = []
wp_np = torch_to_numpy(ws)
for shift in shifts:
wp_mani = manipulation(latent_codes=wp_np,
boundary=direction,
start_distance=np.asarray([0,0,0,0,0,0,0,0,0,0,0]),
end_distance=shift,
steps=1,
layerwise_manipulation=True,
num_layers=16,
manipulation_layers=mani_layers,
is_code_layerwise=True,
is_boundary_layerwise=False)
test_torch = torch.from_numpy(wp_mani[:,0,:,:])
test_torch = test_torch.type(torch.FloatTensor).cuda()
fake_img = G.net.synthesis(test_torch)
fake_imgs.append(fake_img)
fake_imgs = torch.stack(fake_imgs, dim=1).squeeze()
fake_img = fake_imgs.cpu().detach().numpy()
fake_img = convert_array_to_images(fake_img)
#绘画
return list(fake_img)