forked from JinshuChen/MOGAN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathapps.py
226 lines (204 loc) · 8.82 KB
/
apps.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
import argparse
import time
import os
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
from utils_functions import *
from my_models import ResGenerator as G
from my_models import GatedResGenerator as G1
import torchvision
from torchvision import transforms
import matplotlib.pyplot as plt
from tqdm import tqdm
from skimage import io as img
from skimage import transform
import warnings
warnings.filterwarnings("ignore")
seed = 2020
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
print('seed:'+str(seed))
def random_generate(Gs,Zs,NoiseWeight,reals,
start_scale,gen_num,scale_h,scale_w,outputdir,inputdir,part,masks_f=None,G_num_layer=3):
if part=='f':
images_curr = []
n = 0
fpadder = make_padder(1+2+G_num_layer*2)
path = inputdir
scales = len(Gs_f)
# Gs = Gs[0:-1]
for G_f,Z_f,noiseweight_f,real_f in zip(Gs,Zs,NoiseWeight,reals):
g = G(3,256,3,1,0,3,0.01+1.7*(scales-n-1)/scales)
###### remember to change this factor
g.load_state_dict(torch.load(path+'/'+str(n)+'/G_f.pth'))
G_f = g
G_f.to(torch.device("cuda:0"))
del g
G_f = G_f.eval()
# import pdb;pdb.set_trace()
print('scale=%s'%n)
images_prev = images_curr
images_curr = []
h, w = real_f.shape[2], real_f.shape[3]
h, w = int(h*scale_h), int(w*scale_w)
real_f = F.interpolate(real_f,[h,w],mode='bilinear',align_corners=True)
plt.imsave(outputdir+'/r.png', convert_image_np(real_f), vmin=0, vmax=1)
real_f = real_f.squeeze(0).cpu()
# G_f= G_f.to(torch.device("cuda:0"))
for i in tqdm(range(gen_num)):
data_aug = []
if n==0:
z_curr = fpadder(Generate_noise([1,h,w],torch.device("cuda:0"),False,True))
# z_curr = generate_noise([1, Z_f.shape[2] - 2 * fpadder.padding[0], Z_f.shape[3] - 2 * fpadder.padding[0]], device=torch.device("cuda:0"))
# z_curr = z_curr.expand(1, 3, z_curr.shape[2], z_curr.shape[3])
# z_curr = Z_f
else:
z_curr = fpadder(Generate_noise([3,h,w],torch.device("cuda:0"),False,False))
# z_curr = generate_noise([3, Z_f.shape[2] - 2 * fpadder.padding[0], Z_f.shape[3] - 2 * fpadder.padding[0]], device=torch.device("cuda:0"))
# z_curr = fpadder(z_curr)
if images_prev==[]:
I_prev = torch.zeros_like(z_curr).to(torch.device("cuda:0"))
else:
I_prev = images_prev[i]
#SR?
# m = nn.Upsample([h,w],mode='bilinear',align_corners=True)
# I_prev = m(I_prev)
I_prev = F.interpolate(I_prev,[h,w],mode='bilinear',align_corners=True)
I_prev = fpadder(I_prev)
#暂不可用！
# if n<start_scale:
# z_curr = Z_f
###
_augument = data_augmenter(real_f,torch.device("cuda:0"))
# _augument = torch.zeros_like(real_f.unsqueeze(0).to(torch.device("cuda:0")))
for j in range(G_num_layer):
_fpadder = make_padder(2+2*3+(G_num_layer-j)*2)
# data_aug.append(_fpadder(torch.ones_like(_augument).detach()))
# plt.imsave(outputdir+'/a.png', convert_image_np(_augument.detach()), vmin=0, vmax=1)
data_aug.append(_fpadder(_augument).detach())
del _augument
z_in = noiseweight_f * z_curr + I_prev
# z_in = z_in.to(torch.device("cuda:0"))
I_curr = G_f(z_in.detach(),data_aug,I_prev)
images_curr.append(I_curr.detach())
plt.imsave(outputdir+'/f_%s.png' % (i), convert_image_np(I_curr.detach()), vmin=0, vmax=1)
del G_f
n+=1
elif part=='b':
images_curr = []
n = 0
G_num_layer = 3
# bpadder = make_padder(1+2+G_num_layer*2)
bpadder = make_padder(0)
path = inputdir
scales = len(Gs_b)
# Gs = Gs[0:-1]
for G_b,Z_b,noiseweight_b,real_b,mask_f in zip(Gs,Zs,NoiseWeight,reals,masks_f):
print('scale=%s'%n)
g = G1(3,256,3,1,0,3)
g.load_state_dict(torch.load(path+'/'+str(n)+'/G_b.pth'))
G_b = g
G_b.to(torch.device("cuda:1"))
del g
G_b = G_b.eval()
mask_f = mask_f.to(torch.device("cuda:1"))
images_prev = images_curr
images_curr = []
h, w = real_b.shape[2], real_b.shape[3]
h, w = int(h*scale_h), int(w*scale_w)
real_b = F.interpolate(real_b,[h,w],mode='bilinear',align_corners=True)
real_b = real_b.squeeze(0).cpu()
for i in tqdm(range(gen_num)):
if n==0:
z_curr = bpadder(Generate_noise([1,h,w],torch.device("cuda:1"),False,True))
# z_curr = Z_f
else:
z_curr = bpadder(Generate_noise([3,h,w],torch.device("cuda:1"),False,False))
if images_prev==[]:
I_prev = bpadder(torch.zeros_like(z_curr).to(torch.device("cuda:1")))
else:
I_prev = images_prev[i]
#SR?
# m = nn.Upsample([z_curr.shape[2],z_curr.shape[3]],mode='bilinear',align_corners=True)
# I_prev = m(I_prev)
I_prev = F.interpolate(I_prev,[h,w],mode='bilinear',align_corners=True)
# not available for now
if n < start_scale:
z_curr = Z_b
###
z_in = noiseweight_b * z_curr + I_prev
I_curr = G_b(z_in.detach(),I_prev)
images_curr.append(I_curr.detach())
mask = torch.ones_like(mask_f) - mask_f
I_curr = I_curr*mask - mask_f
plt.imsave(outputdir+'/b_%s.png' % (i), convert_image_np(I_curr.detach()), vmin=0, vmax=1)
del G_b
n+=1
def SR():
pass
def harmonization():
pass
def edit():
pass
def paint2image():
pass
def animation(Gs_f,Zs_f,NoiseWeight_f,reals_f,
Gs_b,Zs_b,NoiseWeight_b,reals_b,
start_scale,outputdir):
pass
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--mode', type=str, required=True)
parser.add_argument('--path', type=str, required=True)
parser.add_argument('--part', type=str, required=True)
parser.add_argument('--start_scale', type=int, default=0)
#for random generate
parser.add_argument('--gen_num', type=int, default=50)
parser.add_argument('--scale_h', type=float, default=1)
parser.add_argument('--scale_w', type=float, default=1)
parser = parser.parse_args()
local_time = time.strftime("%m%d_%H:%M:%S", time.localtime())
local_time = local_time.replace(":", "_")
outputdir = '/home/chenjinshu/singan/SinGAN-master/app_result/' + str(parser.mode) + '/' + local_time
inputdir = '/home/chenjinshu/singan/SinGAN-master/' + str(parser.path)
if parser.part=='f':
Gs_f = torch.load('%s/Gs_f.pth'%inputdir)
Zs_f = torch.load('%s/Zs_f.pth'%inputdir)
NoiseWeight_f = torch.load('%s/noiseweight_f.pth'%inputdir)
elif parser.part=='b':
Gs_b = torch.load('%s/Gs_b.pth'%inputdir)
Zs_b = torch.load('%s/Zs_b.pth'%inputdir)
NoiseWeight_b = torch.load('%s/noiseweight_b.pth'%inputdir)
reals_f = torch.load('%s/reals_f.pth'%inputdir)
reals_b = torch.load('%s/reals_b.pth'%inputdir)
masks_f = torch.load('%s/masks_f.pth'%inputdir)
if not os.path.exists(outputdir):
os.makedirs(outputdir)
if parser.mode=='r':
if parser.part=='f':
random_generate(Gs_f,Zs_f,NoiseWeight_f,reals_f,
parser.start_scale,parser.gen_num,
parser.scale_h,parser.scale_w,outputdir,inputdir,parser.part)
elif parser.part=='b':
random_generate(
Gs_b,Zs_b,NoiseWeight_b,reals_b,
parser.start_scale,parser.gen_num,
parser.scale_h,parser.scale_w,outputdir,inputdir,parser.part,masks_f=masks_f)
elif parser.mode=='s':
pass
elif parser.mode=='h':
pass
elif parser.mode=='e':
pass
elif parser.mode=='p':
pass
elif parser.mode=='a':
# not available for now
animation(Gs_f,Zs_f,NoiseWeight_f,reals_f,
Gs_b,Zs_b,NoiseWeight_b,reals_b,
parser.start_scale,outputdir)
else:
print('???')