-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathmain_test.py
171 lines (140 loc) · 7.69 KB
/
main_test.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
import os
import numpy as np
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
import torch
from lib.options import BaseOptions
from lib.options2 import BaseOptions as BaseOptions2
from lib.model import *
from lib.train_util import *
from Constants import consts
from lib.model2 import HGPIFuNet as SemanticNet
import torchvision.transforms as transforms
opt = BaseOptions().parse()
opt2 = BaseOptions2().parse()
def rotateY_by_view(view_id):
"""
input
view_id: 0-front, 1-right, 2-back, 3-left
"""
angle = np.radians(-view_id)
ry = np.array([[np.cos(angle), 0., np.sin(angle)],
[0., 1., 0.],
[-np.sin(angle), 0., np.cos(angle)]]) # (3,3)
ry = np.transpose(ry)
return ry
to_tensor = transforms.Compose([
transforms.Resize(opt.loadSize), # 512 , 512
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
def main(opt):
image_path = 'demo/1.png' # input image
image_mask_path = 'demo/1_mask.png' # The mask corresponding to the picture
normal_path = 'demo/1_normal.png' # image normal
save_file = 'demo/1.obj' # final result path
# load data path
cuda = torch.device('cuda') if len(opt.gpu_ids) > 1 else torch.device('cuda:%d' % opt.gpu_id)
projection_mode = 'orthogonal'
with torch.no_grad():
# load model only once time is ok!
netG_SeIF = HGPIFuNet(opt, projection_mode)
netG_SeIF.to(cuda)
netG_Semantic = SemanticNet(opt2, projection_mode)
netG_Semantic.to(cuda)
# can exchange net with each other!
model_semantic_path = r'checkpoints/model_Semantic/netG_epoch_6_293299'
netG_Semantic.load_state_dict(torch.load(model_semantic_path, map_location=cuda))
netG_Semantic.eval()
model_SeIF_path = 'checkpoints/model_SeIF/netG_epoch_3_293566'
netG_SeIF.load_state_dict(torch.load(model_SeIF_path, map_location=cuda))
netG_SeIF.eval()
# produce a mesh without color
# gen_mesh_iccv(opt, netG.module if len(opt.gpu_ids) > 1 else netG, cuda, data, save_path)
# acoording to the test we can find that the muti_demo is the same as this single demo ;0
# paths = ['209']
# for path in paths:
if True:
projection_mode = 'orthogonal'
cuda = torch.device('cuda') if len(opt.gpu_ids) > 1 else torch.device('cuda:%d' % opt.gpu_id)
# print(cuda)
if os.path.exists(image_path):
print(normal_path)
if not os.path.exists(save_file):
os.makedirs(os.path.dirname(save_file), exist_ok=True)
save_path = save_file
# compute calib and img of data
render_list = []
normal_list = []
calib_list = []
extrinsic_list = []
mask_list = []
# load mask
mask_data = np.round((cv2.imread(image_mask_path)[:, :, 0]).astype(np.float32) / 255.) # (1536, 1024)
mask_data_padded = np.zeros((max(mask_data.shape), max(mask_data.shape)), np.float32) # (1536, 1536)
mask_data_padded[:,
mask_data_padded.shape[0] // 2 - min(mask_data.shape) // 2:mask_data_padded.shape[0] // 2 + min(
mask_data.shape) // 2] = mask_data # (1536, 1536)
# NN resize to (512, 512)
mask_data_padded = cv2.resize(mask_data_padded, (opt.loadSize, opt.loadSize),
interpolation=cv2.INTER_NEAREST)
mask_data_padded = Image.fromarray(mask_data_padded)
# load image
image = cv2.imread(image_path)[:, :, ::-1] # (1536, 1024, 3), np.uint8, {0,...,255}
normal_image = cv2.imread(normal_path)[:, :, ::-1]
image_padded = np.zeros((max(image.shape), max(image.shape), 3), np.uint8) # (1536, 1536, 3)
normal_image_padded = np.zeros((max(normal_image.shape), max(normal_image.shape), 3), np.uint8)
image_padded[:,
image_padded.shape[0] // 2 - min(image.shape[:2]) // 2:image_padded.shape[0] // 2 + min(
image.shape[:2]) // 2,
:] = image # (1536, 1536, 3)
normal_image_padded[:,
normal_image_padded.shape[0] // 2 - min(normal_image.shape[:2]) // 2:normal_image_padded.shape[
0] // 2 + min(
normal_image.shape[:2]) // 2, :] = normal_image # (1536, 1536, 3)
# resize to (512, 512, 3), np.uint8
image_padded = cv2.resize(image_padded, (opt.loadSize, opt.loadSize))
normal_image_padded = cv2.resize(normal_image_padded, (opt.loadSize, opt.loadSize))
image_padded = Image.fromarray(image_padded)
normal_image_padded = Image.fromarray(normal_image_padded)
# load calib and intrinsic
trans_intrinsic = np.identity(4) # trans intrinsic
scale_intrinsic = np.identity(4) # ortho. proj. focal length
scale_intrinsic[0, 0] = 1. / consts.h_normalize_half # const == 2.
scale_intrinsic[1, 1] = -1. / consts.h_normalize_half # const == 2.
scale_intrinsic[2, 2] = -1. / consts.h_normalize_half # const == -2.
# extrinsic: model to cam R|t
extrinsic = np.identity(4)
# randomRot = np.array(dataConfig["randomRot"], np.float32) # by random R
viewRot = rotateY_by_view(view_id=0) # by view direction R
# extrinsic[:3,:3] = np.dot(viewRot, randomRot)
extrinsic[:3, :3] = viewRot.T
mask_data_padded = transforms.ToTensor()(mask_data_padded).float() # 1. inside, 0. outside
mask_list.append(mask_data_padded)
image_padded = to_tensor(image_padded) # (3, 512, 512), float -1 ~ 1
normal_image_padded = to_tensor(normal_image_padded) # (3, 512, 512), float -1 ~ 1
image_padded = mask_data_padded.expand_as(image_padded) * image_padded
normal_image_padded = mask_data_padded.expand_as(normal_image_padded) * normal_image_padded
render_list.append(image_padded)
normal_list.append(normal_image_padded)
intrinsic = np.matmul(trans_intrinsic, scale_intrinsic)
calib = torch.Tensor(np.matmul(intrinsic, extrinsic)).float()
# write by myself ; change calib（旋转矩阵） to ensure project!!!
# rot = calib[:3, :3]
# rot[0, 2] *= -1
# rot[1, 1] *= -1
# calib[:3, :3] = rot
extrinsic = torch.Tensor(extrinsic).float()
calib_list.append(calib) # save calib
extrinsic_list.append(extrinsic) # save extrinsic
#####
data = {'img': torch.stack(render_list, dim=0),
'normal': torch.stack(normal_list, dim=0),
'calib': torch.stack(calib_list, dim=0),
# model will be transformed into a XY-plane-center-aligned-2x2x2-volume of the cam. coord.
'extrinsic': torch.stack(extrinsic_list, dim=0),
'mask': torch.stack(mask_list, dim=0)
}
# every time we do a new encoder photo! and we get a net image feature!
gen_mesh(opt, netG_SeIF.module if len(opt.gpu_ids) > 1 else netG_SeIF, netG_Semantic, cuda, data, save_path)
if __name__ == '__main__':
main(opt)