-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
170 lines (152 loc) · 8.78 KB
/
main.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
import os
os.environ["CUDA_VISIBLE_DEVICES"] = '0'
import argparse
import sys
import time
import cv2
import warnings
import menpo
sys.path.append('external/deep3dfacerecon_pytorch')
sys.path.append('external/face3d')
import numpy as np
from PIL import Image
from external.deep3dfacerecon_pytorch.deep3dfacerecon_api import Deep3dModel
from external.deep3dfacerecon_pytorch.util.nvdiffrast import MeshRenderer
from external.deep3dfacerecon_pytorch.models import networks
from external.deep3dfacerecon_pytorch.util import util
from utils.utils_uv import UV_Helper
from utils.utils_misc import im_menpo2PIL, pil_resize_to_tensor, pil_loader, convert_image, print_and_save_options
from style_transfer_3d import style_transfer_3D
from vgg_extractor import VGG16_Extractor
def main(args):
source_dir = args.source_dir
content_dir = args.content_dir
style_dir = args.style_dir
mask_dir = args.mask_dir
device = args.device
img_size = args.img_size
fov_256 = 2 * np.arctan(args.center_256 / args.focal_256) * 180 / np.pi
renderer_256 = MeshRenderer(
rasterize_fov=fov_256, znear=args.z_near, zfar=args.z_far, rasterize_size=int(2 * args.center_256), use_opengl=True
)
fov_512 = 2 * np.arctan(args.center_512 / args.focal_512) * 180 / np.pi
renderer_512 = MeshRenderer(
rasterize_fov=fov_512, znear=args.z_near, zfar=args.z_far, rasterize_size=int(2 * args.center_512), use_opengl=True
)
renderers = [renderer_256, renderer_512]
deep3dmodel = Deep3dModel()
uv_helper = UV_Helper(args.face_model)
vgg16_extractor = VGG16_Extractor().to(device).eval()
net_recog = networks.define_net_recog(
net_recog=args.net_recog, pretrained_path=args.net_recog_path
).to(device)
full_mask_path = os.path.join('./full_masks', 'bfm_full_mask.png' if args.face_model == 'BFM' else 'multi_pie_mask.png')
full_mask = cv2.resize(cv2.imread(full_mask_path), (img_size, img_size)).astype(np.uint8)
full_mask[full_mask >= 1] = 1
full_mask[full_mask < 1] = 0
im_paths = [os.path.join(source_dir, i) for i in sorted(os.listdir(source_dir)) if i.endswith('png') or i.endswith('jpg')]
lm_paths = [i.replace('png', 'txt').replace('jpg', 'txt') for i in im_paths]
lm_paths = [os.path.join(i.replace(i.split(os.path.sep)[-1], ''), 'detections', i.split(os.path.sep)[-1]) for i in lm_paths]
for i in range(len(im_paths)):
img_name = im_paths[i].split(os.path.sep)[-1].replace('jpg', 'png')
print(i, img_name)
output_dir = os.path.join(args.output_dir, args.exp_name)
if args.check_results_exist and os.path.exists(os.path.join(output_dir, str(args.style_transfer_num), img_name)):
continue
# img = Image.open(im_paths[i]).convert('RGB')
img = Image.open(im_paths[i])
lms = np.loadtxt(lm_paths[i]).astype(np.float32)
lms = lms.reshape([-1, 2])
recon_dict = deep3dmodel.recontruct(img, lms)
if args.use_preprocess:
print('begin to preprocess input image')
img = menpo.image.Image(recon_dict['input'])
inter_results = uv_helper.data_preprocess(args, img, recon_dict, img_name, parsing_mask=None)
print('image preprocess finished!')
content_pil = Image.fromarray(inter_results['coarse_tex']).convert('RGB')
style_pil = Image.fromarray(inter_results['valid_tex']).convert('RGB')
mask_pil = Image.fromarray(inter_results['mask']).convert('RGB')
else:
content_cv2 = uv_helper.get_coarse_tex(recon_dict, img_size, img_size)
content_pil = Image.fromarray(content_cv2).convert('RGB')
style_pil = pil_loader(os.path.join(style_dir, img_name)).convert('RGB')
mask_pil = pil_loader(os.path.join(mask_dir, img_name)).convert('RGB')
if not content_dir is None:
if not os.path.exists(os.path.join(content_dir, img_name)):
continue
content_pil = pil_loader(os.path.join(content_dir, img_name)).convert('RGB')
style_pil_flip = style_pil.transpose(Image.FLIP_LEFT_RIGHT)
# raw_im_pil = Image.fromarray(convert_image(recon_dict['im_hq'][0]))
raw_im_pil = im_menpo2PIL(img)
content_im_orig = pil_resize_to_tensor(content_pil, img_size).to(device)
style_im_orig = pil_resize_to_tensor(style_pil, img_size).to(device)
mask_im_orig = pil_resize_to_tensor(mask_pil, img_size).to(device)
raw_im_orig = pil_resize_to_tensor(raw_im_pil, img_size).to(device)
style_flip_orig = pil_resize_to_tensor(style_pil_flip, img_size).to(device)
datas = {
'option': args,
'content_im': content_im_orig,
'style_im': style_im_orig,
'style_im_flip': style_flip_orig,
'mask_im': mask_im_orig,
'full_mask': full_mask,
'raw_im': raw_im_orig,
'recon_dict': recon_dict,
'uv_helper': uv_helper,
'device': device,
'renderers': renderers,
'vgg16_extractor': vgg16_extractor,
'net_recog': net_recog,
'img_name': img_name
}
start_time = time.time()
style_transfer_3D(datas)
# Report total time
end_time = time.time()
total_time = (end_time - start_time) / 60
print('\nFinished after {:04.3f} minutes\n'.format(total_time))
if __name__ == '__main__':
parser = argparse.ArgumentParser('High Fidelity 3D Face Texture with 3D Style Transfer')
parser.add_argument('--source_dir', '-i', type=str, help='directory of input images', required=True)
parser.add_argument('--style_dir', '-s', type=str, help='directory of style images')
parser.add_argument('--content_dir', '-c', type=str, help='directory of content images')
parser.add_argument('--mask_dir', '-m', type=str, help='directory of valid texture masks')
parser.add_argument('--output_dir', '-o', type=str, help='directory of stylized images', required=True)
parser.add_argument('--inter_save_dir', type=str, help='directory of Intermediate result(content img: coarse tex, style img: valid_tex, mask)')
parser.add_argument('--save_inter_results', action='store_true', help='do not save intermidiate results')
parser.add_argument('--use_preprocess', action='store_false', help='use data preprocess')
parser.add_argument('--exp_name', type=str, required=True)
parser.add_argument('--img_size', type=int, default=512, help='we process all of images with resolution of 512 * 512')
parser.add_argument('--max_scales', type=int, default=2)
parser.add_argument('--max_iter', type=int, default=150)
parser.add_argument('--checkpoint_iter', type=int, default=50)
parser.add_argument('--content_weight', type=float, default=8.0)
parser.add_argument('--style_weight', type=float, default=1.7)
parser.add_argument('--recon_weight', type=float, default=20.0)
parser.add_argument('--learning_rate', type=float, default=0.3)
parser.add_argument('--check_results_exist', action='store_true', help='check if results exsit in the output directory')
parser.add_argument('--device', type=str, default='cuda:0')
parser.add_argument('--face_model', type=str, default='BFM', help='type of uv coords. [BFM, multi_pie]')
parser.add_argument('--style_transfer_num', type=int, default=5)
parser.add_argument('--add_noise', action='store_false', help='if add noise on the feature maps extracted from content image.')
# renderer parameters
parser.add_argument('--focal_256', type=float, default=1160.)
parser.add_argument('--focal_512', type=float, default=2320.)
parser.add_argument('--center_256', type=float, default=128.)
parser.add_argument('--center_512', type=float, default=256.)
parser.add_argument('--camera_d', type=float, default=10.)
parser.add_argument('--z_near', type=float, default=5.)
parser.add_argument('--z_far', type=float, default=15.)
parser.add_argument('--net_recog', type=str, default='r50', choices=['r18', 'r43', 'r50'],
help='face recog network structure')
parser.add_argument('--net_recog_path', type=str,
default='external/deep3dfacerecon_pytorch/checkpoints/recog_model/ms1mv3_arcface_r50_fp16/backbone.pth')
parser.add_argument('--use_predef_M', type=util.str2bool, nargs='?', const=True, default=False,
help='use predefined M for predicted face')
parser.add_argument('--use_style_mask', action='store_true', help='if use the mask to guide transfer of style image')
parser.add_argument('--use_content_mask', action='store_true', help='if use the mask to guide transfer of content image')
args = parser.parse_args()
warnings.filterwarnings("ignore")
# print and save options
print_and_save_options(args)
main(args)