-
Notifications
You must be signed in to change notification settings - Fork 125
/
demo_multi.py
199 lines (164 loc) · 8.51 KB
/
demo_multi.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
import matplotlib
matplotlib.use('Agg')
import os, sys
import yaml
from argparse import ArgumentParser
from tqdm import tqdm
import modules.generator as GEN
import imageio
import numpy as np
from skimage.transform import resize
from skimage import img_as_ubyte
import torch
from sync_batchnorm import DataParallelWithCallback
import depth
from modules.keypoint_detector import KPDetector
from animate import normalize_kp
from scipy.spatial import ConvexHull
from collections import OrderedDict
import pdb
import cv2
from glob import glob
if sys.version_info[0] < 3:
raise Exception("You must use Python 3 or higher. Recommended version is Python 3.7")
def load_checkpoints(config_path, checkpoint_path, cpu=False):
with open(config_path) as f:
config = yaml.load(f,Loader=yaml.FullLoader)
if opt.kp_num != -1:
config['model_params']['common_params']['num_kp'] = opt.kp_num
generator = getattr(GEN, opt.generator)(**config['model_params']['generator_params'],**config['model_params']['common_params'])
if not cpu:
generator.cuda()
config['model_params']['common_params']['num_channels'] = 4
kp_detector = KPDetector(**config['model_params']['kp_detector_params'],
**config['model_params']['common_params'])
if not cpu:
kp_detector.cuda()
if cpu:
checkpoint = torch.load(checkpoint_path, map_location=torch.device('cpu'))
else:
checkpoint = torch.load(checkpoint_path,map_location="cuda:0")
ckp_generator = OrderedDict((k.replace('module.',''),v) for k,v in checkpoint['generator'].items())
generator.load_state_dict(ckp_generator)
ckp_kp_detector = OrderedDict((k.replace('module.',''),v) for k,v in checkpoint['kp_detector'].items())
kp_detector.load_state_dict(ckp_kp_detector)
if not cpu:
generator = DataParallelWithCallback(generator)
kp_detector = DataParallelWithCallback(kp_detector)
generator.eval()
kp_detector.eval()
return generator, kp_detector
def make_animation(source_image, driving_video, generator, kp_detector, relative=True, adapt_movement_scale=True, cpu=False):
sources = []
drivings = []
with torch.no_grad():
predictions = []
depth_gray = []
source = torch.tensor(source_image[np.newaxis].astype(np.float32)).permute(0, 3, 1, 2)
driving = torch.tensor(np.array(driving_video)[np.newaxis].astype(np.float32)).permute(0, 4, 1, 2, 3)
if not cpu:
source = source.cuda()
driving = driving.cuda()
outputs = depth_decoder(depth_encoder(source))
depth_source = outputs[("disp", 0)]
outputs = depth_decoder(depth_encoder(driving[:, :, 0]))
depth_driving = outputs[("disp", 0)]
source_kp = torch.cat((source,depth_source),1)
driving_kp = torch.cat((driving[:, :, 0],depth_driving),1)
kp_source = kp_detector(source_kp)
kp_driving_initial = kp_detector(driving_kp)
# kp_source = kp_detector(source)
# kp_driving_initial = kp_detector(driving[:, :, 0])
for frame_idx in tqdm(range(driving.shape[2])):
driving_frame = driving[:, :, frame_idx]
if not cpu:
driving_frame = driving_frame.cuda()
outputs = depth_decoder(depth_encoder(driving_frame))
depth_map = outputs[("disp", 0)]
gray_driving = np.transpose(depth_map.data.cpu().numpy(), [0, 2, 3, 1])[0]
gray_driving = 1-gray_driving/np.max(gray_driving)
frame = torch.cat((driving_frame,depth_map),1)
kp_driving = kp_detector(frame)
kp_norm = normalize_kp(kp_source=kp_source, kp_driving=kp_driving,
kp_driving_initial=kp_driving_initial, use_relative_movement=relative,
use_relative_jacobian=relative, adapt_movement_scale=adapt_movement_scale)
out = generator(source, kp_source=kp_source, kp_driving=kp_norm,source_depth = depth_source, driving_depth = depth_map)
drivings.append(np.transpose(driving_frame.data.cpu().numpy(), [0, 2, 3, 1])[0])
sources.append(np.transpose(source.data.cpu().numpy(), [0, 2, 3, 1])[0])
predictions.append(np.transpose(out['prediction'].data.cpu().numpy(), [0, 2, 3, 1])[0])
depth_gray.append(gray_driving)
return sources, drivings, predictions,depth_gray
def find_best_frame(source, driving, cpu=False):
import face_alignment
def normalize_kp(kp):
kp = kp - kp.mean(axis=0, keepdims=True)
area = ConvexHull(kp[:, :2]).volume
area = np.sqrt(area)
kp[:, :2] = kp[:, :2] / area
return kp
fa = face_alignment.FaceAlignment(face_alignment.LandmarksType._2D, flip_input=True,
device='cpu' if cpu else 'cuda')
kp_source = fa.get_landmarks(255 * source)[0]
kp_source = normalize_kp(kp_source)
norm = float('inf')
frame_num = 0
for i, image in tqdm(enumerate(driving)):
kp_driving = fa.get_landmarks(255 * image)[0]
kp_driving = normalize_kp(kp_driving)
new_norm = (np.abs(kp_source - kp_driving) ** 2).sum()
if new_norm < norm:
norm = new_norm
frame_num = i
return frame_num
if __name__ == "__main__":
parser = ArgumentParser()
parser.add_argument("--config", required=True, help="path to config")
parser.add_argument("--checkpoint", default='vox-cpk.pth.tar', help="path to checkpoint to restore")
parser.add_argument("--source_folder", default='sup-mat/source.png', help="path to source image")
parser.add_argument("--driving_video", default='sup-mat/source.png', help="path to driving video")
parser.add_argument("--save_folder", default='result.mp4', help="path to output")
parser.add_argument("--relative", dest="relative", action="store_true", help="use relative or absolute keypoint coordinates")
parser.add_argument("--adapt_scale", dest="adapt_scale", action="store_true", help="adapt movement scale based on convex hull of keypoints")
parser.add_argument("--generator", type=str, required=True)
parser.add_argument("--kp_num", type=int, required=True)
parser.add_argument("--find_best_frame", dest="find_best_frame", action="store_true",
help="Generate from the frame that is the most alligned with source. (Only for faces, requires face_aligment lib)")
parser.add_argument("--best_frame", dest="best_frame", type=int, default=None,
help="Set frame to start from.")
parser.add_argument("--cpu", dest="cpu", action="store_true", help="cpu mode.")
parser.set_defaults(relative=False)
parser.set_defaults(adapt_scale=False)
opt = parser.parse_args()
depth_encoder = depth.ResnetEncoder(18, False)
depth_decoder = depth.DepthDecoder(num_ch_enc=depth_encoder.num_ch_enc, scales=range(4))
loaded_dict_enc = torch.load('depth/models/weights_19/encoder.pth')
loaded_dict_dec = torch.load('depth/models/weights_19/depth.pth')
filtered_dict_enc = {k: v for k, v in loaded_dict_enc.items() if k in depth_encoder.state_dict()}
depth_encoder.load_state_dict(filtered_dict_enc)
depth_decoder.load_state_dict(loaded_dict_dec)
depth_encoder.eval()
depth_decoder.eval()
if not opt.cpu:
depth_encoder.cuda()
depth_decoder.cuda()
reader = imageio.get_reader(opt.driving_video)
fps = reader.get_meta_data()['fps']
driving_video = []
try:
for im in reader:
driving_video.append(im)
except RuntimeError:
pass
reader.close()
driving_video = [resize(frame, (256, 256))[..., :3] for frame in driving_video]
generator, kp_detector = load_checkpoints(config_path=opt.config, checkpoint_path=opt.checkpoint, cpu=opt.cpu)
if not os.path.exists(opt.save_folder):
os.makedirs(opt.save_folder)
sources = glob(opt.source_folder+"/*")
for src in tqdm(sources):
source_image = imageio.imread(src)
source_image = resize(source_image, (256, 256))[..., :3]
# predictions = make_animation(source_image, driving_video, generator, kp_detector, relative=opt.relative, adapt_movement_scale=opt.adapt_scale, cpu=opt.cpu)
sources, drivings, predictions,depth_gray = make_animation(source_image, driving_video, generator, kp_detector, relative=opt.relative, adapt_movement_scale=opt.adapt_scale, cpu=opt.cpu)
fn = os.path.basename(src)
imageio.mimsave(os.path.join(opt.save_folder,fn+'.mp4'), [img_as_ubyte(p) for p in predictions], fps=fps)