-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgen_api.py
193 lines (162 loc) · 8.27 KB
/
gen_api.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
import torch
import torch.nn as nn
from networks.generator import Generator_Cycle
import numpy as np
import torchvision
import os
from tqdm import tqdm
import torchvision.transforms as transforms
import glob
from moviepy.editor import ImageSequenceClip
from PIL import Image
from skimage import io
from skimage import img_as_ubyte
import imageio
import cv2
from face_det.tracker.face_tracker import FaceTracker
from utils.warp_func import cv2CropWarpAffine
def load_image(filename, size):
img = Image.open(filename).convert('RGB')
img = img.resize((size, size))
img = np.asarray(img)
img = np.transpose(img, (2, 0, 1)) # 3 x 256 x 256
return img / 255.0
def save_video(save_path, name, vid_target_recon, fps=10, save_frames=False):
vid = (vid_target_recon.permute(0, 2, 3, 4, 1) * 127.5 + 127.5).clamp(0, 255).to(torch.uint8).cpu() # [1, 25, 256, 256, 3]
torchvision.io.write_video(save_path + '%s.mp4' % name, vid[0], fps=fps)
if save_frames:
os.makedirs(os.path.join(save_path + name), exist_ok=True)
for i in range(vid.shape[1]):
io.imsave(os.path.join(save_path + name, '%03d.png' % (i)), vid[0, i].numpy()) # .permute(1, 2, 0)
def data_preprocessing(img_path, size):
img = load_image(img_path, size) # [0, 1]
img = torch.from_numpy(img).unsqueeze(0).float() # [0, 1]
imgs_norm = (img - 0.5) * 2.0 # [-1, 1]
return imgs_norm
def extract_frames_from_a_video(vid_path):
cap = cv2.VideoCapture(vid_path)
frames = []
fps = cap.get(cv2.CAP_PROP_FPS)
while True:
ret, frame = cap.read()
if not ret:
break
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
frames.append(frame)
frames = np.stack(frames) # (N, H, W, 3)
return frames, fps
class Gen_API(nn.Module):
def __init__(self, dataset='vox', size=256, device='cuda:0', save_path='./id_explore', ckpt_path=None, motion_dim=20, id_dim=10):
super(Gen_API, self).__init__()
self.transform = torchvision.transforms.Compose([
transforms.ToTensor(),
transforms.Resize((size, size)),
transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))]
)
self.device = device
self.motion_dim = motion_dim
self.id_dim = id_dim
if dataset == 'vox':
if ckpt_path is not None:
path = ckpt_path
else:
path = 'checkpoints/660000.pt'
print('load from:', path)
else:
raise NotImplementedError
# pass
self.save_path = os.path.join(save_path)
os.makedirs(self.save_path, exist_ok=True)
print('==> loading model')
self.gen = Generator_Cycle(size=size, style_dim=512, motion_dim=motion_dim, id_dim=id_dim, channel_multiplier=1, blur_kernel=[1,3,3,1]).to(self.device)
weight = torch.load(path, map_location='cpu')['gen']
self.gen.load_state_dict(weight)
self.gen.eval()
self.face_tracker = FaceTracker(device=device)
@torch.no_grad()
def get_encoding(self, src):
input_src = self.transform(src).unsqueeze(0).to(self.device)
code_dict, _ = self.gen.enc.forward(input_src, input_drive=None)
wa = code_dict['wa'].cpu().numpy() # (N, 512)
a = code_dict['h_motion'].cpu().numpy() # (N, 20)
b = code_dict['h_id'].cpu().numpy() # (N, 10)
return wa, a, b
def crop_roi(self, image, bbox, expand_scale=1.6):
x1, y1, x2, y2 = bbox.astype(np.int32)
center = (x1+x2)/2, (y1+y2)/2
width = height = (x2-x1 + y2-y1)/2*expand_scale
crop_box = np.array([center[0]-width/2, center[1]-height/2, width, height]).astype(np.int32)
roi = cv2CropWarpAffine(image, crop_box, crop_size=256, interpolate=cv2.INTER_CUBIC)
return roi
@torch.no_grad()
def reenact(self, src, drv, same_id, crop_image=False):
'''
Args:
src: np.ndarray (H, W, 3)
drv: np.ndarray (H, W, 3)
same_id: bool, if the source and driving images are the same person
crop_image: bool, if crop the face region. Note that the too big or too small face region ratio will introduce unexpected artifacts人脸需要占人脸的比例为0.5-0.8，否则会出现异常
Return:
img_recon: np.ndarray (H, W, 3)
'''
if crop_image:
detected_faces = self.face_tracker.face_detector(src, rgb=True)
src = self.crop_roi(src, detected_faces[0][:4])
detected_faces = self.face_tracker.face_detector(drv, rgb=True)
drv = self.crop_roi(drv, detected_faces[0][:4])
input_src = self.transform(src).unsqueeze(0).to(self.device)
input_drv = self.transform(drv).unsqueeze(0).to(self.device)
img_recon = self.gen.inference(input_src, input_drv, same_id=same_id)
img_recon = (img_recon['img_reenacted'].clamp(-1, 1)[0].permute(1,2,0).cpu().numpy()*127.5 + 127.5).astype(np.uint8)
return img_recon
def save_img(self, img: np.ndarray, img_name: str):
os.makedirs(self.save_path, exist_ok=True)
io.imsave(os.path.join(self.save_path, img_name), img)
def video_reenactment(self, source_image, driving_video_path, crop_image=False):
print('==> running')
with torch.no_grad():
predictions = []
source = self.transform(source_image).unsqueeze(0).to(self.device).float()
code_source, _ = self.gen.enc(source, input_drive=None)
driving, fps = extract_frames_from_a_video(driving_video_path)
if crop_image:
detected_bboxes = self.face_tracker.face_detector(source_image, rgb=True)
source_image = self.crop_roi(source_image, detected_bboxes[0][:4])
io.imsave(os.path.join(self.save_path, 'source.jpg'), source_image)
source = self.transform(source_image).unsqueeze(0).to(self.device).float()
code_source, _ = self.gen.enc(source, input_drive=None)
detected_bboxes = self.face_tracker.face_detector(driving[0], rgb=True)
drive_images = []
for i in range(driving.shape[0]):
drive_crop = self.crop_roi(driving[i], detected_bboxes[0][:4])
drive_images.append(drive_crop)
driving = np.stack(drive_images)
for frame_idx in tqdm(range(driving.shape[0])):
driving_frame = self.transform(driving[frame_idx]).unsqueeze(0).to(self.device)
code_drive, _ = self.gen.enc.forward(driving_frame, input_drive=None)
alpha = [code_drive['h_motion'], code_source['h_id']]
img = self.gen.dec(code_source['wa'], alpha, code_source['feats'])
predictions.append(img.clamp(-1, 1)[0].permute(1, 2, 0).cpu().numpy()*0.5 + 0.5) # (-1, 1)
name = os.path.basename(src_path) + 'animated-res-' + os.path.basename(driving_video_path).rsplit('.', 1)[0]
imageio.mimsave(
os.path.join(self.save_path, '%s.mp4' % name),
[np.hstack([cv2.resize(source_image, (256, 256)), driving[i], img_as_ubyte(frame)]) for i, frame in enumerate(predictions)], fps=fps)
print('saved results in', name)
return predictions
if __name__ == '__main__':
size = 256
device = 'cuda:0'
ckpt_path = 'checkpoints/660000.pt'
save_path = './res_manipulation'
demo = Gen_API(dataset='vox', save_path=save_path, size=size, device=device, ckpt_path=ckpt_path, motion_dim=20, id_dim=10)
src_path = 'data/happy_level2_0.jpg'
drv_path = 'data/driving2.mp4'
img_src = np.array(Image.open(src_path).convert('RGB'))
if drv_path.rsplit('.', 1)[1] in {'jpg', 'png', 'bmp', 'jpeg'}:
save_img_path = os.path.join(demo.save_path, '%s-%s' % (os.path.basename(src_path), os.path.basename(drv_path)))
img_drv = np.array(Image.open(drv_path).convert('RGB'))
img_recon = demo.reenact(img_src, img_drv, same_id=True, crop_image=True)
saved_result = np.hstack([cv2.resize(img_src, (size, size)), cv2.resize(img_drv, (size, size)), cv2.resize(img_recon, (size, size))])
io.imsave(save_img_path, saved_result)
elif drv_path.rsplit('.', 1)[1] in {'mp4', 'avi', 'mov'}:
demo.video_reenactment(img_src, drv_path, crop_image=True)