Spaces:
Running
Running
Update src/facerender/animate.py
Browse files- src/facerender/animate.py +91 -72
src/facerender/animate.py
CHANGED
@@ -7,8 +7,7 @@ import numpy as np
|
|
7 |
import warnings
|
8 |
from skimage import img_as_ubyte
|
9 |
import safetensors
|
10 |
-
import safetensors.torch
|
11 |
-
|
12 |
warnings.filterwarnings('ignore')
|
13 |
|
14 |
import imageio
|
@@ -18,9 +17,9 @@ import torchvision
|
|
18 |
from src.facerender.modules.keypoint_detector import HEEstimator, KPDetector
|
19 |
from src.facerender.modules.mapping import MappingNet
|
20 |
from src.facerender.modules.generator import OcclusionAwareGenerator, OcclusionAwareSPADEGenerator
|
21 |
-
from src.facerender.modules.make_animation import make_animation
|
22 |
|
23 |
-
from pydub import AudioSegment
|
24 |
from src.utils.face_enhancer import enhancer_generator_with_len, enhancer_list
|
25 |
from src.utils.paste_pic import paste_pic
|
26 |
from src.utils.videoio import save_video_with_watermark
|
@@ -28,11 +27,11 @@ from src.utils.videoio import save_video_with_watermark
|
|
28 |
try:
|
29 |
import webui # in webui
|
30 |
in_webui = True
|
31 |
-
except:
|
32 |
in_webui = False
|
33 |
|
34 |
|
35 |
-
class AnimateFromCoeff
|
36 |
|
37 |
def __init__(self, sadtalker_path, device):
|
38 |
with open(sadtalker_path['facerender_yaml']) as f:
|
@@ -60,53 +59,72 @@ class AnimateFromCoeff():
|
|
60 |
for param in mapping.parameters():
|
61 |
param.requires_grad = False
|
62 |
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
|
|
|
|
|
|
68 |
else:
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
else:
|
74 |
-
raise AttributeError("
|
75 |
|
76 |
self.kp_extractor = kp_extractor
|
77 |
self.generator = generator
|
78 |
self.he_estimator = he_estimator
|
79 |
self.mapping = mapping
|
|
|
80 |
|
81 |
self.kp_extractor.eval()
|
82 |
self.generator.eval()
|
83 |
self.he_estimator.eval()
|
84 |
self.mapping.eval()
|
85 |
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
kp_detector=None, he_estimator=None,
|
90 |
-
device="cpu"):
|
91 |
|
92 |
checkpoint = safetensors.torch.load_file(checkpoint_path)
|
93 |
|
94 |
if generator is not None:
|
95 |
-
|
96 |
-
|
|
|
97 |
if kp_detector is not None:
|
98 |
-
|
99 |
-
|
|
|
100 |
if he_estimator is not None:
|
101 |
-
|
102 |
-
|
|
|
103 |
|
104 |
return None
|
105 |
|
106 |
-
def load_cpk_facevid2vid(self, checkpoint_path,
|
107 |
-
|
108 |
-
|
109 |
-
|
|
|
|
|
110 |
|
111 |
checkpoint = torch.load(checkpoint_path, map_location=torch.device(device))
|
112 |
|
@@ -118,6 +136,7 @@ class AnimateFromCoeff():
|
|
118 |
he_estimator.load_state_dict(checkpoint['he_estimator'])
|
119 |
if discriminator is not None and 'discriminator' in checkpoint:
|
120 |
discriminator.load_state_dict(checkpoint['discriminator'])
|
|
|
121 |
if optimizer_generator is not None and 'optimizer_generator' in checkpoint:
|
122 |
optimizer_generator.load_state_dict(checkpoint['optimizer_generator'])
|
123 |
if optimizer_discriminator is not None and 'optimizer_discriminator' in checkpoint:
|
@@ -129,45 +148,45 @@ class AnimateFromCoeff():
|
|
129 |
|
130 |
return checkpoint.get('epoch', 0)
|
131 |
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
|
|
|
|
148 |
break
|
149 |
-
if found:
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
return checkpoint.get('epoch', 0)
|
|
|
7 |
import warnings
|
8 |
from skimage import img_as_ubyte
|
9 |
import safetensors
|
10 |
+
import safetensors.torch
|
|
|
11 |
warnings.filterwarnings('ignore')
|
12 |
|
13 |
import imageio
|
|
|
17 |
from src.facerender.modules.keypoint_detector import HEEstimator, KPDetector
|
18 |
from src.facerender.modules.mapping import MappingNet
|
19 |
from src.facerender.modules.generator import OcclusionAwareGenerator, OcclusionAwareSPADEGenerator
|
20 |
+
from src.facerender.modules.make_animation import make_animation
|
21 |
|
22 |
+
from pydub import AudioSegment
|
23 |
from src.utils.face_enhancer import enhancer_generator_with_len, enhancer_list
|
24 |
from src.utils.paste_pic import paste_pic
|
25 |
from src.utils.videoio import save_video_with_watermark
|
|
|
27 |
try:
|
28 |
import webui # in webui
|
29 |
in_webui = True
|
30 |
+
except ImportError:
|
31 |
in_webui = False
|
32 |
|
33 |
|
34 |
+
class AnimateFromCoeff:
|
35 |
|
36 |
def __init__(self, sadtalker_path, device):
|
37 |
with open(sadtalker_path['facerender_yaml']) as f:
|
|
|
59 |
for param in mapping.parameters():
|
60 |
param.requires_grad = False
|
61 |
|
62 |
+
# FaceVid2Vid checkpoint yükleme
|
63 |
+
if 'checkpoint' in sadtalker_path:
|
64 |
+
self.load_cpk_facevid2vid_safetensor(
|
65 |
+
sadtalker_path['checkpoint'],
|
66 |
+
kp_detector=kp_extractor,
|
67 |
+
generator=generator,
|
68 |
+
he_estimator=None,
|
69 |
+
device=device
|
70 |
+
)
|
71 |
else:
|
72 |
+
self.load_cpk_facevid2vid(
|
73 |
+
sadtalker_path['free_view_checkpoint'],
|
74 |
+
kp_detector=kp_extractor,
|
75 |
+
generator=generator,
|
76 |
+
he_estimator=he_estimator,
|
77 |
+
device=device
|
78 |
+
)
|
79 |
+
|
80 |
+
# MappingNet checkpoint yükleme
|
81 |
+
if sadtalker_path.get('mappingnet_checkpoint') is not None:
|
82 |
+
self.load_cpk_mapping(
|
83 |
+
sadtalker_path['mappingnet_checkpoint'],
|
84 |
+
mapping=mapping,
|
85 |
+
device=device
|
86 |
+
)
|
87 |
else:
|
88 |
+
raise AttributeError("mappingnet_checkpoint path belirtmelisiniz.")
|
89 |
|
90 |
self.kp_extractor = kp_extractor
|
91 |
self.generator = generator
|
92 |
self.he_estimator = he_estimator
|
93 |
self.mapping = mapping
|
94 |
+
self.device = device
|
95 |
|
96 |
self.kp_extractor.eval()
|
97 |
self.generator.eval()
|
98 |
self.he_estimator.eval()
|
99 |
self.mapping.eval()
|
100 |
|
101 |
+
def load_cpk_facevid2vid_safetensor(self, checkpoint_path,
|
102 |
+
generator=None, kp_detector=None,
|
103 |
+
he_estimator=None, device="cpu"):
|
|
|
|
|
104 |
|
105 |
checkpoint = safetensors.torch.load_file(checkpoint_path)
|
106 |
|
107 |
if generator is not None:
|
108 |
+
state = {k.replace('generator.', ''): v
|
109 |
+
for k, v in checkpoint.items() if k.startswith('generator.')}
|
110 |
+
generator.load_state_dict(state)
|
111 |
if kp_detector is not None:
|
112 |
+
state = {k.replace('kp_extractor.', ''): v
|
113 |
+
for k, v in checkpoint.items() if k.startswith('kp_extractor.')}
|
114 |
+
kp_detector.load_state_dict(state)
|
115 |
if he_estimator is not None:
|
116 |
+
state = {k.replace('he_estimator.', ''): v
|
117 |
+
for k, v in checkpoint.items() if k.startswith('he_estimator.')}
|
118 |
+
he_estimator.load_state_dict(state)
|
119 |
|
120 |
return None
|
121 |
|
122 |
+
def load_cpk_facevid2vid(self, checkpoint_path,
|
123 |
+
generator=None, discriminator=None,
|
124 |
+
kp_detector=None, he_estimator=None,
|
125 |
+
optimizer_generator=None, optimizer_discriminator=None,
|
126 |
+
optimizer_kp_detector=None, optimizer_he_estimator=None,
|
127 |
+
device="cpu"):
|
128 |
|
129 |
checkpoint = torch.load(checkpoint_path, map_location=torch.device(device))
|
130 |
|
|
|
136 |
he_estimator.load_state_dict(checkpoint['he_estimator'])
|
137 |
if discriminator is not None and 'discriminator' in checkpoint:
|
138 |
discriminator.load_state_dict(checkpoint['discriminator'])
|
139 |
+
# Optimizeler varsa yükle
|
140 |
if optimizer_generator is not None and 'optimizer_generator' in checkpoint:
|
141 |
optimizer_generator.load_state_dict(checkpoint['optimizer_generator'])
|
142 |
if optimizer_discriminator is not None and 'optimizer_discriminator' in checkpoint:
|
|
|
148 |
|
149 |
return checkpoint.get('epoch', 0)
|
150 |
|
151 |
+
def load_cpk_mapping(self, checkpoint_path,
|
152 |
+
mapping=None, discriminator=None,
|
153 |
+
optimizer_mapping=None, optimizer_discriminator=None,
|
154 |
+
device='cpu'):
|
155 |
+
|
156 |
+
# 1) .tar ise içeriği aç ve .pth bul
|
157 |
+
if checkpoint_path.endswith(".tar"):
|
158 |
+
tmpdir = tempfile.mkdtemp()
|
159 |
+
with tarfile.open(checkpoint_path, "r") as tar:
|
160 |
+
tar.extractall(path=tmpdir)
|
161 |
+
found = False
|
162 |
+
for root, _, files in os.walk(tmpdir):
|
163 |
+
for fname in files:
|
164 |
+
if fname.endswith(".pth"):
|
165 |
+
checkpoint_path = os.path.join(root, fname)
|
166 |
+
found = True
|
167 |
+
break
|
168 |
+
if found:
|
169 |
break
|
170 |
+
if not found:
|
171 |
+
raise FileNotFoundError(f"{checkpoint_path} içinde .pth dosyası bulunamadı.")
|
172 |
+
|
173 |
+
# 2) Klasör yüklendiyse archive/data.pkl’e bak
|
174 |
+
if os.path.isdir(checkpoint_path):
|
175 |
+
possible = os.path.join(checkpoint_path, "archive", "data.pkl")
|
176 |
+
if os.path.isfile(possible):
|
177 |
+
checkpoint_path = possible
|
178 |
+
|
179 |
+
# 3) checkpoint’i yükle
|
180 |
+
checkpoint = torch.load(checkpoint_path, map_location=torch.device(device))
|
181 |
+
|
182 |
+
# 4) State dict’leri ata
|
183 |
+
if mapping is not None and 'mapping' in checkpoint:
|
184 |
+
mapping.load_state_dict(checkpoint['mapping'])
|
185 |
+
if discriminator is not None and 'discriminator' in checkpoint:
|
186 |
+
discriminator.load_state_dict(checkpoint['discriminator'])
|
187 |
+
if optimizer_mapping is not None and 'optimizer_mapping' in checkpoint:
|
188 |
+
optimizer_mapping.load_state_dict(checkpoint['optimizer_mapping'])
|
189 |
+
if optimizer_discriminator is not None and 'optimizer_discriminator' in checkpoint:
|
190 |
+
optimizer_discriminator.load_state_dict(checkpoint['optimizer_discriminator'])
|
191 |
+
|
192 |
+
return checkpoint.get('epoch', 0)
|
|
|
|