forked from VinAIResearch/CPM
-
Notifications
You must be signed in to change notification settings - Fork 0
/
makeup.py
231 lines (211 loc) · 9.59 KB
/
makeup.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
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
import cv2
import numpy as np
import tensorflow as tf
import torch
import utils.net as net
from blend_modes import darken_only, hard_light, normal
from PIL import Image
from torchvision import transforms
from utils.api import PRN
from utils.models import Segmentor
from utils.render import prepare_tri_weights, render_by_tri, render_texture
from utils.utils import de_norm, to_tensor, to_var
class Makeup:
def __init__(self, args):
# if args.pattern:
self.pattern = Segmentor(args)
self.pattern.test_model(args.checkpoint_pattern)
self.color = net.Generator_branch(64, 6).cuda()
self.color.load_state_dict(torch.load(args.checkpoint_color))
self.color.eval()
if args.prn:
self.prn = PRN(is_dlib=True)
def get_mask(self, img):
x_tensor = to_tensor(img / 255)
x_tensor = torch.from_numpy(x_tensor).unsqueeze(0).cuda()
pr_mask = self.pattern.model.predict(x_tensor)
pr_mask = pr_mask[0, 0, :, :].detach().cpu().numpy()
return pr_mask
def makeup(self, img_A, img_B):
transform = transforms.Compose(
[
transforms.Resize(256),
transforms.ToTensor(),
transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5]),
]
)
img_A = transform(Image.fromarray(img_A))
img_B = transform(Image.fromarray(img_B))
img_A = img_A[None, :, :, :]
img_B = img_B[None, :, :, :]
real_org = to_var(img_A).cuda()
real_ref = to_var(img_B).cuda()
# Get makeup result
fake_A, fake_B = self.color(real_org, real_ref)
result = de_norm(fake_A.detach()[0]).cpu().numpy().transpose(1, 2, 0)
result = cv2.resize(result, (256, 256), cv2.INTER_CUBIC)
return result
def prn_process(self, face):
# --- face
self.face = cv2.resize(face, (256, 256))
self.pos = self.prn.process(self.face)
self.vertices = self.prn.get_vertices(self.pos)
# self.face = face/255
self.h, self.w, _ = self.face.shape
self.triangles = self.prn.triangles
vis_colors = np.ones((self.vertices.shape[0], 1))
face_mask = render_texture(self.vertices.T, vis_colors.T, self.triangles.T, self.h, self.w, c=1)
self.face_mask = np.squeeze(face_mask > 0).astype(np.float32)
self.weights, self.dst_triangle_buffer = prepare_tri_weights(self.vertices.T, self.triangles.T, self.h, self.w)
uv_face_eye = np.array(Image.open("./PRNet/uv-data/uv_face_eyes.png"))[:, :, :3] / 255
new_colors = self.prn.get_colors_from_texture(uv_face_eye)
new_colors = (new_colors > 0).astype("uint8")
mask_out_eye = render_by_tri(
new_colors.T,
self.triangles.T,
self.weights,
self.dst_triangle_buffer,
self.h,
self.w,
c=3,
)
self.mask_out_eye = (mask_out_eye > 0).astype("uint8") # [:, :, np.newaxis]
tf.reset_default_graph()
def prn_process_target(self, face):
# --- face
face = cv2.resize(face, (256, 256))
pos = self.prn.process(face)
vertices = self.prn.get_vertices(pos)
# self.face = face/255
h, w, _ = face.shape
triangles = self.prn.triangles
vis_colors = np.ones((vertices.shape[0], 1))
face_mask = render_texture(vertices.T, vis_colors.T, triangles.T, h, w, c=1)
face_mask = np.squeeze(face_mask > 0).astype(np.float32)
weights, dst_triangle_buffer = prepare_tri_weights(vertices.T, triangles.T, h, w)
# uv_face = cv2.imread("./PRNet/uv-data/uv_face.png")[:, :, 0] / 255.0
texture = cv2.remap(
face,
pos[:, :, :2].astype(np.float32),
None,
interpolation=cv2.INTER_LINEAR,
borderMode=cv2.BORDER_CONSTANT,
borderValue=(0),
)
tf.reset_default_graph()
return texture
def get_texture(self):
texture = cv2.remap(
self.face,
self.pos[:, :, :2].astype(np.float32),
None,
interpolation=cv2.INTER_LINEAR,
borderMode=cv2.BORDER_CONSTANT,
borderValue=(0),
)
return texture
def get_seg(self):
texture = cv2.remap(
self.face_seg,
self.pos[:, :, :2].astype(np.float32),
None,
interpolation=cv2.INTER_NEAREST,
borderMode=cv2.BORDER_CONSTANT,
borderValue=(0),
)
return texture
def render_texture(self, texture, patt_only=False):
new_colors = self.prn.get_colors_from_texture(texture)
new_image = render_by_tri(
new_colors.T,
self.triangles.T,
self.weights,
self.dst_triangle_buffer,
self.h,
self.w,
c=3,
)
# new_face = self.face_mask[:, :, np.newaxis]*new_image + (1-self.face_mask[:, :, np.newaxis])*self.face/255
# if patt_only:
# return new_face
# else:
return new_image
def blend_imgs(self, source, reference, blend_mode="normal", alpha=0.8):
""""""
# blurred_mask = cv2.GaussianBlur(np.stack([self.face_mask, self.face_mask, self.face_mask], axis=2)*255, (25, 25), 0)
blurred_mask = cv2.GaussianBlur(self.mask_out_eye * 255, (25, 25), 0)
extend_mask = self.mask_out_eye.copy() * 255
gray = cv2.cvtColor(extend_mask, cv2.COLOR_BGR2GRAY)
contours, hierarchy = cv2.findContours(gray.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
new_mask = np.where(extend_mask == np.array([255, 255, 255]), blurred_mask, extend_mask)
refer = reference * self.mask_out_eye + source * (1 - self.mask_out_eye)
source = cv2.cvtColor(source.astype("uint8"), cv2.COLOR_RGB2RGBA)
source[:, :, 3] = np.ones((256, 256)) * 255
refer = cv2.cvtColor(refer.astype("uint8"), cv2.COLOR_RGB2RGBA)
refer[:, :, 3] = new_mask[:, :, 0]
if blend_mode == "normal":
blended_img = normal(source.astype("float"), refer.astype("float"), alpha).astype("uint8")
elif blend_mode == "darken_only":
blended_img = darken_only(source.astype("float"), refer.astype("float"), alpha).astype("uint8")
elif blend_mode == "hard_light":
blended_img = hard_light(source.astype("float"), refer.astype("float"), alpha).astype("uint8")
# Image.fromarray(np.concatenate([source.astype('uint8'), blended_img[:, :, :3], tar.astype('uint8')], axis=1))
return blended_img[:, :, :3]
def get_blur_mask(self, source_seg):
seg = cv2.resize(source_seg, (256, 256), interpolation=cv2.INTER_NEAREST)
if len(seg.shape) == 3:
seg = seg[:, :, 0]
facial_mask = (seg == 1) + (seg == 2) + (seg == 3) + (seg == 6) + (seg == 7) + (seg == 9)
facial_mask = facial_mask.astype("uint8")
facial_mask = cv2.dilate(facial_mask, np.ones((10, 10), np.uint8), iterations=1)
# mask_out_eye = cv2.dilate(self.mask_out_eye, np.ones((10, 10),np.uint8), iterations = 1)
facial_mask = (facial_mask == 1) * (self.mask_out_eye[:, :, 0] == 1)
facial_mask = np.stack([facial_mask, facial_mask, facial_mask], axis=2).astype("uint8")
blurred_mask = cv2.GaussianBlur(facial_mask * 255, (25, 25), 0)
extend_mask = facial_mask.copy() * 255
gray = cv2.cvtColor(extend_mask, cv2.COLOR_BGR2GRAY)
contours, hierarchy = cv2.findContours(gray.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
extend_mask = cv2.drawContours(extend_mask, contours, -1, (255, 255, 255), 30)
new_mask = np.where(extend_mask == np.array([255, 255, 255]), blurred_mask, facial_mask)
return new_mask
def blend_imgs_2(self, source, reference, source_seg, blend_mode="normal", alpha=0.9):
""""""
# seg = np.array(Image.open(list_segs_A[0]))
new_mask = self.get_blur_mask(source_seg)
refer = reference * self.mask_out_eye + source * (1 - self.mask_out_eye)
source = cv2.cvtColor(source.astype("uint8"), cv2.COLOR_RGB2RGBA)
source[:, :, 3] = np.ones((256, 256)) * 255
refer = cv2.cvtColor(refer.astype("uint8"), cv2.COLOR_RGB2RGBA)
refer[:, :, 3] = new_mask[:, :, 0]
if blend_mode == "normal":
blended_img = normal(source.astype("float"), refer.astype("float"), alpha).astype("uint8")
elif blend_mode == "darken_only":
blended_img = darken_only(source.astype("float"), refer.astype("float"), alpha).astype("uint8")
elif blend_mode == "hard_light":
blended_img = hard_light(source.astype("float"), refer.astype("float"), alpha).astype("uint8")
# Image.fromarray(np.concatenate([source.astype('uint8'), blended_img[:, :, :3], tar.astype('uint8')], axis=1))
return blended_img[:, :, :3]
def location_to_crop(self, mini=False):
if mini:
idx = np.where(self.mask_out_eye[:, :, 0] == 1)
x2, y2, x1, y1 = (
np.min(idx[1]),
np.max(idx[1]),
np.min(idx[0]),
np.max(idx[0]),
)
max_idx = np.argmax([y1 - x1, y2 - x2]) + 1
if max_idx == 1:
y2 = x2 + (y1 - x1)
else:
y1 = x1 + (y2 - x2)
return x2, y2, x1, y1
else:
idx = np.where(self.mask_out_eye[:, :, 0] == 1)
x2, y2, x1, y1 = (
np.min(idx[1]),
np.max(idx[1]),
np.min(idx[0]),
np.max(idx[0]),
)
return x2, y2, x1, y1