Spaces:
Running
Running
guocheng66
commited on
Commit
•
57f6383
1
Parent(s):
c04f3e8
Upload 6 files
Browse files
app.py
ADDED
@@ -0,0 +1,185 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from PIL import ImageColor
|
3 |
+
|
4 |
+
import onnxruntime
|
5 |
+
import cv2
|
6 |
+
import numpy as np
|
7 |
+
|
8 |
+
# The common resume photo size is 35mmx45mm
|
9 |
+
RESUME_PHOTO_W = 350
|
10 |
+
RESUME_PHOTO_H = 450
|
11 |
+
|
12 |
+
|
13 |
+
# modified from https://github.com/opencv/opencv_zoo/blob/main/models/face_detection_yunet/yunet.py
|
14 |
+
class YuNet:
|
15 |
+
def __init__(
|
16 |
+
self,
|
17 |
+
modelPath,
|
18 |
+
inputSize=[320, 320],
|
19 |
+
confThreshold=0.6,
|
20 |
+
nmsThreshold=0.3,
|
21 |
+
topK=5000,
|
22 |
+
backendId=0,
|
23 |
+
targetId=0,
|
24 |
+
):
|
25 |
+
self._modelPath = modelPath
|
26 |
+
self._inputSize = tuple(inputSize) # [w, h]
|
27 |
+
self._confThreshold = confThreshold
|
28 |
+
self._nmsThreshold = nmsThreshold
|
29 |
+
self._topK = topK
|
30 |
+
self._backendId = backendId
|
31 |
+
self._targetId = targetId
|
32 |
+
|
33 |
+
self._model = cv2.FaceDetectorYN.create(
|
34 |
+
model=self._modelPath,
|
35 |
+
config="",
|
36 |
+
input_size=self._inputSize,
|
37 |
+
score_threshold=self._confThreshold,
|
38 |
+
nms_threshold=self._nmsThreshold,
|
39 |
+
top_k=self._topK,
|
40 |
+
backend_id=self._backendId,
|
41 |
+
target_id=self._targetId,
|
42 |
+
)
|
43 |
+
|
44 |
+
@property
|
45 |
+
def name(self):
|
46 |
+
return self.__class__.__name__
|
47 |
+
|
48 |
+
def setBackendAndTarget(self, backendId, targetId):
|
49 |
+
self._backendId = backendId
|
50 |
+
self._targetId = targetId
|
51 |
+
self._model = cv2.FaceDetectorYN.create(
|
52 |
+
model=self._modelPath,
|
53 |
+
config="",
|
54 |
+
input_size=self._inputSize,
|
55 |
+
score_threshold=self._confThreshold,
|
56 |
+
nms_threshold=self._nmsThreshold,
|
57 |
+
top_k=self._topK,
|
58 |
+
backend_id=self._backendId,
|
59 |
+
target_id=self._targetId,
|
60 |
+
)
|
61 |
+
|
62 |
+
def setInputSize(self, input_size):
|
63 |
+
self._model.setInputSize(tuple(input_size))
|
64 |
+
|
65 |
+
def infer(self, image):
|
66 |
+
# Forward
|
67 |
+
faces = self._model.detect(image)
|
68 |
+
return faces[1]
|
69 |
+
|
70 |
+
|
71 |
+
class ONNXModel:
|
72 |
+
def __init__(self, model_path, input_w, input_h):
|
73 |
+
self.model = onnxruntime.InferenceSession(model_path)
|
74 |
+
self.input_w = input_w
|
75 |
+
self.input_h = input_h
|
76 |
+
|
77 |
+
def preprocess(self, rgb, mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)):
|
78 |
+
# convert the input data into the float32 input
|
79 |
+
img_data = (
|
80 |
+
np.array(cv2.resize(rgb, (self.input_w, self.input_h)))
|
81 |
+
.transpose(2, 0, 1)
|
82 |
+
.astype("float32")
|
83 |
+
)
|
84 |
+
|
85 |
+
# normalize
|
86 |
+
norm_img_data = np.zeros(img_data.shape).astype("float32")
|
87 |
+
|
88 |
+
for i in range(img_data.shape[0]):
|
89 |
+
norm_img_data[i, :, :] = img_data[i, :, :] / 255
|
90 |
+
norm_img_data[i, :, :] = (norm_img_data[i, :, :] - mean[i]) / std[i]
|
91 |
+
|
92 |
+
# add batch channel
|
93 |
+
norm_img_data = norm_img_data.reshape(1, 3, self.input_h, self.input_w).astype(
|
94 |
+
"float32"
|
95 |
+
)
|
96 |
+
return norm_img_data
|
97 |
+
|
98 |
+
def forward(self, image):
|
99 |
+
input_data = self.preprocess(image)
|
100 |
+
output_data = self.model.run(["argmax_0.tmp_0"], {"x": input_data})
|
101 |
+
|
102 |
+
return output_data
|
103 |
+
|
104 |
+
|
105 |
+
def make_resume_photo(rgb, background_color):
|
106 |
+
h, w, _ = rgb.shape
|
107 |
+
bgr = cv2.cvtColor(rgb, cv2.COLOR_RGB2BGR)
|
108 |
+
|
109 |
+
# Initialize models
|
110 |
+
face_detector = YuNet("models/face_detection_yunet_2023mar.onnx")
|
111 |
+
face_detector.setInputSize([w, h])
|
112 |
+
human_segmentor = ONNXModel(
|
113 |
+
"models/human_pp_humansegv2_lite_192x192_inference_model.onnx", 192, 192
|
114 |
+
)
|
115 |
+
|
116 |
+
# yunet uses opencv bgr image format
|
117 |
+
detections = face_detector.infer(bgr)
|
118 |
+
|
119 |
+
results = []
|
120 |
+
for idx, det in enumerate(detections):
|
121 |
+
# bounding box
|
122 |
+
pt1 = np.array((det[0], det[1]))
|
123 |
+
pt2 = np.array((det[0] + det[2], det[1] + det[3]))
|
124 |
+
|
125 |
+
# face landmarks
|
126 |
+
landmarks = det[4:14].reshape((5, 2))
|
127 |
+
right_eye = landmarks[0]
|
128 |
+
left_eye = landmarks[1]
|
129 |
+
|
130 |
+
angle = np.arctan2(right_eye[1] - left_eye[1], (right_eye[0] - left_eye[0]))
|
131 |
+
rmat = cv2.getRotationMatrix2D((0, 0), -angle, 1)
|
132 |
+
|
133 |
+
# apply rotation
|
134 |
+
rotated_bgr = cv2.warpAffine(bgr, rmat, (bgr.shape[1], bgr.shape[0]))
|
135 |
+
rotated_pt1 = rmat[:, :-1] @ pt1
|
136 |
+
rotated_pt2 = rmat[:, :-1] @ pt2
|
137 |
+
|
138 |
+
face_w, face_h = rotated_pt2 - rotated_pt1
|
139 |
+
up_length = int(face_h / 4)
|
140 |
+
down_length = int(face_h / 3)
|
141 |
+
crop_h = face_h + up_length + down_length
|
142 |
+
crop_w = int(crop_h * (RESUME_PHOTO_W / RESUME_PHOTO_H))
|
143 |
+
|
144 |
+
pt1 = np.array(
|
145 |
+
(rotated_pt1[0] - (crop_w - face_w) / 2, rotated_pt1[1] - up_length)
|
146 |
+
).astype(np.int32)
|
147 |
+
pt2 = np.array((pt1[0] + crop_w, pt1[1] + crop_h)).astype(np.int32)
|
148 |
+
|
149 |
+
resume_photo = rotated_bgr[pt1[1] : pt2[1], pt1[0] : pt2[0], :]
|
150 |
+
|
151 |
+
rgb = cv2.cvtColor(resume_photo, cv2.COLOR_BGR2RGB)
|
152 |
+
mask = human_segmentor.forward(rgb)
|
153 |
+
mask = mask[0].transpose(1, 2, 0)
|
154 |
+
mask = cv2.resize(
|
155 |
+
mask.astype(np.uint8), (resume_photo.shape[1], resume_photo.shape[0])
|
156 |
+
)
|
157 |
+
|
158 |
+
resume_photo = cv2.cvtColor(resume_photo, cv2.COLOR_BGR2RGB)
|
159 |
+
resume_photo[mask == 0] = ImageColor.getcolor(background_color, "RGB")
|
160 |
+
resume_photo = cv2.resize(resume_photo, (RESUME_PHOTO_W, RESUME_PHOTO_H))
|
161 |
+
results.append(resume_photo)
|
162 |
+
|
163 |
+
return results
|
164 |
+
|
165 |
+
|
166 |
+
title = "Resume Photo Maker"
|
167 |
+
|
168 |
+
demo = gr.Interface(
|
169 |
+
fn=make_resume_photo,
|
170 |
+
inputs=[
|
171 |
+
gr.Image(type="numpy", label="input"),
|
172 |
+
gr.ColorPicker(label="background color"),
|
173 |
+
],
|
174 |
+
outputs=gr.Gallery(label="output"),
|
175 |
+
examples=[
|
176 |
+
["images/elon.jpg", "#FFFFFF"],
|
177 |
+
["images/9_Press_Conference_Press_Conference_9_45.jpg", "#FFFFFF"],
|
178 |
+
],
|
179 |
+
title=title,
|
180 |
+
allow_flagging="never",
|
181 |
+
article="<p style='text-align: center;'><a href='https://github.com/bot66/resume-photo-maker' target='_blank'>Github Repo</a></p>",
|
182 |
+
)
|
183 |
+
|
184 |
+
if __name__ == "__main__":
|
185 |
+
demo.launch()
|
images/9_Press_Conference_Press_Conference_9_45.jpg
ADDED
images/elon.jpg
ADDED
models/face_detection_yunet_2023mar.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8f2383e4dd3cfbb4553ea8718107fc0423210dc964f9f4280604804ed2552fa4
|
3 |
+
size 232589
|
models/human_pp_humansegv2_lite_192x192_inference_model.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:34edc335d7833f5a96bb2dadafb1d9da24bac072a26b447c18dd021ea8f29215
|
3 |
+
size 12219997
|
requirements.txt
ADDED
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
aiofiles==23.2.1
|
2 |
+
altair==5.1.2
|
3 |
+
annotated-types==0.6.0
|
4 |
+
anyio==3.7.1
|
5 |
+
attrs==23.1.0
|
6 |
+
certifi==2023.7.22
|
7 |
+
charset-normalizer==3.3.2
|
8 |
+
click==8.1.7
|
9 |
+
colorama==0.4.6
|
10 |
+
coloredlogs==15.0.1
|
11 |
+
contourpy==1.2.0
|
12 |
+
cycler==0.12.1
|
13 |
+
exceptiongroup==1.1.3
|
14 |
+
fastapi==0.104.1
|
15 |
+
ffmpy==0.3.1
|
16 |
+
filelock==3.13.1
|
17 |
+
flatbuffers==23.5.26
|
18 |
+
fonttools==4.44.0
|
19 |
+
fsspec==2023.10.0
|
20 |
+
gradio==4.3.0
|
21 |
+
gradio_client==0.7.0
|
22 |
+
h11==0.14.0
|
23 |
+
httpcore==1.0.1
|
24 |
+
httpx==0.25.1
|
25 |
+
huggingface-hub==0.19.2
|
26 |
+
humanfriendly==10.0
|
27 |
+
idna==3.4
|
28 |
+
importlib-resources==6.1.1
|
29 |
+
Jinja2==3.1.2
|
30 |
+
jsonschema==4.19.2
|
31 |
+
jsonschema-specifications==2023.7.1
|
32 |
+
kiwisolver==1.4.5
|
33 |
+
markdown-it-py==3.0.0
|
34 |
+
MarkupSafe==2.1.3
|
35 |
+
matplotlib==3.8.1
|
36 |
+
mdurl==0.1.2
|
37 |
+
mpmath==1.3.0
|
38 |
+
numpy==1.26.1
|
39 |
+
onnxruntime==1.16.1
|
40 |
+
opencv-python==4.8.1.78
|
41 |
+
orjson==3.9.10
|
42 |
+
packaging==23.2
|
43 |
+
pandas==2.1.2
|
44 |
+
Pillow==10.1.0
|
45 |
+
protobuf==4.25.0
|
46 |
+
pydantic==2.4.2
|
47 |
+
pydantic_core==2.10.1
|
48 |
+
pydub==0.25.1
|
49 |
+
Pygments==2.16.1
|
50 |
+
pyparsing==3.1.1
|
51 |
+
python-dateutil==2.8.2
|
52 |
+
python-multipart==0.0.6
|
53 |
+
pytz==2023.3.post1
|
54 |
+
PyYAML==6.0.1
|
55 |
+
referencing==0.30.2
|
56 |
+
requests==2.31.0
|
57 |
+
rich==13.6.0
|
58 |
+
rpds-py==0.12.0
|
59 |
+
semantic-version==2.10.0
|
60 |
+
shellingham==1.5.4
|
61 |
+
six==1.16.0
|
62 |
+
sniffio==1.3.0
|
63 |
+
starlette==0.27.0
|
64 |
+
sympy==1.12
|
65 |
+
tomlkit==0.12.0
|
66 |
+
toolz==0.12.0
|
67 |
+
tqdm==4.66.1
|
68 |
+
typer==0.9.0
|
69 |
+
typing_extensions==4.8.0
|
70 |
+
tzdata==2023.3
|
71 |
+
urllib3==2.0.7
|
72 |
+
uvicorn==0.24.0.post1
|
73 |
+
websockets==11.0.3
|