-
Notifications
You must be signed in to change notification settings - Fork 32
/
metfaces.py
157 lines (136 loc) · 7.22 KB
/
metfaces.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
# Copyright 2020 NVIDIA Corporation
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import argparse
import json
import numpy as np
import os
import PIL.Image
import scipy.ndimage
from tqdm import tqdm
_examples = '''examples:
# Run x, y, z
python %(prog)s --output=tmp
'''
def extract_face(face, source_images, output_dir, rng, target_size=1024, supersampling=4, enable_padding=True, random_shift=0.0, retry_crops=False, rotate_level=True):
def rot90(v) -> np.ndarray:
return np.array([-v[1], v[0]])
# Sanitize facial landmarks.
face_spec = face['face_spec']
landmarks = (np.float32(face_spec['landmarks']) + 0.5) * face_spec['shrink']
assert landmarks.shape == (68, 2)
lm_eye_left = landmarks[36 : 42] # left-clockwise
lm_eye_right = landmarks[42 : 48] # left-clockwise
lm_mouth_outer = landmarks[48 : 60] # left-clockwise
# Calculate auxiliary vectors.
eye_left = np.mean(lm_eye_left, axis=0)
eye_right = np.mean(lm_eye_right, axis=0)
eye_avg = (eye_left + eye_right) * 0.5
eye_to_eye = eye_right - eye_left
mouth_left = lm_mouth_outer[0]
mouth_right = lm_mouth_outer[6]
mouth_avg = (mouth_left + mouth_right) * 0.5
eye_to_mouth = mouth_avg - eye_avg
# Choose oriented crop rectangle.
if rotate_level:
# Orient according to tilt of the input image
x = eye_to_eye - rot90(eye_to_mouth)
x /= np.hypot(*x)
x *= max(np.hypot(*eye_to_eye) * 2.0, np.hypot(*eye_to_mouth) * 1.8)
y = rot90(x)
c0 = eye_avg + eye_to_mouth * 0.1
else:
# Do not match the tilt in the source data, i.e., use an axis-aligned rectangle
x = np.array([1, 0], dtype=np.float64)
x *= max(np.hypot(*eye_to_eye) * 2.0, np.hypot(*eye_to_mouth) * 1.8)
y = np.flipud(x) * [-1, 1]
c0 = eye_avg + eye_to_mouth * 0.1
# Load.
img = PIL.Image.open(os.path.join(source_images, face['source_path'])).convert('RGB')
# Calculate auxiliary data.
qsize = np.hypot(*x) * 2
quad = np.stack([c0 - x - y, c0 - x + y, c0 + x + y, c0 + x - y])
# Keep drawing new random crop offsets until we find one that is contained in the image
# and does not require padding
if random_shift != 0:
for _ in range(1000):
# Offset the crop rectange center by a random shift proportional to image dimension
# and the requested standard deviation (by default 0)
c = (c0 + np.hypot(*x)*2 * random_shift * rng.normal(0, 1, c0.shape))
quad = np.stack([c - x - y, c - x + y, c + x + y, c + x - y])
crop = (int(np.floor(min(quad[:,0]))), int(np.floor(min(quad[:,1]))), int(np.ceil(max(quad[:,0]))), int(np.ceil(max(quad[:,1]))))
if not retry_crops or not (crop[0] < 0 or crop[1] < 0 or crop[2] >= img.width or crop[3] >= img.height):
# We're happy with this crop (either it fits within the image, or retries are disabled)
break
else:
# rejected N times, give up and move to next image
# (does not happen in practice with the MetFaces data)
print('rejected image %s' % face['source_path'])
return
# Shrink.
shrink = int(np.floor(qsize / target_size * 0.5))
if shrink > 1:
rsize = (int(np.rint(float(img.size[0]) / shrink)), int(np.rint(float(img.size[1]) / shrink)))
img = img.resize(rsize, PIL.Image.ANTIALIAS)
quad /= shrink
qsize /= shrink
# Crop.
border = max(int(np.rint(qsize * 0.1)), 3)
crop = (int(np.floor(min(quad[:,0]))), int(np.floor(min(quad[:,1]))), int(np.ceil(max(quad[:,0]))), int(np.ceil(max(quad[:,1]))))
crop = (max(crop[0] - border, 0), max(crop[1] - border, 0), min(crop[2] + border, img.size[0]), min(crop[3] + border, img.size[1]))
if crop[2] - crop[0] < img.size[0] or crop[3] - crop[1] < img.size[1]:
img = img.crop(crop)
quad -= crop[0:2]
# Pad.
pad = (int(np.floor(min(quad[:,0]))), int(np.floor(min(quad[:,1]))), int(np.ceil(max(quad[:,0]))), int(np.ceil(max(quad[:,1]))))
pad = (max(-pad[0] + border, 0), max(-pad[1] + border, 0), max(pad[2] - img.size[0] + border, 0), max(pad[3] - img.size[1] + border, 0))
if enable_padding and max(pad) > border - 4:
pad = np.maximum(pad, int(np.rint(qsize * 0.3)))
img = np.pad(np.float32(img), ((pad[1], pad[3]), (pad[0], pad[2]), (0, 0)), 'reflect')
h, w, _ = img.shape
y, x, _ = np.ogrid[:h, :w, :1]
mask = np.maximum(1.0 - np.minimum(np.float32(x) / pad[0], np.float32(w-1-x) / pad[2]), 1.0 - np.minimum(np.float32(y) / pad[1], np.float32(h-1-y) / pad[3]))
blur = qsize * 0.02
img += (scipy.ndimage.gaussian_filter(img, [blur, blur, 0]) - img) * np.clip(mask * 3.0 + 1.0, 0.0, 1.0)
img += (np.median(img, axis=(0,1)) - img) * np.clip(mask, 0.0, 1.0)
img = PIL.Image.fromarray(np.uint8(np.clip(np.rint(img), 0, 255)), 'RGB')
quad += pad[:2]
# Transform.
super_size = target_size * supersampling
img = img.transform((super_size, super_size), PIL.Image.QUAD, (quad + 0.5).flatten(), PIL.Image.BILINEAR)
if target_size < super_size:
img = img.resize((target_size, target_size), PIL.Image.ANTIALIAS)
# Save face image.
img.save(os.path.join(output_dir, f"{face['obj_id']}-{face['face_idx']:02d}.png"))
def main():
parser = argparse.ArgumentParser(
description='MetFaces dataset processing tool',
epilog=_examples,
formatter_class=argparse.RawDescriptionHelpFormatter
)
parser.add_argument('--json', help='MetFaces metadata json file path', required=True)
parser.add_argument('--source-images', help='Location of MetFaces raw image data', required=True)
parser.add_argument('--output-dir', help='Where to save output files')
parser.add_argument('--random-shift', help='Standard deviation of random crop rectangle jitter', type=float, default=0.0, metavar='SHIFT')
parser.add_argument('--retry-crops', help='Retry random shift if crop rectangle falls outside image (up to 1000 times)', dest='retry_crops', default=False, action='store_true')
parser.add_argument('--no-rotation', help='Keep the original orientation of images', dest='no_rotation', default=False, action='store_true')
args = parser.parse_args()
os.makedirs(args.output_dir, exist_ok=True)
rng = np.random.RandomState(12345) # fix the random seed for reproducibility
with open(args.json, encoding="utf8") as fin:
faces = json.load(fin)
for f in tqdm(faces):
extract_face(f, source_images=args.source_images, output_dir=args.output_dir, rng=rng,
random_shift=args.random_shift, retry_crops=args.retry_crops, rotate_level=not args.no_rotation)
if __name__ == "__main__":
main()