-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathVGG_Face_Resnet50.py
executable file
·156 lines (108 loc) · 5.13 KB
/
VGG_Face_Resnet50.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
from collections import defaultdict
from glob import glob
from random import choice, sample
import cv2
import numpy as np
import pandas as pd
import keras
from keras import backend as K
from keras.callbacks import ModelCheckpoint, ReduceLROnPlateau
from keras.layers import Input, Dense, GlobalMaxPool2D, GlobalAvgPool2D, Concatenate, Multiply, Dropout, Subtract
from keras.models import Model, load_model
from keras.optimizers import Adam
from keras_vggface.utils import preprocess_input
from keras_vggface.vggface import VGGFace
train_file_path = "../../train_relationships.csv"
train_folders_path = "../../train/"
val_famillies = "F09"
all_images = glob(train_folders_path + "*/*/*.jpg")
train_images = [x for x in all_images if val_famillies not in x]
val_images = [x for x in all_images if val_famillies in x]
train_person_to_images_map = defaultdict(list)
ppl = [x.split("/")[-3] + "/" + x.split("/")[-2] for x in all_images]
for x in train_images:
train_person_to_images_map[x.split("/")[-3] + "/" + x.split("/")[-2]].append(x)
val_person_to_images_map = defaultdict(list)
for x in val_images:
val_person_to_images_map[x.split("/")[-3] + "/" + x.split("/")[-2]].append(x)
relationships = pd.read_csv(train_file_path)
relationships = list(zip(relationships.p1.values, relationships.p2.values))
relationships = [x for x in relationships if x[0] in ppl and x[1] in ppl]
train = [x for x in relationships if val_famillies not in x[0]]
val = [x for x in relationships if val_famillies in x[0]]
def read_img(path):
img = cv2.imread(path)
img = np.array(img).astype(np.float)
return preprocess_input(img, version=2)
def gen(list_tuples, person_to_images_map, batch_size=16):
ppl = list(person_to_images_map.keys())
while True:
batch_tuples = sample(list_tuples, batch_size // 2)
labels = [1] * len(batch_tuples)
while len(batch_tuples) < batch_size:
p1 = choice(ppl)
p2 = choice(ppl)
if p1 != p2 and (p1, p2) not in list_tuples and (p2, p1) not in list_tuples:
batch_tuples.append((p1, p2))
labels.append(0)
for x in batch_tuples:
if not len(person_to_images_map[x[0]]):
print(x[0])
X1 = [choice(person_to_images_map[x[0]]) for x in batch_tuples]
X1 = np.array([cv2.resize(read_img(x), dsize=(198, 198), interpolation=cv2.INTER_CUBIC) for x in X1])
X2 = [choice(person_to_images_map[x[1]]) for x in batch_tuples]
X2 = np.array([cv2.resize(read_img(x), (198, 198), interpolation=cv2.INTER_CUBIC) for x in X2])
yield [X1, X2], labels
def baseline_model():
input_1 = Input(shape=(198, 198, 3))
input_2 = Input(shape=(198, 198, 3))
base_model = VGGFace(model='resnet50', include_top=False)
base_model.summary()
#base_model.load_weights('rcmalli_vggface_tf_notop_vgg16.h5')
for x in base_model.layers[:-3]:
x.trainable = True
x1 = base_model(input_1)
x2 = base_model(input_2)
# x1_ = Reshape(target_shape=(7*7, 2048))(x1)
# x2_ = Reshape(target_shape=(7*7, 2048))(x2)
#
# x_dot = Dot(axes=[2, 2], normalize=True)([x1_, x2_])
# x_dot = Flatten()(x_dot)
x1 = Concatenate(axis=-1)([GlobalMaxPool2D()(x1), GlobalAvgPool2D()(x1)])
x2 = Concatenate(axis=-1)([GlobalMaxPool2D()(x2), GlobalAvgPool2D()(x2)])
x3 = Subtract()([x1, x2])
x3 = Multiply()([x3, x3])
x = Multiply()([x1, x2])
x = Concatenate(axis=-1)([x, x3])
x = Dense(512, activation="relu")(x)
x = Dropout(0.3)(x)
x = Dense(1024, activation="relu")(x)
x = Dropout(0.5)(x)
out = Dense(1, activation="sigmoid")(x)
model = Model([input_1, input_2], out)
model.compile(loss="binary_crossentropy", metrics=['acc'], optimizer=Adam(0.00001))
model.summary()
return model
file_path = "vgg_face_out.h5"
checkpoint = ModelCheckpoint(file_path, monitor='val_acc', verbose=1, save_best_only=True, mode='max')
reduce_on_plateau = ReduceLROnPlateau(monitor="val_acc", mode="max", factor=0.1, patience=20, verbose=1)
callbacks_list = [checkpoint, reduce_on_plateau]
model = baseline_model()
model.fit_generator(gen(train, train_person_to_images_map, batch_size=16), use_multiprocessing=True,
validation_data=gen(val, val_person_to_images_map, batch_size=16), epochs=50, verbose=2,
workers=4, callbacks=callbacks_list, steps_per_epoch=200, validation_steps=100)
test_path="../../test/"
def chunker(seq, size=32):
return (seq[pos:pos + size] for pos in range(0, len(seq), size))
from tqdm import tqdm
submission = pd.read_csv('../../sample_submission.csv')
predictions = []
for batch in tqdm(chunker(submission.img_pair.values)):
X1 = [x.split("-")[0] for x in batch]
X1 = np.array([cv2.resize(read_img(test_path + x), dsize=(198, 198), interpolation=cv2.INTER_CUBIC) for x in X1])
X2 = [x.split("-")[1] for x in batch]
X2 = np.array([cv2.resize(read_img(test_path + x), dsize=(198, 198), interpolation=cv2.INTER_CUBIC) for x in X2])
pred = model.predict([X1, X2]).ravel().tolist()
predictions += pred
submission['is_related'] = predictions
submission.to_csv("vgg_face32.csv", index=False)