-
Notifications
You must be signed in to change notification settings - Fork 0
/
ssv2.py
146 lines (119 loc) · 5.71 KB
/
ssv2.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
#pip install tensorflow tensorflow-datasets
import os
import tensorflow as tf
import pandas as pd
from tensorflow.keras.preprocessing import image
from tensorflow.keras import layers
# Hyperparameters
IMG_SIZE = 224
BATCH_SIZE = 32
EPOCHS = 10
LEARNING_RATE = 0.001
CLIP_LEN = 8
FRAME_SAMPLE_RATE = 2
class VideoClsDataset(tf.keras.utils.Sequence):
"""Load your own video classification dataset."""
def __init__(self, anno_path, data_path, mode='train', clip_len=8,
frame_sample_rate=2, crop_size=224, short_side_size=256,
new_height=256, new_width=340, keep_aspect_ratio=True,
num_segment=1, num_crop=1, test_num_segment=10, test_num_crop=3):
self.anno_path = anno_path
self.data_path = data_path
self.mode = mode
self.clip_len = clip_len
self.frame_sample_rate = frame_sample_rate
self.crop_size = crop_size
self.short_side_size = short_side_size
self.new_height = new_height
self.new_width = new_width
self.keep_aspect_ratio = keep_aspect_ratio
self.num_segment = num_segment
self.test_num_segment = test_num_segment
self.num_crop = num_crop
self.test_num_crop = test_num_crop
self.dataset_samples, self.label_array = self._load_annotations()
self.data_transform = self._build_transforms()
def _load_annotations(self):
cleaned = pd.read_csv(self.anno_path, header=None, delimiter=' ')
dataset_samples = list(cleaned.values[:, 0])
label_array = list(cleaned.values[:, 1])
return dataset_samples, label_array
def _build_transforms(self):
if self.mode == 'train':
return tf.keras.Sequential([
layers.Resizing(self.short_side_size, self.short_side_size, interpolation='bilinear'),
layers.RandomCrop(self.crop_size, self.crop_size),
layers.RandomFlip('horizontal'),
layers.Rescaling(1./255),
layers.Normalization(mean=[0.485, 0.456, 0.406], variance=[0.229, 0.224, 0.225])
])
elif self.mode == 'validation':
return tf.keras.Sequential([
layers.Resizing(self.short_side_size, self.short_side_size, interpolation='bilinear'),
layers.CenterCrop(self.crop_size, self.crop_size),
layers.Rescaling(1./255),
layers.Normalization(mean=[0.485, 0.456, 0.406], variance=[0.229, 0.224, 0.225])
])
elif self.mode == 'test':
return tf.keras.Sequential([
layers.Resizing(self.short_side_size, self.short_side_size, interpolation='bilinear'),
layers.Rescaling(1./255),
layers.Normalization(mean=[0.485, 0.456, 0.406], variance=[0.229, 0.224, 0.225])
])
def __len__(self):
return len(self.dataset_samples) // BATCH_SIZE
def __getitem__(self, index):
batch_samples = self.dataset_samples[index * BATCH_SIZE:(index + 1) * BATCH_SIZE]
batch_labels = self.label_array[index * BATCH_SIZE:(index + 1) * BATCH_SIZE]
batch_videos = [self.load_video(sample) for sample in batch_samples]
batch_videos = tf.stack(batch_videos, axis=0)
batch_videos = self.data_transform(batch_videos)
batch_labels = tf.convert_to_tensor(batch_labels)
return batch_videos, batch_labels
def load_video(self, sample):
video_path = os.path.join(self.data_path, sample)
video_reader = tf.io.read_file(video_path)
video = tf.io.decode_video(video_reader)
# Sample frames from the video
video = video[::self.frame_sample_rate]
if len(video) < self.clip_len:
pad_len = self.clip_len - len(video)
paddings = tf.constant([[0, pad_len], [0, 0], [0, 0], [0, 0]])
video = tf.pad(video, paddings)
video = video[:self.clip_len]
return video
def build_model(num_classes):
base_model = tf.keras.applications.ResNet50(include_top=False, weights='imagenet', input_shape=(IMG_SIZE, IMG_SIZE, 3))
base_model.trainable = False
model = tf.keras.Sequential([
base_model,
tf.keras.layers.GlobalAveragePooling2D(),
tf.keras.layers.Dense(num_classes, activation='softmax')
])
return model
def compile_model(model):
model.compile(
optimizer=tf.keras.optimizers.Adam(learning_rate=LEARNING_RATE),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
def train_model(model, train_dataset, val_dataset):
history = model.fit(
train_dataset,
epochs=EPOCHS,
validation_data=val_dataset
)
return history
def main():
train_dataset = VideoClsDataset(anno_path='path/to/train_annotations.txt', data_path='path/to/train_videos', mode='train')
val_dataset = VideoClsDataset(anno_path='path/to/val_annotations.txt', data_path='path/to/val_videos', mode='validation')
train_dataset = tf.data.Dataset.from_generator(lambda: train_dataset, output_types=(tf.float32, tf.int32), output_shapes=((BATCH_SIZE, CLIP_LEN, IMG_SIZE, IMG_SIZE, 3), (BATCH_SIZE,)))
val_dataset = tf.data.Dataset.from_generator(lambda: val_dataset, output_types=(tf.float32, tf.int32), output_shapes=((BATCH_SIZE, CLIP_LEN, IMG_SIZE, IMG_SIZE, 3), (BATCH_SIZE,)))
num_classes = len(set(train_dataset.label_array))
model = build_model(num_classes)
compile_model(model)
history = train_model(model, train_dataset, val_dataset)
model.save('ssv2_model.h5')
print("Model trained and saved successfully.")
if __name__ == "__main__":
main()