-
Notifications
You must be signed in to change notification settings - Fork 2
/
split.py
61 lines (48 loc) · 2.05 KB
/
split.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
import tensorflow as tf
import os
files = tf.data.Dataset.list_files("data/*.tfrecord")
ds = tf.data.TFRecordDataset(files)
ds = ds.shuffle( buffer_size=10000)
l = list(ds.as_numpy_iterator())
# The tfrecond tensor format is really basic; we need to know about the fields in the schema
LABELED_TFREC_FORMAT = {
"image/filename": tf.io.FixedLenFeature([], tf.string), # tf.string means bytestring
"image/object/class/text": tf.io.VarLenFeature(tf.string), # shape [] means single element
}
for x in l:
# Show that we can extract filename and tagged classes from each record
record = tf.io.parse_single_example(x, LABELED_TFREC_FORMAT)
filenameTensor = record['image/filename']
filename = bytes.decode(filenameTensor.numpy())
classTensor = record['image/object/class/text']
values = classTensor.values
classNames = []
for v in values:
className = bytes.decode(v.numpy())
classNames.append(className)
print(filename, classNames)
total = len(l)
print("Loaded records in dataset:", total)
evalSize = round(total * 0.20)
trainingSize = total - evalSize
print("Eval size", evalSize)
print("Training size", trainingSize)
# Can't use ds.take and ds.skip because the order is not repeatable; duplicate items
eval = l[0:evalSize]
training = l[evalSize:total]
print("Extracted ", len(eval), " eval items")
print("Left with ", len(training), " training items")
print("Moving eval files")
for x in eval:
record = tf.io.parse_single_example(x, LABELED_TFREC_FORMAT)
filenameTensor = record['image/filename']
filename = bytes.decode(filenameTensor.numpy())
tfrecordFilename = filename.replace('.jpg', '.tfrecord')
os.rename('data/' + tfrecordFilename, 'eval/' + tfrecordFilename)
print("Moving training files")
for x in training:
record = tf.io.parse_single_example(x, LABELED_TFREC_FORMAT)
filenameTensor = record['image/filename']
filename = bytes.decode(filenameTensor.numpy())
tfrecordFilename = filename.replace('.jpg', '.tfrecord')
os.rename('data/' + tfrecordFilename, 'training/' + tfrecordFilename)