-
Notifications
You must be signed in to change notification settings - Fork 5
/
convert-to-hdf5.py
35 lines (30 loc) · 1.17 KB
/
convert-to-hdf5.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
import numpy as np
import h5py
import os
from argparse import ArgumentParser
def save_to_hdf5(args):
dataset = args.dataset
# read from disk
# use float 16 to save space
boxes = np.fromfile('data_only_im2p_' +dataset+ '_output.h5-boxes',
sep=" ", dtype=np.float16)
feats = np.fromfile('data_only_im2p_' +dataset+ '_output.h5-feats',
sep=" ", dtype=np.float16)
print dataset, "raw data shape: ", boxes.shape, feats.shape
# reshape data
boxes = np.reshape(boxes, ((boxes.shape[0]/50/4), 50, 4))
feats = np.reshape(feats, ((feats.shape[0]/50/4096), 50, 4096))
print dataset, "converted data shape: ", boxes.shape, feats.shape
# write to hdf5
print 'start writing to h5 file'
h5file = h5py.File(os.path.join("./data",'im2p_' +dataset+ '_output.h5'), 'w')
# redundant boxes
# h5file.create_dataset('boxes', data=boxes)
h5file.create_dataset('feats', data=feats)
h5file.close()
print 'close h5 file'
if __name__ == '__main__':
parser = ArgumentParser()
parser.add_argument('--dataset', default='val')
args = parser.parse_args()
save_to_hdf5(args)