Why does tf.read_file() and tf.decode_raw() change size?

Asked Aug 07 '19 at 02:14

Active Aug 07 '19 at 09:22

Viewed 448 times

I want to upload an npy file to tensor. So I used tf.decode_raw() after tf.read_file(), but the value and size are different from the original npy file. Can't improve this? ( I know it works with tf.convert_to_tensor(), but I think I should use read_file() to do tf.train.slice_input_producer() for running deep learning. )

file_reader = tf.read_file('./data/test_LR/T2_npy/T2_168_C6M5_1.npy')
ab = tf.decode_raw(file_reader, out_type = tf.uint8)
sess = tf.Session()
a = sess.run(ab)
print(np.shape(a))

# result
(49280,)

an = np.load('./data/test_LR/T2_npy/T2_168_C6M5_1.npy')
print(np.shape(an))

# result
(192,256) -> 49152

edited Aug 07 '19 at 09:22

DrBwts

3,470
6
38
62

asked Aug 07 '19 at 02:14

LUCY

My guess is that `npy` files store a header. – Mateen Ulhaq Aug 07 '19 at 02:28
Possible duplicate of [What is the way data is stored in \*.npy?](https://stackoverflow.com/questions/4090080/what-is-the-way-data-is-stored-in-npy) – Mateen Ulhaq Aug 07 '19 at 02:29
See also: https://stackoverflow.com/questions/48889482/feeding-npy-numpy-files-into-tensorflow-data-pipeline – Mateen Ulhaq Aug 07 '19 at 02:30
1

Thank you so much for your advice!! It works well and have a nice day :) – LUCY Aug 07 '19 at 07:01

Why does tf.read_file() and tf.decode_raw() change size?

0 Answers0