microsoft / VoTT

Visual Object Tagging Tool: An electron app for building end to end Object Detection Models from Images and Videos.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[linux pack error] snap release 2.2.0 can't run on manjaro, and export tfrecord always add 'image/encoded' related 42 B header

Neutree opened this issue · comments

Describe the bug

Two issue:

  • snap release 2.2.0 can't run on manjaro 20.1, but v2.1.0 can run normally

  • export tfrecord always add 'image/filename' related 42 B header to image bytes on linux like b'v!\x00\x00\x00\x00\x00\x00\x9f\xdd\n(\n\xf3B\n\xc4>\n\rimage/encoded\x12\xb2>\n\xaf>\n\xac>\xff\xd8\xff\xe0, but no this header on Windows, the same version(v2.1.0), so maybe linux pack issue?

To Reproduce
Steps to reproduce the behavior:

  • open vott v2.1.0, create project
  • export as TFRecord
  • open file with code:
raws = tf.data.TFRecordDataset(tfrecord_files)
# for raw in raws:
#     example = tf.train.Example()
#     example.ParseFromString(raw.numpy())
#     print(example)
feature_description = {
    "image/encoded": tf.io.FixedLenFeature([], tf.string),
}
def _parse_func(example_proto):
    return tf.io.parse_single_example(example_proto, feature_description)
parsed_dataset = raws.map(_parse_func)
for record in parsed_dataset:
    print(record['image/encoded'].numpy()[:128])
    break

Expected behavior

image bytes should startswith 0xff0xd8 if jpeg but not others

Screenshots

  • v2.2.0 run and v2.1.0 run output

image

  • load image bytes:

image

Desktop (please complete the following information):

  • OS: Manjaro 20.1 x86_64 KDE 5.73.0/Plasma5.19.4
  • Browser: local bin