[bugfix] - COCO dataset reads images as text files

[hbrylkowski]

After trying to run tensor2tensor on COCO dataset, I got error:

Traceback (most recent call last):
  [...]
  File "/root/.local/lib/python3.5/site-packages/tensor2tensor/bin/t2t_trainer.py", line 401, in <module>
    tf.app.run()
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/platform/app.py", line 125, in run
    _sys.exit(main(argv))
  File "/root/.local/lib/python3.5/site-packages/tensor2tensor/bin/t2t_trainer.py", line 384, in main
    generate_data()
  File "/root/.local/lib/python3.5/site-packages/tensor2tensor/bin/t2t_trainer.py", line 279, in generate_data
    registry.problem(problem_name).generate_data(data_dir, tmp_dir)
  File "/root/.local/lib/python3.5/site-packages/tensor2tensor/data_generators/image_utils.py", line 372, in generate_data
    self.dev_filepaths(data_dir, self.dev_shards, shuffled=False))
  File "/root/.local/lib/python3.5/site-packages/tensor2tensor/data_generators/generator_utils.py", line 490, in generate_dataset_and_shuffle
    generate_files(train_gen, train_paths)
  File "/root/.local/lib/python3.5/site-packages/tensor2tensor/data_generators/generator_utils.py", line 165, in generate_files
    for case in generator:
  File "/root/.local/lib/python3.5/site-packages/tensor2tensor/data_generators/mscoco.py", line 129, in mscoco_generator
    encoded_image_data = f.read()
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/lib/io/file_io.py", line 132, in read
    pywrap_tensorflow.ReadFromStream(self._read_buf, length, status))
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/lib/io/file_io.py", line 100, in _prepare_value
    return compat.as_str_any(val)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/util/compat.py", line 107, in as_str_any
    return as_str(value)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/util/compat.py", line 80, in as_text
    return bytes_or_text.decode(encoding)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xff in position 0: invalid start byte

So my PR only changes to reading images as bytefiles.

[afrozenator]

Thanks @hbrylkowski !

[ZainySong]

Sorry to bother you. Can I ask a question about the t2t-decoder? How should I use the t2t-decoder when the file is image? I used the image transformer, and I trained the model by using Cifar10. But I cannot use the t2t-decoder, there is an error all the time.

/compat.py", line 80, in as_text
return bytes_or_text.decode(encoding)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x8b in position 1: invalid start byte

Can you tell me your command when you use the COCO dataset to do the decode? Or do you know how can I solve my error?