Search code examples
cntk

CNTK reader for base64 encoded images in Python

I'm migrating training and evaluation configurations for CNTK from Brainscript over to Python. Because our training data is created in a map/reduce framework, I'm storing images and labels in a huge text file that contains the base64 encoded image as one of its columns. That all worked fine, but I still have not found a way of doing the equivalent in Python.

My CNTK.exe configuration is similar to this example configuration:

deserializers = ({
        type = "Base64ImageDeserializer" ; module = "ImageReader"
        file = "myFile.tsv"
...

All Python examples (for example this one) use the ImageDeserializer, which reads from images that are individual files. I have not found anything that sounds like a base64 image deserializer in the Python code of cntk.io.

How can I use base64 encoded images in CNTK via Python?

A related ask: The Brainscript Base64ImageDeserializer accepts files that contain a sequence ID in the first column, which is critical for us to identify individual examples at test time. How can I use that in Python?

Solution

  • The Base64 deserializer support for Python was merged to master. For sample usage please see /bindings/python/cntk/io/tests/io_tests.py test_base64_image_deserializer.

    Regarding getting labels - currently there is no easy way, you can compose CNTKTextFormat that will contain the ids. But this is cumbersome, we are discussing to make this easier.