Overview¶
Old vs. new approach: https://stackoverflow.com/questions/37340129/tensorflow-training-on-my-own-image Since with the introduction of tf.data in release r1.4, we can create a batch of images without placeholders and without queues.
Often it is helpful to read the documentation first! https://www.tensorflow.org/datasets/overview That’s the corresponding git project https://github.com/tensorflow/datasets
However it is sad/more difficult because we (probably) will have a super huge dataset, so check out: https://www.tensorflow.org/datasets/beam_datasets https://beam.apache.org/
Another post, coming with a really nice discription: https://cs230-stanford.github.io/tensorflow-input-data.html#building-an-image-data-pipeline