Deep-learning libraries include a few functions to manipulate images in the pipeline. For a simple resize or random flip, Keras has you covered. But your ingenuity — or curiosity— outpaces the ready-made.
With your training notebook primed, your model itching to see the first batch, you imagine how your images would respond to that snippit you learned over the weekend. Was it Pillow, Sci-kit Image or OpenCV? Then the real question:
Do imported libraries interfere with your state-of-the-art pipeline?
tf.data pipeline is fantastic, as are the complementary (and complimentary) TPUs. Model training is an iterative process, but it doesn’t…