The tensorflow version is too old and the library is not working as expected. You are welcome to add your use-cases in the Issues as Feature request to be considered in the new versions. Sorry for the inconvenience.

A simple wrapper of TensorFlow for Converting, Importing (and Soon, Training) Images in tensorflow.

Installation:

pip install imageflow

Usage:

import imageflow

Just calling the following function will make a filename.tfrecords file in the directory converted_data in your projects root(where you call this method).

convert_images(images, labels, filename)

The images should be an array of shape [-1, height, width, channel] and has the same rows as the labels

# Distorted images for training
images, labels = distorted_inputs(filename='../my_data_raw/train.tfrecords', batch_size=FLAGS.batch_size,
                                      num_epochs=FLAGS.num_epochs,
                                      num_threads=5, imshape=[32, 32, 3], imsize=32)

# Normal images for validation
val_images, val_labels = inputs(filename='../my_data_raw/validation.tfrecords', batch_size=FLAGS.batch_size,
                                    num_epochs=FLAGS.num_epochs,
                                    num_threads=5, imshape=[32, 32, 3])

Dependencies:

• TensorFlow ( => version 0.7.0)
• Numpy
• Pillow