Loading other images into tensorflow, apart from MNIST

Question

Iam interested in applying a convolutional neural networks using tensorflow. However, the only tutorial i have seen is loading the MNIST dataset. I have tried replicating the procedure done there and reading chunks of tutorials around the interwebs but it's not working. Here is my code so far

import tensorflow as tf
import os
import numpy as np

filename = os.getcwd() + '/sample_images/*.png'
filename_queue = tf.train.string_input_producer(tf.train.match_filenames_once(filename))

image_reader = tf.WholeFileReader()

_, image_file = image_reader.read(filename_queue)


image = tf.image.decode_png(image_file, 3)
image = tf.image.rgb_to_grayscale(image, name=None)
image = tf.image.resize_images(image, 28, 28, method=0, align_corners=False)

data = []
with tf.Session() as sess:
    tf.initialize_all_variables().run()

    coord = tf.train.Coordinator()
    threads = tf.train.start_queue_runners(coord=coord)

    image_tensor = sess.run([image])
    data.append(image_tensor)
    #print(image_tensor)
    coord.request_stop()
    coord.join(threads)

xx = np.asarray(data)
print xx[0].shape

Basically, i want to do the following:- - Load images in from the folder with their names

• resize each image to 28*28

• change it to grey scale

• turn it into a tensor and add it to a training set

• create it's target(from it's label and add it to a numpy array)

• repeat for all images in the folder

when i am done, pass the dataset and target to a tensorflow RNN

All help will be highly appreciated

Answer 1

Take a look at the generic_input_producer of TensorVision. The input function takes a list of images and returns a tensor representing a batch of images. Preprocessing, including resize is also done in there.