Tensorflow - mnist neural network testing with my own images

Question

I am trying to write a script that will allow me to draw an image of a digit, and then determine which digit matches the model trained by MNIST.

Here is my code:

import random import image from tensorflow.examples.tutorials.mnist import input_data import tensorflow as tf import numpy as np import scipy.ndimage mnist = input_data.read_data_sets("MNIST_data/", one_hot=True) x = tf.placeholder(tf.float32, [None, 784]) W = tf.Variable(tf.zeros([784, 10])) b = tf.Variable(tf.zeros([10])) y = tf.nn.softmax(tf.matmul(x, W) + b) y_ = tf.placeholder(tf.float32, [None, 10]) cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y), reduction_indices=[1])) train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy) init = tf.initialize_all_variables() sess = tf.Session() sess.run(init) for i in range(1000): batch_xs, batch_ys = mnist.train.next_batch(1000) sess.run(train_step, feed_dict= {x: batch_xs, y_: batch_ys}) print ("done with training") data = np.ndarray.flatten(scipy.ndimage.imread("im_01.jpg", flatten=True)) result = sess.run(tf.argmax(y,1), feed_dict={x: [data]}) print (' '.join(map(str, result))) 

For some reason, the results are always wrong, but have 92% accuracy when I use the standard testing method.

I think the problem may be how I encoded the image:

 data = np.ndarray.flatten(scipy.ndimage.imread("im_01.jpg", flatten=True)) 

I tried looking in tensorflow code for the next_batch () function to see how they did it, but I have no idea how I can compare it with my approach.

The problem may be elsewhere.

Any help would be helpful to make an accuracy of 80 +%.

Answer 1

I found my mistake: it encoded the opposite, black - 255, not 0.

  data = np.vectorize(lambda x: 255 - x)(np.ndarray.flatten(scipy.ndimage.imread("im_01.jpg", flatten=True))) 

Fixed.