How Can I Test Own Image To Cifar-10 Tutorial On Tensorflow?

I trained Tensorflow Cifar10 model and I would like to feed it with own single image (32*32, jpg/png). I want to see label and probability of each label as an output, but I having

Solution 1:

The video https://youtu.be/d9mSWqfo0Xw shows an example for classifying a single image.

After the network has already trained by python cifar10_train.py we evaluate the individual image deer6.png of CIFAR-10 database and an own photo of a matchbox. The most important modifications of the original source code of the TF tutorial are the following:

First it is necessary to convert these images to a binary form that the cifar10_input.py can read. It can be done easily by using the code snippet that can be found at How to create dataset similar to cifar-10

Then in order to read the converted images (called input.bin) we need modify the function input() in cifar10_input.py:

  else:
    #filenames = [os.path.join(data_dir, 'test_batch.bin')]
    filenames = [os.path.join(data_dir, 'input.bin')]

(data_dir is equal to './')

Finally to get the label we have modify the function eval_once() in source cifar10_eval.py:

      #while step < num_iter and not coord.should_stop():
      #  predictions = sess.run([top_k_op])
      print(sess.run(logits[0]))
      classification = sess.run(tf.argmax(logits[0], 0))
      cifar10classes = ["airplane", "automobile", "bird", "cat", "deer", "dog", "frog", "horse", "ship", "truck"]
      print(cifar10classes[classification])

      #true_count += np.sum(predictions)
      step += 1

      # Compute precision @ 1.
      precision = true_count / total_sample_count
      # print('%s: precision @ 1 = %.3f' % (datetime.now(), precision))

And of course there are some small modifications that you will need to make.