Passing wrong argument for TensorFlow method - image

Question

I am trying to classify two objects. I would like to get Accuracy and Cross Entropy from the evaluate.py script.
Here is the code I'm trying.
evaluate.py (by tensorflow for poets)
#!/usr/bin/python
#
# Copyright 2017 Google Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import sys
import argparse
import numpy as np
import PIL.Image as Image
import tensorflow as tf
import scripts.retrain as retrain
from scripts.count_ops import load_graph
def evaluate_graph(graph_file_name):
with load_graph(graph_file_name).as_default() as graph:
ground_truth_input = tf.placeholder(
tf.float32, [None, 5], name='GroundTruthInput')
image_buffer_input = graph.get_tensor_by_name('input:0')
final_tensor = graph.get_tensor_by_name('final_result:0')
accuracy, _ = retrain.add_evaluation_step(final_tensor, ground_truth_input)
logits = graph.get_tensor_by_name("final_training_ops/Wx_plus_b/add:0")
xent = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
labels=ground_truth_input,
logits=logits))
#image_dir = 'tf_files/flower_photos'
image_dir = 'tf_files/test_images'
testing_percentage = 10
validation_percentage = 10
validation_batch_size = 100
category = 'testing'
image_lists = retrain.create_image_lists(
image_dir, testing_percentage,
validation_percentage)
class_count = len(image_lists.keys())
ground_truths = []
filenames = []
for label_index, label_name in enumerate(image_lists.keys()):
for image_index, image_name in enumerate(image_lists[label_name][category]):
image_name = retrain.get_image_path(
image_lists, label_name, image_index, image_dir, category)
ground_truth = np.zeros([1, class_count], dtype=np.float32)
ground_truth[0, label_index] = 1.0
ground_truths.append(ground_truth)
filenames.append(image_name)
accuracies = []
xents = []
with tf.Session(graph=graph) as sess:
for filename, ground_truth in zip(filenames, ground_truths):
image = Image.open(filename).resize((224, 224), Image.ANTIALIAS)
image = np.array(image, dtype=np.float32)[None, ...]
image = (image - 128) / 128.0
feed_dict = {
image_buffer_input: image,
ground_truth_input: ground_truth}
eval_accuracy, eval_xent = sess.run([accuracy, xent], feed_dict)
accuracies.append(eval_accuracy)
xents.append(eval_xent)
return np.mean(accuracies), np.mean(xents)
if __name__ == "__main__":
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
accuracy, xent = evaluate_graph(*sys.argv[1:])
print('Accuracy: %g' % accuracy)
print('Cross Entropy: %g' % xent)
However when I run the above script for prediction I get the following error:
ValueError: Cannot feed value of shape (1, 224, 224) for Tensor
u'input:0', which has shape '(1, 224, 224, 3)'
How can I solve this error?

Answer 1

It seems like you are feeding grayscale images into input placeholder. Grayscale images have only 1 channel, hence the shape (224, 224, ) (dimension of size 1 is omitted), while pretrained network you are trying requires RGB images with 3 channels and shape (224, 224, 3)
If your images are actualy RGB, you might have an error here:
image = np.array(image, dtype=np.float32)[None, ...]
this indexing: [None, ...] does not seem necessary.
If your images are actually grayscale, you may convert them into RGB format using PIL.convert() (one channel will be repeated 3 times):
image = image.convert("RGB")
Although with channel duplication running 3-channel CNN is inefficient (computation is performed for the same data 3 times) and likely will perform worse than with colored images, this should run the script and will get you on track quickly.