[83] Example to run a deep learning demo

src/main/python/tensorframes_snippets/preparation_inceptionv3.py

@@ -0,0 +1,189 @@
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from datetime import datetime
+import math
+import os.path
+import time
+
+
+import numpy as np
+import tensorflow as tf
+
+from preprocessing import inception_preprocessing
+import datasets.imagenet as imagenet
+from nets import inception
+import datasets.dataset_utils as dataset_utils
+
+import tensorflow as tf
+from tensorflow.python.training import saver as tf_saver
+from tensorflow.python.framework import graph_util
+
+slim = tf.contrib.slim
+
+default_image_size = 299
+
+
+####### Download the network data
+# The URL of the checkpointed data.
+url = "http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz"
+# The name of the checkpoint file:
+checkpoint_file = 'inception_v3.ckpt'
+# Specify where you want to download the model to
+checkpoints_dir = '/tmp/checkpoints'
+
+checkpoint_path = os.path.join(checkpoints_dir, checkpoint_file)
+if not tf.gfile.Exists(checkpoints_dir):
+    tf.gfile.MakeDirs(checkpoints_dir)
+
+if not tf.gfile.Exists(checkpoint_path):
+    print('Downloading the model...')
+    dataset_utils.download_and_uncompress_tarball(url, checkpoints_dir)
+
+#### TEST TO REMOVE
+
+s = tf.constant("This is string")
+r = tf.decode_raw(s, tf.int8)
+s2 = tf.as_string(r)
+sess = tf.InteractiveSession()
+print(s2.eval())
+
+###### Building the computation graph
+
+# All this code can be run once. It assembles the computation graph, fills it with the checkpointed
+# coefficients, and then saves it as a protocol buffer description.
+
+# Build the graph
+g = tf.Graph()
+with g.as_default():
+    # Keep for now a placeholder that will eventually be filled with the content of the image.
+    # This code only accepts JPEG images, which is the most common image format.
+    image_string = tf.placeholder(tf.string, [], name="image_input")
+
+    # Decode string into matrix with intensity values
+    image = tf.image.decode_jpeg(image_string, channels=3)
+
+    # Resize the input image, preserving the aspect ratio
+    # and make a central crop of the resulted image.
+    # The crop will be of the size of the default image size of
+    # the network.
+    processed_image = inception_preprocessing.preprocess_image(image,
+                                                         default_image_size,
+                                                         default_image_size,
+                                                         is_training=False)
+
+    # Networks accept images in batches.
+    # The first dimension usually represents the batch size.
+    # In our case the batch size is one.
+    processed_images  = tf.expand_dims(processed_image, 0)
+
+    # Create the model, use the default arg scope to configure
+    # the batch norm parameters. arg_scope is a very conveniet
+    # feature of slim library -- you can define default
+    # parameters for layers -- like stride, padding etc.
+    # Note: like the Arabian nights, inception defines 1001 classes
+    # to include a background class (the first).
+    with slim.arg_scope(inception.inception_v3_arg_scope()):
+        logits, _ = inception.inception_v3(processed_images,
+                               num_classes=1001,
+                               is_training=False)
+
+    # In order to get probabilities we apply softmax on the output.
+    probabilities = tf.nn.softmax(logits)
+
+    # Just focus on the top predictions
+    top_pred = tf.nn.top_k(tf.squeeze(probabilities), k=5, name="top_predictions")
+
+    # These are the outputs we will be requesting from the network.
+    output_nodes = [probabilities, top_pred.indices, top_pred.values]
+
+# Create the saver
+with g.as_default():
+    model_variables = slim.get_model_variables('InceptionV3')
+    saver = tf_saver.Saver(model_variables, reshape=False)
+
+def get_op_name(tensor):
+    return tensor.name.split(":")[0]
+
+# Export the network
+with g.as_default():
+    with tf.Session() as sess:
+        saver.restore(sess, checkpoint_path)
+        # The add_shapes option is important: Spark requires this extra shape information to infor the
+        # correct types.
+        input_graph_def = g.as_graph_def(add_shapes=True)
+        output_tensor_names = [node.name for node in output_nodes]
+        output_node_names = [n.split(":")[0] for n in output_tensor_names]
+        output_graph_def = graph_util.convert_variables_to_constants(
+            sess,
+            input_graph_def,
+            output_node_names,
+            variable_names_blacklist=[])
+
+# The variable 'output_graph_def' now contains all the description of the computation.
+# The variables in the 'output_nodes' list will be used to know what to output.
+
+####### Testing the computation graph
+
+# This code performs a sanity check, by running the network against some image content downloaded from the internet.
+
+g2 = tf.Graph()
+with g2.as_default():
+    tf.import_graph_def(output_graph_def, name='')
+
+#### Download an image
+import requests
+
+# Example picture:
+# Specify where you want to download the model to
+images_dir = '/tmp/image_data'
+
+if not tf.gfile.Exists(images_dir):
+    tf.gfile.MakeDirs(images_dir)
+
+image_url = 'https://upload.wikimedia.org/wikipedia/commons/thumb/8/85/WeaverAntDefense.JPG/640px-WeaverAntDefense.JPG'
+image_url = 'https://www.tensorflow.org/images/cropped_panda.jpg'
+image_path = os.path.join(images_dir, image_url.split('/')[-1])
+
+if not tf.gfile.Exists(image_path):
+    response = requests.get(image_url)
+    if response.status_code == 200:
+        with open(image_path, 'wb') as f:
+            f.write(response.content)
+
+image_data = tf.gfile.FastGFile(image_path, 'rb').read()
+
+with g2.as_default():
+    input_node2 = g2.get_operation_by_name(get_op_name(image))
+    output_nodes2 = [g2.get_tensor_by_name(n) for n in output_tensor_names]
+    with tf.Session() as sess:
+        (probabilities_, indices_, values_) = sess.run(output_nodes2, {'image_input:0':image_data})
+
+names = imagenet.create_readable_names_for_imagenet_labels()
+for i in range(5):
+    index = indices_[i]
+    print('Probability %d %0.2f => [%s]' % (index, values_[i], names[index]))
+
+
+###### Perform some evaluation with TensorFrames
+
+# This code takes the network and a directory that contains some image content. It shows how to process the content
+# using Spark dataframes and Tensorframes.
+
+import tensorframes as tfs
+sc.setLogLevel('INFO')
+
+raw_images_miscast = sc.binaryFiles("file:"+images_dir)
+raw_images = raw_images_miscast.map(lambda x: (x[0], bytearray(x[1])))
+
+df = spark.createDataFrame(raw_images).toDF('image_uri', 'image_data')
+df
+
+with g2.as_default():
+    index_output = tf.identity(g2.get_tensor_by_name('top_predictions:1'), name="index")
+    value_output = tf.identity(g2.get_tensor_by_name('top_predictions:0'), name="value")
+    pred_df = tfs.map_rows([index_output, value_output], df, feed_dict={'image_input':'image_data'})
+
+pred_df.select('index', 'value').show()
+