{"id":14947,"date":"2018-05-21T06:00:16","date_gmt":"2018-05-21T06:00:16","guid":{"rendered":"https:\/\/data-flair.training\/blogs\/?p=14947"},"modified":"2021-05-14T11:00:19","modified_gmt":"2021-05-14T05:30:19","slug":"cnn-tensorflow-cifar-10","status":"publish","type":"post","link":"https:\/\/data-flair.training\/blogs\/cnn-tensorflow-cifar-10\/","title":{"rendered":"Convolutional Neural Network (CNN)| CIFAR 10 TensorFlow"},"content":{"rendered":"

In our previous Tensorflow tutorial<\/strong>, we discussed MNIST<\/strong> with TensorFlow. Today we\u2019ll be learning how to build a Convolutional Neural Network (CNN) using TensorFlow in CIFAR 10 Model. <\/span><\/p>\n

Moreover, in this Convolution Neural Network Tutorial, we will see CIFAR 10 CNN TensorFlow model architecture and also the predictions for this model. Along with this, we will learn training and launching of CIFAR 10 model with TensorFlow Convolutional Neural Network example.<\/span><\/p>\n

So, let’s begin the Convolutional Neural Network (CNN) in TensorFlow.<\/p>\n

Working With Convolutional Neural Network<\/span><\/h2>\n

Before we start, it\u2019ll be good to understand the working of a convolutional neural network.<\/span><\/p>\n

Basically, we will be working on the CIFAR 10 dataset, which is a dataset used for object recognition and consists of 60,000 32×32 images which contain one of the ten object classes including aeroplane, automobile, car, bird, dog, frog, horse, ship, and truck.<\/span><\/p>\n

\"Working<\/a>

Working of Convolutional Neural Network<\/p><\/div>\n

You can download the dataset from\u00a0<\/span>https:\/\/www.cs.toronto.edu\/~kriz\/cifar.html<\/span><\/a>. You can use other datasets as well, but for the sake of convenience, CIFAR 10 is used here.<\/span><\/p>\n

CIFAR 10 TensorFlow Model Architecture<\/span><\/h2>\n

This Convolutional neural network Model achieves a peak performance of about 86% accuracy within a few hours of training time on a GPU. Following is a list of the files you\u2019ll be needing: <\/span>
\n<\/i><\/p>\n

cifar10_input.py<\/i><\/strong> \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Reads the native CIFAR-10 binary file format.<\/strong><\/span>
\ncifar10.py <\/i> <\/strong> \u00a0<\/strong>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Builds the CIFAR-10 model.<\/strong><\/span>
\ncifar10_train.py<\/i><\/strong> \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Trains a CIFAR-10 model on a CPU or GPU.<\/strong><\/span>
\ncifar10_multi_gpu_train.py<\/i><\/strong>\u00a0 \u00a0 \u00a0\u00a0Trains a CIFAR-10 model on multiple GPUs<\/strong>.<\/span>
\ncifar10_eval.py<\/strong> \u00a0\u00a0\u00a0\u00a0\u00a0<\/span><\/i> \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Evaluates the predictive performance of a CIFAR-10 model.<\/strong><\/span><\/p>\n

a. Inputs<\/h3>\n

Basically, the input part of the CIFAR 10 CNN TensorFlow model is built by the functions inputs() and distorted_inputs() which read images from the CIFAR 10 binary data files. <\/span><\/p>\n

These files contain fixed byte length records, so you can use <\/span>tf.FixedLengthRecordReader<\/span><\/i>. You can look at Reading Data to learn more about how the Reader class works.<\/span><\/p>\n

First, crop the images are up to 24 x 24 pixels. To make the model insensitive to the dynamic range they are approximately whitened. For training, you can apply a series of random distortions like flipping the image horizontally, changing the brightness or the contrast, to artificially increase the data set size<\/span><\/p>\n

Prediction for CIFAR 10 Model<\/span><\/h2>\n

The prediction part of the CIFAR 10 Convolutional Neural Network model is constructed by the inference() function which adds operations to compute the logic of the predictions. The following are the layers you need to build for the model to work properly:<\/span><\/p>\n