卷积神经网络进行图像识别

简介

这是一个机器学习和计算机视觉的项目，框架设计得非常好，需要自己填充核心代码。该项目以一个图像识别任务由浅入深，从实现knn、pca-knn完成30%-40%左右的准确率，到使用多层感知机实现50%的准确率，再到最后使用卷积神经网络实现60%左右的准确率，很有意思。

---

Machine Learning and Computer Vision

Problem 1: Install Tensorflow

Follow the directions on https://www.tensorflow.org/install/ to install Tensorflow on your computer.

Note: You will not need GPU support for this assignment so don’t worry if you don’t have one. Furthermore, installing with GPU support is often more difficult to configure so it is suggested that you install the CPU only version. However, if you have a GPU and would like to install GPU support feel free to do so at your own risk 😃

Note: On windows, Tensorflow is only supported in python3, so you will need to install python3 for this assignment.

Run the following cell to verify your instalation.

import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))

b'Hello, TensorFlow!'

Problem 2: Downloading CIFAR10

Download the CIFAR10 dataset. You will need the python version.
Extract the data to ./data
Once extracted run the following cell to view a few example images.

import numpy as np

# unpickles raw data files
def unpickle(file):
    import pickle
    import sys
    with open(file, 'rb') as fo:
        if sys.version_info[0] < 3:
            dict = pickle.load(fo)
        else:
           dict = pickle.load(fo, encoding='bytes') 
    return dict

# loads data from a single file
def getBatch(file):
    dict = unpickle(file)
    data = dict[b'data'].reshape(-1,3,32,32).transpose(0,2,3,1)
    labels = np.asarray(dict[b'labels'], dtype=np.int64)
    return data,labels

# loads all training and testing data
def getData(path='./data'):
    classes = [s.decode('UTF-8') for s in unpickle(path+'/batches.meta')[b'label_names']]
    
    trainData, trainLabels = [], []
    for i in range(5):
        data, labels = getBatch(path+'/data_batch_%d'%(i+1))
        trainData.append(data)
        trainLabels.append(labels)
    trainData = np.concatenate(trainData)
    trainLabels = np.concatenate(trainLabels)
    
    testData, testLabels = getBatch(path+'/test_batch')
    return classes, trainData, trainLabels, testData, testLabels

# training and testing data that will be used in the following problems
classes, trainData, trainLabels, testData, testLabels = getData()

# display some example images
import matplotlib.pyplot as plt
%matplotlib inline

plt.figure(figsize=(14, 6))
for i in range(14):
    plt.subplot(2,7,i+1)
    plt.imshow(trainData[i])
    plt.title(classes[trainLabels[i]])
plt.show()

print ('train shape: ' + str(trainData.shape) + ', ' + str(trainLabels.shape))
print ('test shape : ' + str(testData.shape) + ', ' + str(testLabels.shape))

train shape: (50000, 32, 32, 3), (50000,)
test shape : (10000, 32, 32, 3), (10000,)

Below are some helper functions that will be used in the following problems.

# a generator for batches of data
# yields data (batchsize, 3, 32, 32) and labels (batchsize)
# if shuffle, it will load batches in a random order
def DataBatch(data, label, batchsize, shuffle=True):
    n = data.shape[0]
    if shuffle:
        index = np.random.permutation(n)
    else:
        index = np.arange(n)
    for i in range(int(np.ceil(n/batchsize))):
        inds = index[i*batchsize : min(n,(i+1)*batchsize)]
        yield data[inds], label[inds]

# tests the accuracy of a classifier
def test(testData, testLabels, classifier):
    batchsize=50
    correct=0.
    for data,label in DataBatch(testData,testLabels,batchsize):
        prediction = classifier(data)
        #print (prediction)
        correct += np.