tf_tutorial_dataset_cifar.py

# -*- coding: utf-8 -*-
"""
Created on Tue May 22 16:25:18 2018

@author: Lin Chen
This file is adapted from github project: 
    image-classification-CIFAR-10-using-tensorflow
https://github.com/TapanBhavsar/image-classification-CIFAR-10-using-tensorflow
"""

import tensorflow as tf
import pickle
import sys
import os
import time
import numpy as np
import glob
import cv2

class_name = ["aeroplane","automobile","bird","cat","deer","dog","frog","horse","ship","truck"]

#this function gives you predicted class name by CNN with highest probabilistic class
def classify_name(predicts):
    max =predicts[0,0]
    temp =0
    for i in range(len(predicts[0])):
        #check higher probable class 
        if predicts[0,i]>max:
                max = predicts[0,i]
                temp = i
    # print higher probale class name
    print(class_name[temp])

def load_dataset(dirpath='data\cifar-10-batches-py'):# give path as example "/home/username/folder name"
    X, y = [], []
    # take data from the data batch
    for path in glob.glob('%s/data_batch_*' % dirpath):
        with open(path, 'rb') as f:
            batch = pickle.load(f, encoding='latin1')#,encoding='latin1' (if gives error of encoding)
        # append all data and labels from the 5 data betch
        X.append(batch['data'])
        y.append(batch['labels'])
    # devide by 255 for making value 0 to 1
    X = np.concatenate(X) /np.float32(255)
    # making labels as int
    y = np.concatenate(y).astype(np.int32)
    #seperate in to RGB colors
    X = np.dstack((X[:, :1024], X[:, 1024:2048], X[:, 2048:]))
    # reshape data into 4D tensor with compatible to CNN model
    X = X.reshape((X.shape[0], 32, 32, 3))
    # initialize labels for training ,validation and testing 
    Y_train = np.zeros((40000,10),dtype = np.float32)
    Y_valid = np.zeros((10000,10), dtype = np.float32)
    y_test = np.zeros((10000,10),dtype = np.int32)
    
    # divide 40000 as training data and it's labels
    X_train = X[-40000:]
    y_train = y[-40000:]
    #devide 10000 as validation data and it's labelss
    X_valid = X[:-40000]
    y_valid = y[:-40000]
    
    # make training labels compatables with CNN model
    for i in range(40000):
        a = y_train[i]
        Y_train[i,a] = 1

    # make validation labels compatables with CNN model
    for i in range(10000):
        a = y_valid[i]
        Y_valid[i,a] = 1
    
    # load test set
    path = '%s/test_batch' % dirpath
    with open(path, 'rb') as f:
        batch = pickle.load(f, encoding='latin1')#,encoding='latin1'
    X_test = batch['data'] /np.float32(255)
    X_test = np.dstack((X_test[:, :1024], X_test[:, 1024:2048], X_test[:, 2048:]))
    X_test = X_test.reshape((X_test.shape[0], 32, 32, 3))
    y_t = np.array(batch['labels'], dtype=np.int32)
    # make test labels compatables with CNN model
    for i in range(10000):
        a = y_t[i]
        y_test[i,a] = 1

    # normalize to zero mean and unity variance
    offset = np.mean(X_train, 0)
    scale = np.std(X_train, 0).clip(min=1)
    X_train = (X_train - offset) / scale
    X_valid = (X_valid - offset) / scale
    X_test = (X_test - offset) / scale
    return X_train, Y_train, X_valid, Y_valid, X_test, y_test