分类任务数据集制作
众所周知，在深度学习图像分类、目标检测等任务的训练过程中需要大量的训练图片；为了方便快速批量的读取数据，往往都会先把图片数据集转化成特殊的文件格式，常见的如：.mat, .npy, .pkl, .csv 等。本文主要介绍如何由现有图片生成.pkl文件。
图片的整理
首先，将图片按类别整理成如下目录格式：
- 第一层目录为/jpg
- 第二层目录如图
显然每个文件夹下放的是各类的图片。
转换程序
# img2pkl.py
from __future__ import division, print_function, absolute_import
import os
import sys
# import tarfile
import numpy as np
import pickle
import random
from PIL import Image
# 主要函数接口
def load_data(dirname="17flowers", pklpath='17flowers/jpg/',
resize_pics=(224, 224), shuffle=True, one_hot=False):
dataset_file = os.path.join(dirname, 'dataset.pkl')
X, Y = build_image_dataset_from_dir(directory=pklpath,
dataset_file=dataset_file,
resize=resize_pics,
filetypes=['.jpg', '.jpeg'],
convert_gray=False,
shuffle_data=shuffle,
categorical_Y=one_hot)
return X, Y
####
def build_image_dataset_from_dir(directory,
dataset_file="dataset.pkl",
resize=None, convert_gray=None,
filetypes=None, shuffle_data=False,
categorical_Y=False):
try:
X, Y = pickle.load(open(dataset_file, 'rb'))
except Exception:
X, Y = image_dirs_to_samples(directory, resize, convert_gray, filetypes)
if categorical_Y:
Y = to_categorical(Y, np.max(Y) + 1) # First class is '0'
if shuffle_data:
X, Y = shuffle(X, Y)
pickle.dump((X, Y), open(dataset_file, 'wb'))
return X, Y
########
def image_dirs_to_samples(directory, resize=None, convert_gray=None,
filetypes=None):
print("Starting to parse images...")
if filetypes:
if filetypes not in [list, tuple]: filetypes = list(filetypes)
samples, targets = directory_to_samples(directory, flags=filetypes)
for i, s in enumerate(samples):
samples[i] = load_image(s)
if resize:
samples[i] = resize_image(samples[i], resize[0], resize[1])
if convert_gray:
samples[i] = convert_color(samples[i], 'L')
samples[i] = pil_to_nparray(samples[i])
samples[i] /= 255.
print("Parsing Done!")
return samples, targets
# 类别one-hot编码
def to_categorical(y, nb_classes):
""" to_categorical.
Convert class vector (integers from 0 to nb_classes)
to binary class matrix, for use with categorical_crossentropy.
Arguments:
y: `array`. Class vector to convert.
nb_classes: `int`. Total number of classes.
"""
y = np.asarray(y, dtype='int32')
if not nb_classes:
nb_classes = np.max(y)+1
Y = np.zeros((len(y), nb_classes))
for i in range(len(y)):
Y[i, y[i]] = 1.
return Y
# 将数据打乱
def shuffle(*arrs):
""" shuffle.
Shuffle given arrays at unison, along first axis.
Arguments:
*arrs: Each array to shuffle at unison as a parameter.
Returns:
Tuple of shuffled arrays.
"""
arrs = list(arrs)
for i, arr in enumerate(arrs):
assert len(arrs[0]) == len(arrs[i])
arrs[i] = np.array(arr)
p = np.random.permutation(len(arrs[0]))
return tuple(arr[p] for arr in arrs)
# 遍历各文件夹和文件夹中的图片
def directory_to_samples(directory, flags=None):
""" Read a directory, and list all subdirectories files as class sample """
samples = []
targets = []
label = 0
classes = sorted(os.walk(directory).next()[1])
for c in classes:
c_dir = os.path.join(directory, c)
for sample in os.walk(c_dir).next()[2]:
if not flags or any(flag in sample for flag in flags):
samples.append(os.path.join(c_dir, sample))
targets.append(label)
label += 1
return samples, targets
# 使用PIL库读取图片
def load_image(in_image):
""" Load an image, returns PIL.Image. """
img = Image.open(in_image)
return img
# 图片尺度变换函数
def resize_image(in_image, new_width, new_height, out_image=None,
resize_mode=Image.ANTIALIAS):
""" Resize an image.
Arguments:
in_image: `PIL.Image`. The image to resize.
new_width: `int`. The image new width.
new_height: `int`. The image new height.
out_image: `str`. If specified, save the image to the given path.
resize_mode: `PIL.Image.mode`. The resizing mode.
Returns:
`PIL.Image`. The resize image.
"""
img = in_image.resize((new_width, new_height), resize_mode)
if out_image:
img.save(out_image)
return img
#
def convert_color(in_image, mode):
""" Convert image color with provided `mode`. """
return in_image.convert(mode)
# 加载图片并转换为numpy多维数组
def pil_to_nparray(pil_image):
""" Convert a PIL.Image to numpy array. """
pil_image.load()
return np.asarray(pil_image, dtype="float32")
if __name__ == "__main__":
load_data(dirname="dataSet", resize_pics=(224, 224), True, True)使用
如下说明了如何使用这个数据转换模块，如果是初次调用将会先生成.pkl文件。
import image2pkl
X, Y = image2pkl.load_data(one_hot=True, resize_pics=(224,224))
# 得到的X是训练样本，Y是对应的标签；用法和MNIST手写体识别的数据集一样也可以直接使用生成的.pkl文件，如下：
'''using .pkl file directly'''
import cPickle as pickle
file = open("dataset.pkl")
data, labels = pickle.load(file)
print data.shape
print labels.shape
# alpha控制训练和测试数据的比例
alpha = 0.3
trainIndex = data.shape[0] - int(data.shape[0] * alpha)
testIndex = int(data.shape[0] * alpha)
trainData = data[:trainIndex]
testData = data[-testIndex:]
trainLabels = labels[:trainIndex]
testLabels = labels[-testIndex:]