Skip to content

TIFOSI528/CIFAR-10

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
 
 
 
 
 
 

Repository files navigation

CIFAR-10

CIFAR是一个影响力很大的图像分类数据集,分为了CIFAR-10 和CIFAR-100 两个问题,其中的图片是由Alex Krizhevsky, Vinod Nair和Geoffrey Hinton收集的。

CIFAR-10包含了来自10个不同种类,总共60,000张32x32的彩色图片,每一类含有6000张图片。进一步地,其中50,000张为训练图片,剩余的10,000张为测试图片。

下面是数据集中的样例图片:

使用图片生成器ImageDataGenerator对数据进行提升:

train_datagen = ImageDataGenerator(
	featurewise_center=True,              # Set input mean to 0 over the dataset, feature-wise.
	featurewise_std_normalization=True,         # Divide inputs by std of the dataset, feature-wise.
	rotation_range=20,          # Degree range for random rotations.
	shear_range=0.2,            # Shear Intensity (Shear angle in counter-clockwise direction as radians)
	zoom_range=0.2,             # Range for random zoom.
	fill_mode='nearest',        # Points outside the boundaries of the input are filled according to the given mode.
	horizontal_flip=True,		# Randomly flip inputs horizontally.
	)

使用CNN模型,添加BN和正则化,可以实现~91%的准确率。

下面是模型训练过程中accuracy和loss的变化情况:

可以看到,模型在训练后期出现了过拟合,还需要进一步改进。

Reference

  • Learning Multiple Layers of Features from Tiny Images, Alex Krizhevsky, 2009.

About

CIFAR-10图像分类

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages