深度残差网络：ResNet

ResNet:

一、介绍

caffe-fast-rcnn（Caffe、FSRCNN、FastRCNN）

name: "ResNet_50_1by2"
layer {
name: "data"
type: "Input"
top: "data"
input_param { shape: { dim: 1 dim: 3 dim: 224 dim: 224 } }
//　第一个维度是图片数，第二个是通道数，后面的是图片的长宽
}
layer {
name: "conv_1"
type: "Convolution"
bottom: "data"
top: "conv_1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
pad: 3
kernel_size: 7
stride: 2
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0
}
}
}

shape {

dim: 1 #num，可自行定义

dim: 3 #通道数，表示RGB三个通道

dim: 32 #图像的长和宽，通过 _train_test.prototxt文件中数据输入层的crop_size获取

dim: 32}

二、训练

http://www.cnblogs.com/ml-cv/p/5719531.html 深度残差网(deep residual networks)的训练过程

1、下载基于python的训练代码：

https://github.com/dnlcrl/deep-residual-networks-pyfunt



2、pyfunt需要安装：

@ubuntu:~$ sudo pip install git+git://github.com/dnlcrl/PyFunt.git
Downloading/unpacking git+git://github.com/dnlcrl/PyFunt.git
Cloning git://github.com/dnlcrl/PyFunt.git to /tmp/pip-MS88tP-build

customize UnixCCompiler

warning: no files found matching 'setupegg.py'
warning: no files found matching 'bscript'
warning: no files found matching 'bento.info'
warning: no files found matching '*' under directory 'doc'
warning: no files found matching 'tox.ini'
warning: no previously-included files matching '*_subr_*.f' found under directory 'pyfunt/linalg/src/id_dist/src'
no previously-included directories found matching 'doc/build'
no previously-included directories found matching 'doc/source/generated'
no previously-included directories found matching '*/__pycache__'
warning: no previously-included files matching '*~' found anywhere in distribution
warning: no previously-included files matching '*.bak' found anywhere in distribution
warning: no previously-included files matching '*.swp' found anywhere in distribution
warning: no previously-included files matching '*.pyo' found anywhere in distribution
Successfully installed numpy tqdm cython torchfile pyfunt
Cleaning up...

3、

@ubuntu:~/deep-residual-networks-pyfunt$ git clone https://github.com/dnlcrl//PyDatSet Cloning into 'PyDatSet'...
remote: Counting objects: 185, done.
remote: Total 185 (delta 0), reused 0 (delta 0), pack-reused 185
Receiving objects: 100% (185/185), 29.90 KiB | 11.00 KiB/s, done.
Resolving deltas: 100% (111/111), done.
Checking connectivity... done.

@ubuntu:~/deep-residual-networks-pyfunt/PyDatSet$ sudo python setup.py install
[sudo] password for wei:
/usr/lib/python2.7/distutils/dist.py:267: UserWarning: Unknown distribution option: 'install_requires'
warnings.warn(msg)
running install
running build
running build_py
running install_lib
creating /usr/local/lib/python2.7/dist-packages/pydatset
copying build/lib.linux-x86_64-2.7/pydatset/gtsrb.py -> /usr/local/lib/python2.7/dist-packages/pydatset
copying build/lib.linux-x86_64-2.7/pydatset/__init__.py -> /usr/local/lib/python2.7/dist-packages/pydatset
copying build/lib.linux-x86_64-2.7/pydatset/sfddd.py -> /usr/local/lib/python2.7/dist-packages/pydatset
copying build/lib.linux-x86_64-2.7/pydatset/tiny_imagenet.py -> /usr/local/lib/python2.7/dist-packages/pydatset
copying build/lib.linux-x86_64-2.7/pydatset/cifar10.py -> /usr/local/lib/python2.7/dist-packages/pydatset
copying build/lib.linux-x86_64-2.7/pydatset/mnist.py -> /usr/local/lib/python2.7/dist-packages/pydatset
copying build/lib.linux-x86_64-2.7/pydatset/data_augmentation.py -> /usr/local/lib/python2.7/dist-packages/pydatset
byte-compiling /usr/local/lib/python2.7/dist-packages/pydatset/gtsrb.py to gtsrb.pyc
byte-compiling /usr/local/lib/python2.7/dist-packages/pydatset/__init__.py to __init__.pyc
byte-compiling /usr/local/lib/python2.7/dist-packages/pydatset/sfddd.py to sfddd.pyc
byte-compiling /usr/local/lib/python2.7/dist-packages/pydatset/tiny_imagenet.py to tiny_imagenet.pyc
byte-compiling /usr/local/lib/python2.7/dist-packages/pydatset/cifar10.py to cifar10.pyc
byte-compiling /usr/local/lib/python2.7/dist-packages/pydatset/mnist.py to mnist.pyc
byte-compiling /usr/local/lib/python2.7/dist-packages/pydatset/data_augmentation.py to data_augmentation.pyc
running install_egg_info
Writing /usr/local/lib/python2.7/dist-packages/pydatset-0.1.egg-info
wei@ubuntu:~/deep-residual-networks-pyfunt/PyDatSet$

https://www.cs.toronto.edu/%7Ekriz/cifar.html The CIFAR-10 dataset

Download
If you're going to use this dataset, please cite the tech report at the bottom of this page.
Version     Size    md5sum
CIFAR-10 python version     163 MB  c58f30108f718f92721af3b95e74349a
CIFAR-10 Matlab version     175 MB  70270af85842c9e89bb428ec9976c926
CIFAR-10 binary version (suitable for C programs)   162 MB  c32a1d4ab5d03f1284b67883e8d87530

参考资料：

http://blog.csdn.net/forest_world/article/details/53035009 LeNet、AlexNet、GoogLeNet、VGG、ResNet

http://www.cnblogs.com/daihengchen/p/5761304.html 使用caffe测试自己的图片

http://blog.csdn.net/lg1259156776/article/details/52550865 神经网络与深度学习 Caffe部署中的几个train-test-solver-prototxt-deploy等说明<三>

http://www.kaiminghe.com/ Kaiming He

http://blog.csdn.net/sunbaigui/article/details/50906002 [caffe]深度学习之MSRA图像分类模型Deep Residual Network(深度残差网络)解读

http://blog.csdn.net/yichenmoyan/article/details/51885433 使用Keras搭建深度残差网络

http://blog.csdn.net/heyongluoyao8/article/details/52478715 梯度下降优化算法综述

http://mp.weixin.qq.com/s?__biz=MzIzNDQyNjI5Mg==&mid=100000125&idx=1&sn=72ba0e3e301281c13349f1a1821bad0d&chksm=68f7dba65f8052b0762594489c785ed67f19e111cf2c44dc4522e941989e85d8ee2a03203d26&mpshare=1&scene=23&srcid=1202GeZsjHGcHixoK41RU2mS#rd 深度学习所有文章