horse007666 / resnet

ResNet Recurrence

This repository contains codes using TensorFlow to recurrence the ResNet CNN

1. https://arxiv.org/pdf/1512.03385.pdf
2. https://arxiv.org/pdf/1603.05027.pdf

datasets: mnist, cifar50

preparation

1. operation system only Ubuntu16 support the newest TensorFlow, check version the operation system

lsb_release -a

update the operation system to Ubuntu 16.04.2 LTS

2. python installation and update must install python3 to use the newest TensorFlow

sudo apt install python3
sudo apt-get install --upgrade python3

3. install pip and pip3 first, install setuptools

get --no-check-certificate  https://pypi.python.org/packages/source/s/setuptools/setuptools-19.6.tar.gz#md5=c607dd118eae682c44ed146367a17e26
tar -zxvf setuptools-19.6.tar.gz
cd setuptools-19.6
python3 setup.py build
sudo python3 setup.py install

second, install pip,pip3

wget --no-check-certificate  https://pypi.python.org/packages/source/p/pip/pip-8.0.2.tar.gz#md5=3a73c4188f8dbad6a1e6f6d44d117eeb
tar -zxvf pip-8.0.2.tar.gz
cd pip-8.0.2

python3 setup.py build
sudo python3 setup.py install

third, update pip,pip3

python3 -m pip install --upgrade pip  

4. begin installing newest TensorFlow(1.8.0)

pip3 install tensorflow    

use the following commands to update python packages

pip list --outdated
sudo pip  install  --upgrade SomePackage

5. install ROOT ROOT is a CERN developed software tool used in high energy physics for big data processing, statistical analysis, visualisation and storage. just follow the website to install ROOT: https://root.cern.ch/

dataset introduction

1. MNIST

2. cifar10

3. ImageNet

using TensorFlow to construct a CNN

the CNN(convolutional neural network) usually contains 3 parts:

1. covolutional layer
2. pooling layer
3. fully connected layer

function of all part:

1. fetch the features of the input images
2. downsample the image

in order to accelerate convergence ReLU used

python3 Plain_Net.py 

using TensorFlow to construct ResNet

python3 Plain_Net.py 

Plot and compare to other result

use ROOT for plotting:

root Plot.c

For the MNIST dataset: The loss function vs. training epoches, the loss function indicate that the CNN converge after about 200 epoches.

use 1000 training data to validate

use 10000 test data set to validate