ImageNet Classifier with TensorFlow

• Deep CNN with AlexNet for Classifying/Recognizing objects
• Built using Python, TensorFlow, TFlearn and AWS EC2
• Used g2.8x GPU instances to speed up the classification process
• Built a model with an error rate of 6.2%
• Used factional max pooling with sparse CNN to improve accuracy

Technical Specifications

• Python 2.7
• TensorFlow 0.10.0
• TFLearn 0.2.1
• CUDA 8
• CuDnn v5

Convolutional Neural Network

AlexNet Classifier

Dataset Links

CIFAR 10 dataset

102 Category Flower Dataset

Executing the model

Initiate EC2 instance on AWS with the following specification

• g2.8x
• 80GB SSD
• 32 vCPU

Installing initial dependencies

sudo apt-get update
sudo apt-get -y dist-upgrade
sudo apt-get install python
sudo apt install python3-pip
sudo apt-get install -y libglu1-mesa libxi-dev libxmu-dev libglu1-mesa-dev gcc g++ gfortran build-essential git wget linux-image-generic libopenblas-dev python-dev liblapack-dev libblas-dev build-essential cmake git unzip pkg-config linux-image-generic linux-image-extra-virtual linux-source linux-headers-generic 

Installing compression libraries

sudo apt-get install zlib1g-dev python-imaging

Exporting TensorFlow 0.10.0 for Python 2.7

export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.10.0rc0-cp27-none-linux_x86_64.whl

Installing TensorFlow

sudo pip install --upgrade $TF_BINARY_URL

Installing and upgrading Python and Pip

sudo apt-get install python
sudo apt install python-pip
sudo pip install --upgrade pip

Downloading and installing

wget -qO- https://github.com/tflearn/tflearn/tarball/0.2.1 | tar xvz
cd tflearn-tflearn-a55c1fd/
sudo python setup.py install

Installing scipy stack dependencies

sudo pip install pillow numpy scipy h5py

Install CUDA

https://developer.nvidia.com/cuda-downloads

Installing CuDNN

https://developer.nvidia.com/cudnn

SCP the code into the remote server and run the code using python 2.7

python alex_net.py