Tensorflow: From Building of Database to Training and Prediction of Deep Convolutional Neural Network

This is a very Simple example of showing how to build image dataset from your own collection of images, how to train multiple class classifier using tensorflow CNN and how to predict the class of an object in an unseen image.

This code is designed so that a newbie user become able to train a tensorflow CNN model using his limited CPU resources. By loading small batches of images in the HDF5 data format is the key for doing so.

We took an example of face recognition. The small VGG FACE network is shown in the Figure

Step 1 :

Build HDF5 database from your own images using the following Command. You can set a few simple input parameters in a vggfaces.json file such as input image size, number images in a miniBatch

 python buildDatabase.py

note:

Before building your own training hdf5 dataset, you must put training images using the following directory structure

. ├── rootFolder

└── ClassA
	└── image1.png
	└── image2.png
	└──
	└──
	└── imagen.png
   |
   
   └── ClassB
	└── image1.png
	└── image2.png
	└──
	└──
	└── imagen.png
     |
     |
     |
     |
     
     └── ClassN
	└── image1.png
	└── image2.png
	└──
	└──
	└── imagen.png

Step 2: Train a tensorflow CNN

Train your model executing the following command python trainTesnorCNN.py

Step 3: Predicting a class of unkown image

Using the trained model, you can predict eg the identity of a face in an unseen image

 python predict_TensorFlowCNN.py

Creating your own CNN Archicture

you can define your own CNN by adding your own method to the class ConvNNTF define in multiClassCNN/ConvNN.py file.