Hand Sign recognition using CNN

The goal of this project is, given a hand sign image, predict the digit shown in image. The preview of image follows:

Quick start:

• clone or download the repository and run signs.py
  signs.py

Outline

1. create basic functions
2. build a model using basic function's
3. running the model to predict

Basic functions:

1. load data : using h5py module data is being loaded
2. intitializing parameters : xavier intialisation
3. create placeholder: useful during running of model
4. feed forward propagation: CONV2D -> RELU -> MAXPOOL -> CONV2D -> RELU -> MAXPOOL -> FLATTEN -> FULLYCONNECTED (architecture of CNN)
5. computing cost : softmax entropy loss
6. random mini batches : creates random mini batches for training
7. one hot vector : to convert the labels into categorial ( as shown in above image)

Model :

• input : training data, testing data, learning rate, num epochs, mini batch size

• ouput(after running the model) : training accuracy, testing accuracy, plot showing the cost function value after each epoch

Results:

• Train Accuracy: 0.98333335
• Test Accuracy: 0.8833333