image classification of rock paper scissor hands

In this post we are going to setup a simple CNN to be able to classify images of hands playing rock, paper, scissor game. This data-set will be loaded from tensorflow_datasets module

Use pip install tensorflow-datasets if you don’t have this module installed already.

UNDERSTANDING THE DATA-SET

The data-set contains images of people playing the rock, paper scissor games as shown in the picture below.It consist of 2,892 images having only train and test splits. Each image has a shape of [300, 300, 3] with 3 output classes(i.e rock, scissor, paper

LOADING THE DATA-SET

To load the data-set the first thing we will need to do is import the necessary libraries. We will then use the tfds.load() to load (downloads and then load on the first time)our data-set while setting with _info and as_supervised to True.

#import the necessary libraries
	from tensorflow import keras
	import tensorflow as tf
	import os,datetime
	import tensorflow_datasets as tfds

	#Loading the datase
	df, info = tfds.load('rock_paper_scissors', with_info = True, as_supervised = True)
view raw
ROCK,PAPER,SCISSOR hosted with ❤ by GitHub

Let’s try to go over some parts of the code that might not be clear

• While using the tfds.load() on line 8, setting with_info = True returns information about our data-set which is then stored in the variable we declared (i.e info).
• as_supervised = True loads our data-set as a(image, label) tuple structure.

PREPROCESS THE DATA

Remember we have just train and test split we need to get our validation split. We will use 10% of the train data as our validation split.

#10 % of the train data as validation data
	num_validation = 0.1 * info.splits['train'].num_examples

	#Turning it to an integer as a float may cause problem along the way
	num_validation = tf.cast(num_validation, tf.int64)
view raw
ROCK,PAPER,SCISSOR hosted with ❤ by GitHub

Before feeding our data into the CNN it will have to go through some form of preprocessing.

Each pixels of the image in our data-set ranges from 0 to 255 which we will scale to between 0 and 1 with the help of a small function

#deep-learning #ai #machine-learning #deep learning