Description

METHOD

The downloaded hand gesture images dataset will be classified in three steps: 

The first step:

create variable containing data from dataset using base_path . command : 

base_path = '/content/drive/My Drive/datasets/RockPaperScissor/rps-cv-images'

Second step:

convert data into array using ImageDataGenerator function from Keras library:

from keras.preprocessing.image import ImageDataGenerator

Third step: 

create your own generator with the code below

generator = ImageDataGenerator(

    horizontal_flip = True,

    vertical_flip = True,

    height_shift_range = .2,

    validation_split = 0.2

)

Figure 1 ImageDataGenerator

in the code above there is a value of 0.2 which means the dataset will be divided into 80% for training, and 20% for validation

horizontal_flip, vertical_flip, and height_shift_range is another function of ImageDataGenerator() to change image position.

The dataset is divided into 2 parts, train_image for training and val_image for validation

Adding subset='validation' indicates that the validation data uses 20% of the dataset as defined in the generator

Figure 2 train_image_code 

target_size=(224,224) will make the image size 224 x 224 to equalize all images

class_mode made 'categorical' because we will define 3 classes: rock, scissors, and paper. If you only need 2 classes, then it's better to use 'binary’

shuffle will fetch data randomly if it is True

Figure 3 val_image_code

RESULT

when the program code is execute it produces output

Figure 4  Output Program

CONCLUSION

The dataset contains a total of 2188 images corresponding to the 'Rock' (726 images), 'Paper' (710 images) and 'Scissors' (752 images) hand gestures of the Rock-Paper-Scissors game

on the data training found 1751 images, and data validation found 437 images

The images are separated into three sub-folders named 'rock', 'paper', and 'scissors' according to their respective class