hassansajjad229/data-augmentation-arguments-keras--explained

Real-time data augmentation properties are used to generate the images when dateset is limited.Certain arguments are available in Keras to generate images with different properties.

Data-Augmentation-Arguments-keras--Explained

Real-time data augmentation properties are used to generate the images when dateset is limited.Certain arguments are available in Keras to generate images with different properties.

ImageDataGenerator class is used to generate images using data augmentation.

Arugments of ImageDataGenerator Class is explained step by step,so one will be able to use them to generate dataset. Arugments are explained as follows

1.ROTATION this angle would help to rotate the image between the negative to positive range described by us.For example in present case it would rotate the image between 60 to -60

2.WIDTH-SHIFT AND HEIGHT SHIFT It would help us to shift the image between the described range

3.BRIGHTNESS it is use to control the brightness range of the images

4.ZOOM Zoom the image to any value given below i.e between 0.5 to 2.0

5.SHEAR TRANSFORMATION Image will be distorted along the axis

6.CHANNEL SHIFT It would randomly shift the channel value by the mentioned channel shift range 7.FLIPS Images would be randomly flipped on vertical and horizantal axis

8.NORMALIZATION When features in a dataset have different range of values,in ML,we use technique known as normalization.Normalization change the values numeric columns in dataset to a common scale Standard deviation is use to normalize the data SD tells us about spreat out of values.LOW SD most number are close to mean value Formula for SD is x.mean=x_train.mean() x.train_norm=x.train.std() x_train_norm=(x_train-x_mean)/x.std()

a)Featurewise Normalization Featurewise , data is mostly normalized feature-wise as the aim is to study relationship across samples and being able to predict well about new samples

b)Samplewise Normalization Samplewise normalization take each sample and normalize it's features such as the end up being a unit vector