Data augmentation

Is the process by why you add different prospective to your dataset by changing the image a bit and feeding it back into the network. It can include zooming into the image, cropping it , flipping it upside down and side words. This not only helps in adding variety to the dataset but helps the model learn rather than memorising the data. Some of the techniques include:

Random crop

As the name suggests it randomly crops the image in different sizes

Rotate

Rotates the image in different angles

Dihedral

turns the image by 90 degrees to make 4 images out of a single one

Brightness

adds different images of varying brightness from 0-being black to 1 being white

Mirror Image

Adds a mirror image of the data and feeds it into the network

Jitter

Converts every pixel to its average of neighbouring pixels causing jitter

Built With

• python