Given 400 images. The data was split into train(60%),test(20%) and validation(20%).
The training set was then augmented using opencv to generate new images and it's correspoinding masks, bringing the total number of images to 3120.
This dataset was trained using Mobilenetv3small-Unet.