Train on a test set and valildate on a validation set - on which the model is not trained - every time we finish training

Hyperparameters tuning based on validation results

what are the hyperparameters? learning rate, batch size, number of epochs, regularization

Test data to evaluate the model

what are different model improvements we can do if the test goes bad? architectures, loss functions, optimizer

Image augmentation

ways to do that: rotating, flipping, cropping, resizing, noise augmentation