Data Augmentation and Optimizer Tuning for Polyp Segmentation

Automatic segmentation of polyps is a very challenging problem in the scope of medical imaging. This challenge is often faced due to a lack of quality datasets. In this research, we explore the effect on the accuracy of colorectal cancer polyp segmentation due to augmentation and various optimizers while training the segmentation model. The effect of augmentation and optimizers on the polyp segmentation is studied separately. The augmentation effect is observed by changing the percentage of augmentation in each experiment whereas the optimizer effect is studied by changing the optimizer for each experiment. The experiments are performed with 8 optimizers and 10 different augmentation strategies.