Implementation of Unet(deep learning) using pytorch

---

Overview

Data

The dataset contains bimedical images and labels

Data augmentation

The data contains 38 images(1024*1024) in total. So data augumentation techniques were used along with a non-rigid transformation technique to generate a few more samples for training.

Model

Training

The model is trained for 55 epochs.

Loss function for the training is crossentropy.

Results

Use the trained model to do segmentation on test images, the result is statisfactory. *** Target image ************ Predicted image *************** Original image

---