This repository contains the code for the segmentation of MRI images using U-Net architecture. The dataset used for training the model is the MRI Image Semantic Segmentation Dataset. The model is trained on the training set of the dataset and tested on the validation set. The models are trained using the PyTorch library. They are evaluated using DiceLoss and IOU.
All of the model thats used for the segmentation of MRI images is derived from U-Net Architecture from Segmentation Model Pytorch with [imagenet] as the starting encoder weights. The models that's used are :
- U-Net (EfficientNet-b0)
- EfficientUNet++ (timm-efficientnet-b5)
We also do some modification on the model, which is adding some layers to the model to make it more complex and have more parameters to train.