Fu-TransHNet

This paper presents a novel deep learning fusion method called Fu-TransHNet, which aims to improve the accuracy of colon polyp segmentation.

Configuration

We adopt NVIDIA GTX3090 for GPU acceleration training

Pytorch = 2.1.2+cu121

Python = 3.8.19

Model Overview

Experiments

GPUs with a memory capacity of 4GB or more are adequate for this experiment. GPU-accelerated training with NVIDIA GeForce RTX 4090 D

1.Preparing necessary data:

Dataset: CVC-ClinicDB, CVC-ColonDB, CVC-EndoScene, ETIS-LaribPolypDB and Kvasir

• Download the dataset and put the unzipped data into . /data.
• In accordance with (https://github.com/ChenxiLuo-code1/Fu-TransHNet), run process_train.py and process_test.py to process the train and test data, producing data_{train, val, test}.npy and mask_{train, val, test}.npy.

2.Training:

• run train_isic.py；Need to change the default save path or other hyperparameters.
• Save the model parameter file *.pth.
• Model checkpoints have been given: “w3-Trans109.pth”

3.Testing:

• run test_isic.py；Need to change the default save path or other hyperparameters.