us_superres's Introduction

Ultrasound Image Super Resolution

This repository consists of code that can be used to run image super resolution algorithms on ultrasound data. The code is modular in such a way that you can add other networks into the models folder and use them. Please take a look at the comments in the code for specific details.

Details about important folders and files in the repo

Data: Contains dataloading scripts and data trasnforms
Model: Image SR Network scripts
Options: YAML files containing training and test options
build_dataset.py: Reproducible script to download and partition different US datasets
trainer.py: Contains the training code
main.py: Run this file to train the model
test.py: Run this to test .pt model

Training Instructions:

Edit the train YAML file given in the Options folder to provide training options.
- YAML file should contain details about data path, experiment name, network to be used, network design params, learning rate, epochs
- Ensure the Batch size is 1 to avoid memory errors.
- Path to save the results of training must be provided.
Ensure that the data folder contains train and val folders in it. The path to data folder must be specified in the YAML file.
Run the main.py file with the path to YAML file as an argument (-h for help)
- The training progress in printed on the terminal and is logged into a text file.
- A directory with the results of training(log, best model, images) is created at the location specified by the YAML file.

Testing Instructions:

Edit the test YAML file given in the Options folder to provide test options.
- YAML file should contain the test folder path, model path, network structure similar to the model you are testing.
- Path to save the results of training must be provided.
Run the test.py file with the path to YAML file as an argument (-h for help)
- Every test image's PSNR and SSIM is printed on the terminal and is logged into a text file.
- The results directory specified in the YAML will contain a folder with the experiment name. This folder will have the model predictions for every test image along with a log file.

For queries, contact [email protected]

us_superres's People