This repository is a fork of PFENet with minimal changes to use their architecture. All credit belongs to the authors of the original repository. Please refer to their original README for more information.
This repository and the Notebooks expect
-
The backbones must be placed in an
initmodel/folder (please refer to the original README ). -
A folder a folder
dataset/must exist, and contain the dataset from Standardized evaluation methodology and reference database for evaluating IVUS image segmentation. Please refer to the authors of their original paper for the dataset. -
We provide weights pretrained using Dataset A from 1, designed to be used for one-shot predictions of the lumen of Dataset B from 1. For other weigths please refer to the original README.
In descriptions on how to use the model and predict IVUS imaging are described in the notebooks Understanding PFENet and Novel Data.
To train the model execute sh test.sh ivus split0_resnet50. Make sure to
modify config/ivus/ivus_split0_resnet50.yaml to use the correct parameters.
Fine-tuning is possible setting weights to the location of the pretrained
weights.