Barlow Adaptor

This repository contains the code for Cross-Dataset Adaptation for Instrument Classification in Cataract Surgery Videos, accepted at MICCAI 2023

Environment File

Create a new conda environment with the config file given in the repository as follows:

conda env create --file=VIU.yaml
conda activate VIU

Data Directory Structure

root
|--train
    |--label1
        |--img
        |--img
        ...
    |--label2
    ...
|--val
    |--label1
        |--img
        |--img
        ...
    |--label2
    ...
|--test
    |--label1
        |--img
        |--img
        ...
    |--label2
    ...

General file descriptions

• transform_utils.py - data transforms defined here
• data_utils.py - functions to generate dataloaders for different datasets
• model.py - model architectures defined here
• train.py - driver code for training and testing model
• test.py - driver code for testing models

Example Usage for Training

python train.py <dataset1 name> <dataset2 name> <save_path> "cuda:0" 16 0.0001 1e-3 0.1 loss7 True 1

Citation

@misc{paranjape2023crossdataset,
      title={Cross-Dataset Adaptation for Instrument Classification in Cataract Surgery Videos}, 
      author={Jay N. Paranjape and Shameema Sikder and Vishal M. Patel and S. Swaroop Vedula},
      year={2023},
      eprint={2308.04035},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}