Trustworthy Endoscopic Super-Resolution

Trustworthy Endoscopic Super-Resolution.
Julio Silva-Rodríguez, Ender Konukoglu ⋅ Computer Vision Lab, ETH Zurich.

Install

• Install in your environment a compatible torch version with your GPU. For example:

conda create -n cfm python=3.11 -y
conda activate cfm
pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu124

git clone https://github.com/******/Endoscopic-CFM.git
cd Endoscopic-CFM
pip install -r requirements.txt

Preparing the datasets

• Configure data paths (see ./docs/local_data/constants.py).
• Download, and configure datasets (see ./docs/local_data/datasets/README.md).

Preparing SR models

• Download, and configure SR models (see .../models_weights/README.md).

Usage

We present the basic usage here.

(a) Run predictions using the super-resolution model:

• python run_msr.py --task surgisr4k_lowres_swinir.yml

(b) Train the Reconstruction Error Network:

• python run_errornet.py --task surgisr4k_lowres_swinir.yml --errornet_config errornet_2layer.yml

(c) Create and validate the Conformal Failure Masks:

• python run_cfm.py --task surgisr4k_lowres_swinir.yml --errornet_config errornet_2layer.yml --target_psnr 22 --alpha 0.05

You will find the final and intermediate results at ./docs/local_data/experiments/.

Citation

If you find this repository useful, please consider citing the following sources.

@inproceedings{endoscopic_cfm,
    title={Trustworthy Endoscopic Super-Resolution},
    author={Julio Silva-Rodríguez and Ender Konukoglu},
    booktitle={arXiv preprint arXiv:2604.18001},
    year={2026}
}