endomapper/ColonSLAM

Topological SLAM in colonoscopies leveraging deep features and topological priors (MICCAI 2024)

ColonSLAM: Metric-Topological SLAM for colonoscopy

ColonSLAM is a metric-topological SLAM algorithm able to build a topological graph starting from the metric submaps built by CudaSIFT-SLAM. Additionally, it can localize a second sequence of the same patient against the previously built topological map.

Installation

To install the ColonSLAM environment, simply use conda as:

conda create -f environment.yml

Download trained models and evaluation data

First, download the trained models, which can be found here.

The images used for evaluation can be found here.

Usage

To run the topological SLAM run the following command:

cd ColonSLAM
python slam.py --resume=[PATH_TO_MODELS]/endofm_cls_0_0_224_schedule_resize_none/best_model.pth --use_lightglue --experiment_name=colonslam --sim_threshold=0.95 --use_mlp --matches_threshold=100 --window_size=5 --voting_threshold=0.20 --filter

To run the map reuse experiment, run the following command:

cd ColonSLAM
python slam_reuse.py --resume=[PATH_TO_MODELS]/endofm_cls_0_0_224_schedule_resize_none/best_model.pth --experiment_name=colonslam-reuse --sim_threshold=0.80 --use_mlp --mode=slam --voting_threshold=0.15 --sequence_map 027 --start_processing 34 --sequence_loc 035 --window_size=5 --filter 

Related Publications:

Javier Morlana, Juan D. Tardós J.M.M. Montiel, Topological SLAM in colonoscopies leveraging deep features and topological priors, MICCAI 2024. PDF

@inproceedings{morlana2024topological,
  title={Topological SLAM in colonoscopies leveraging deep features and topological priors},
  author={Morlana, Javier and Tard{\'o}s, Juan D and Montiel, Jos{\'e} MM},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={733--743},
  year={2024},
  organization={Springer}
}