Welcome to the official implementation of "Graph Self-Supervised Learning for Endoscopic Image Matching"

Pre-trained models

Pre-trained models can be downloaded following the link. You may save the models to ./models folder

Download link

Training and testing datasets for endoscopic images matching

Download link

unzip the two zip files in the ./data folder and then update your configuration yaml files accordingly (see below)

Runners and configuration files

"run_trainning.py", configuration yaml file in "config/config_train.yaml"
"run_validation.py", configuration yaml file in "config/config_validation.yaml"
"run_generate_training_data.py", configuration yaml file in "config/config_generate_training_data.yaml"
"run_matching_demo.py", configuration yaml file in "config/config_validation.yaml"

Setup for Dev on local machine

This code base is tested only on Ubuntu 20.04 LTS, TitanV and RTX2080-ti NVIDIA GPUs.

Install local environment and requirements First install Anaconda3 then install the requirements as follows:

conda create -n crns---self-sup-image-matching python=3.8

a new virtual environment is now created in ~/anaconda3/envs/crns---self-sup-image-matching Now activate the virtual environment by running:

source activate crns---self-sup-image-matching

In case you would like stop your venv conda deactivate
To install dependencies, cd to the directory where requirements.txt is located and run the following command in your shell:

cat requirements.txt | xargs -n 1 -L 1 pip3 install

Install docker and nvidia-docker and then run these commands:

nvidia-docker build -t <YOUR_DOCKER_IMAGE_NAME>:<YOUR_IMAGE_VERSION> .

sudo NV_GPU='0' nvidia-docker run -i -t --entrypoint=/bin/bash --runtime=nvidia <YOUR_DOCKER_IMAGE_NAME>:<YOUR_IMAGE_VERSION>

How to use Facial Process on local machine:

To test photos from local folder: run this command python local_processing.py To process photos coming from the front on your machine: run this command python main.py

Git pre-commit hooks

if not already installed from the requirements.txt then first install pre-commit and black using these commands: pip3 install pre-commit and pip3 install black

run pre-commit install to set up the git hook scripts

You can also flake8 <YOURSCRIPT>.py to check if your python script is compliant with the project

or directly fix your script using black <YOURSCRIPT>.py

Docker build

[comming soon]

abenhamadou/graph-self-supervised-learning-for-endoscopic-image-matching