de(ep)tangle

This repository contains the implementation of Fast detection of slender bodies in high density microscopy data paper.

<p align="center"> <img src="https://github.com/kirkegaardlab/deeptanglelabel/blob/main/docs/figures/tracking.gif" height="236" /> <img src="https://github.com/kirkegaardlab/deeptanglelabel/blob/main/docs/figures/following.gif" height="236" /> <img src="https://github.com/kirkegaardlab/deeptanglelabel/blob/main/docs/figures/dense.png" width="720" /> </p>

Installation

To run the code one must first install the dependencies. You can do this in a virtual environment:

python3 -m venv venv
source venv/bin/activate

Start by installing jax following instructions at their repository. Install the remaining dependencies afterwards:

pip install -r requirements.txt

If you need to use the model and the auxiliary functions outside this repository, you can install it from the root folder by

pip install -e .

Train

To train the model, there is a train script used for the model presented in the paper. The possible arguments can be seen by using the help flag.

python3 train.py --help

An example of a training run would be

python3 train.py --batch_size=32 --eval_interval=10 --nworms=100,200 --save

Usage

Example scripts such as detection and tracking can be found in the examples folder

We include a Dockerfile (cpu only). For linux we provide a script to run the relevant commands:

(sudo) sh docker_run.sh

Weights

The weights used in the paper can be downloaded from here or by using the following commmand

wget https://sid.erda.dk/share_redirect/cEjIpG1yQl -O weights.zip