Ultrack's CTC submission

This repository contains the code to reproduce the results of the Ultrack submission to the Cell Tracking Challenge.

For more information about the challenge see https://celltrackingchallenge.net

For more information about Ultrack see https://github.com/royerlab/ultrack

Requirements

• Docker: https://docs.docker.com/get-docker/
• NVIDIA GPU with at least 12 GB of RAM. Execution tested on 24 GB.
• Gurobi Web License Service (WLS) because it's running in a docker container. See https://github.com/royerlab/ultrack/tree/main/docker#gurobi-support for additional instructions.

Setup

1. Clone the repository

   git clone https://github.com/royerlab/ultrack_CTC_submission

2. Change to the repository directory

   cd ultrack_CTC_submission

3. Download pre-trained weights

   bash download.sh

4. Build docker image

   docker build -t ultrack_ctc .

Reproducing CTC submission

Once setup is complete, you can run the experiments by running the docker container.

docker run --rm -it --gpus all \
    -v weights/:/app/weights \
    -v <YOUR_DATA_DIR>:/app/data \
    -v <LARGE_DATA_STORAGE>:/wkdir \
    -e WK_DIR=/wkdir \
    -v <PATH TO YOUR GUROBI WSL LICENSE>:/opt/gurobi/gurobi.lic \
    ultrack-ctc

The variable $WK_DIR and /wkdir are useful to store intermediate results, for the TRIF dataset this can be more than 500GBs, which otherwise will be saved in your main storage unit.

General usage

For general usage and application of our tracking algorithm on your own data see https://github.com/royerlab/ultrack.

Additional notes

To train your own model, see train.sh and train.py. This requires at least 24 GB of GPU memory depending on the dataset.