MT3

Introduction

This repository contains the code for the paper "Next Generation Multitarget Trackers: Random Finite Set Methods vs Transformer-based Deep Learning" (https://arxiv.org/abs/2104.00734). The code for MT3 was developed as joint effort by Juliano Pinto, Georg Hess, William Ljungbergh, Yuxuan Xia, Lennart Svensson and Henk Wymeersch, and was partially based on the code available at the repositories for DETR and Deformable-DETR.

Setting up

In order to set up a conda environment with all the necessary dependencies, run the command:

 conda env create -f conda-env/environment-<gpu/cpu>.yml

Running experiments

Run an experiment using the train.py script. Example usage:

src/training.py -tp configs/tasks/task1.yaml -mp configs/models/mt3.yaml

Training hyperparameters such as batch size, learning rate, checkpoint interval, etc, are found in the file configs/models/mt3.yaml.

Evaluating experiments

After an experiment has generated checkpoints, you can evaluate its average GOSPA score using the eval.py script. The evaluation hyperparameters can be found inside configs/eval/default.yaml. Example usage:

src/eval.py -rp src/results/experiment_name -tp configs/tasks/task1.yaml