HyperTrack：A Unified Framework for Hyperspectral Object Tracking

Official implementation of HyperTrack, including models and training&testing codes.

Introduction

A new unified hyperspectral tracking framework (e.g. VIS, NIR, and RedNIR Tracking).

• HyperTrack has high performance on hyperspectal tracking tasks.

• A single model cross three kinds of hyperspectral images.

Results

On HOT2023 tracking benchmarks

On HOT2022 tracking benchmark

Usage

Installation

Create and activate a conda environment:

conda create -n hypertrack python=3.7
conda activate hypertrack

Install the required packages:

bash install_hypertrack.sh

Data Preparation

Put the training datasets in ./data/. It should look like:

$<PATH_of_HyperTrack>
-- data
    -- Test
        |-- VIS
        |-- NIR
        |-- RedNIR
        ...
    -- Train
        |-- VIS
        |-- NIR
        |-- RedNIR
        ...

Path Setting

Run the following command to set paths:

cd <PATH_of_HyperTrack>
python tracking/create_default_local_file.py --workspace_dir . --data_dir ./data --save_dir ./output

You can also modify paths by these two files:

./lib/train/admin/local.py  # paths for training
./lib/test/evaluation/local.py  # paths for testing

Training

Download the pretrained foundation model (Mixformer) and put it under ./pretrained_networks/.

python run_training.py

You can train models with various modalities and variants by modifying run_train.py.

Testing

HOT2022/HOT2023

python test.py

You can test models with various modalities and variants by modifying test.py.

Acknowledgment

• This repo is based on Mixformer which is an excellent work.
• We thank for the PyTracking library, which helps us to quickly implement our ideas.