The soure code of the paper "A unified spatial-spectral-temporal network for hyperspectral object tracking".

Quick Start

1. Install the environment

Use the Anaconda

conda create -n csstrack python=3.8
conda activate csstrack
bash install.sh

2. Set project paths

Run the following command to set paths for this project

python tracking/create_default_local_file.py --workspace_dir . --data_dir ./data --save_dir ./output

After running this command, you can also modify paths by editing these two files

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

3. Hyperspectral Video Dataset

• The HOT2020 is from "https://www.hsitracking.com/".
• The IMEC25 dataset is from paper "Histograms of oriented mosaic gradients for snapshot spectral image description".
• The data should look like:

    (1). The format of training dataset:
        rootDir |-
                   videoName1
                       |- HSI
                           |- 0001.png
                           |- 0002.png
                           ...
                           |- XXXX.png
                           |- groundturth_rect.txt
                   videoName2
                       |- HSI
                           |- 0001.png
                           |- 0002.png
                           ...
                           |- XXXX.png
                           |- groundturth_rect.txt
                   ...
                   videoNameN
                       |- HSI
                           |- 0001.png
                           |- 0002.png
                           ...
                           |- XXXX.png
                           |- groundturth_rect.txt

  (2). The format of testing dataset:
      rootDir |-
                 test_HSI
                     |- videoName1
                         |- groundturth_rect.txt
                         |- HSI
                              |- 0001.png
                              |- 0002.png
                              |- ...
                              |- XXXX.png
                     |- videoName2
                         |- groundturth_rect.txt
                         |- HSI
                              |- 0001.png
                              |- 0002.png
                              |- ...
                              |- XXXX.png
                     ...
                     |- videoNameM
                         |- groundturth_rect.txt
                         |- HSI
                              |- 0001.png
                              |- 0002.png
                              |- ...
                              |- XXXX.png

4. Train & Test in HOT2020

(a) cd CSSTrack-HOT2020/

(b) Train: Download pretrained model and put in the folder "pretrained_models", which is available in
- https://pan.baidu.com/s/1vBFqFkpCHO9vqRR-Q0O1zw
- Access code: vr63

I. Change the path of training data in lib/train/admin/local.py (Line 25: self.hot2020_dir='/data/xx/HOT2020/train')
II. Run: python tracking/train.py --script csstrack --config CSSTrack-ep30-s256 --save_dir ./output --mode single --nproc_per_node 1

(c) Test: Download testing model of HOT2020 in
- https://pan.baidu.com/s/1a9Byn-R9zL89AVIx1loJiA - Access code: sumc

I. Change the path of training data in lib/train/admin/local.py (Line 20: settings.hot2020_path = '/data/xx/HOT2020/test')
II. Run: python tracking/test_epoch.py --checkpoint_path ../CSSTrack_ep0030_final.pth.tar

5. Train & Test in IMEC25

(a) cd CSSTrack-IMEC25/

(b) Train: Download pretrained model and put in the folder "pretrained_models", which is available in
- https://pan.baidu.com/s/1vBFqFkpCHO9vqRR-Q0O1zw
- Access code: vr63

I. Change the path of training data in lib/train/admin/local.py (Line 25: self.imec25_dir='/data/xxx/HOT/IMEC25/train')
II. Run: python tracking/train.py --script csstrack --config CSSTrack-ep30-s256 --save_dir ./output --mode single --nproc_per_node 1

(c) Test: Download testing model of IMEC25 in
- https://pan.baidu.com/s/1Ty_MKeiEwd2977A-6i4nfw - Access code: adpe

I. Change the path of training data in lib/train/admin/local.py (Line 20: settings.imec25_path = '/data/xxx/HOT/IMEC25/test')
II. Run: python tracking/test_epoch.py --checkpoint_path ../CSSTrack_ep0030_final.pth.tar

6. Cite

@article{LI2025111389,
title = {A unified spatial-spectral-temporal network for hyperspectral object tracking},
author = {Zhuanfeng Li and Jing Wang and Jue Zhang and Dong Zhao and Guanyiman Fu and Jiangtao Wang and Jianfeng Lu},
journal = {Pattern Recognition},
volume = {174},
pages = {113005},
year = {2026},
issn = {0031-3203},
doi = {https://doi.org/10.1016/j.patcog.2025.113005},
}

7. Concat

lizhuanfeng@hytc.edu.cn; If you have any questions, just contact me.

Acknowledgments

• Thanks for the AQATrack and PyTracking library, which helps us to quickly implement our ideas.