🎯 CADTrack: Learning Contextual Aggregation with Deformable Alignment for Robust RGBT Tracking

This repository contains the official implementation of CADTrack, a novel framework for robust RGB-Thermal (RGBT) object tracking. CADTrack addresses key challenges of modality discrepancies and spatial misalignment via three innovative components: Mamba-based Feature Interaction (MFI) for efficient cross-modal interaction, Contextual Aggregation Module (CAM) for dynamic multi-layer feature fusion, and Deformable Alignment Module (DAM) for spatiotemporal alignment. Included are training/evaluation codes, models, and results.

🚀 New

• 🎉 Paper Accepted at AAAI 2026!
• 📦 Code & Models Released – Full implementation now publicly available.

✨ Key Features & Contributions

Figure 1: Overall framework of CADTrack.

Figure 2: Details of MFI.

Figure 3: The structure of CAM.

Figure 4: Deformable alignment of DAM.

⚙️ Installation

Create and activate a conda environment:

cd path/to/CADTrack
conda create -n CADTrack python=3.10
conda activate CADTrack

Download mamba_install and install the required packages:

bash install_cadtrack.sh

📂 Data Preparation

Download the following datasets and place them under ./data/:

• GTOT, RGBT210, RGBT234, LasHeR
• VTUAV

$<PATH_of_CADTrack>
-- data
    -- GTOT
        |-- BlackCar
        |-- Black5wan1
        ...
    -- RGBT210
        |-- afterrain
        |-- aftertree
        ...
    -- RGBT234
        |-- afterrain
        |-- aftertree
        ...
    -- LasHeR/train
        |-- 1boygo
        |-- 1handsth
        ...
    -- LasHeR/test
        |-- 1blackteacher
        |-- 1boycoming
        ...
    -- VTUAV/train
        |-- animal_002
        |-- bike_002
        ...
    -- VTUAV/test_ST
        |-- animal_001
        |-- bike_003
        ...
    -- VTUAV/test_LT
        |-- animal_003
        |-- animal_004
        ...

🔧 Setup & Configuration

Run the following command to set paths:

cd <PATH_of_CADTrack>
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 model and put it under ./pretrained/.

bash train.sh

You can train models with various variants by modifying train.sh.

📊 Testing

Testing on Benchmark Datasets

Modify the <DATASET_PATH> and <SAVE_PATH> in./RGBT_workspace/test_rgbt_mgpus.py, then run:

bash test.sh

Evaluation Tools

• GTOT/RGBT210/RGBT234/LasHeR: Use the Evaluation Toolkit
• VTUAV: Follow the VTUAV_Evaluation

📜 Poster

📝 Citation

If you find CADTrack is helpful for your research, please consider citing:

@inproceedings{li2026cadtrack,
  title={CADTrack: Learning Contextual Aggregation with Deformable Alignment for Robust RGBT Tracking},
  author={Li, Hao and Wang, Yuhao and Hu, Xiantao and Hao, Wenning and Zhang, Pingping and Wang, Dong and Lu, Huchuan},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={40},
  number={8},
  pages={6109--6117},
  year={2026}
}

🙏 Acknowledgments

This repo is based on STTrack and IDEA which are excellent works.

Star ⭐ this repo if you like our work!