/HMFT

Hierarchical Multi-modal Fusion Tracker for RGB-T tracking (CVPR2022)

Primary LanguagePython

Code for Hierarchical Multi-modal Fusion Tracker in CVPR2022 paper Visible-Thermal UAV Tracking: A Large-Scale Benchmark and New Baseline, which is a strong baseline for RGB-T tracking.

Framework

alt text Three complementary modules are introduced for multi-modal fusion

  • Complementary Image Fusion(CIF): CIF aims to use a shared backbone to extract complementary information and Kullback–Leibler divergence loss is introduced to unify the feature distribution.
  • Discriminative Feature Fusion(DFF): DFF aims to build individual representations for both modalities and learns a channel-wise modality weight to fuse them.
  • Adaptive Decision Fusion(ADF): ADF is to adaptively provide the final response by considering the results of two branches and the modality confidence.

Model Zoo

The pretrained model are available at [GoogleDrive] and [BaiduDisk]

Get Started

Set up Anaconda environment

conda create -n HMFT python=3.6
conda activate HMFT
cd $Path_to_HMFT$
bash install.sh

Run demo sequence

cd $Path_to_HMFT$/mfDiMP/pytracking
python demo.py

Test on current benchmarks

cd $Path_to_HMFT$/mfDiMP/pytracking
python run_VTUAV.py
python run_GTOT.py
python run_RGBT210.py
python run_RGBT234.py

Training

Results

alt text The results can be found at [GoogleDrive] and [BaiduDisk]

Citation

If you find our work useful, please cite

@InProceedings{Zhang_CVPR22_VTUAV, author = {Zhang Pengyu and Jie Zhao and Dong Wang and Huchuan Lu and Xiang Ruan}, title = {Visible-Thermal UAV Tracking: A Large-Scale Benchmark and New Baseline}, booktitle = {Proceedings of the IEEE conference on computer vision and pattern recognition}, year = {2022} }