FMA-Net: Flow-driven Motion-Aware Network for Multi-Object Tracking in Satellite Videos

This repository provides the official code for the paper:

FMA-Net: Flow-driven Motion-Aware Network for Multi-Object Tracking in Satellite Videos
IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2025

---

🎯 Key idea

FMA-Net introduces a motion-sensitive tracking paradigm tailored for satellite video.
The framework consists of:

Component	Role
FME	Estimates pixel-level motion cues to perceive fine-grained dynamic patterns
MAF	Aggregates multi-frame features guided by motion sensitivity
MAR	Refines association and preserves identity continuity

Experiments on SatVideoDT and SatMTB confirm that FMA-Net establishes state-of-the-art performance across tracking metrics.

---

🏛️ Architecture

---

📦 Installation

Environment

• Python 3.8
• PyTorch 1.11.0
• CUDA 11.3

Dependencies

pip install -r requirements.txt

MMCV + MMDetection

pip install -U openmim
pip install mmcv-full==1.7.0 -f https://download.openmmlab.com/mmcv/dist/cu116/torch1.12/index.html
pip install mmdet==2.25.1

COCOAPI (required）

# COCOAPI=/path/to/clone/cocoapi
git clone https://github.com/cocodataset/cocoapi.git $COCOAPI
cd $COCOAPI/PythonAPI
make
python setup.py install --user

DCNv2 (required）

cd lib/models
git clone https://github.com/lucasjinreal/DCNv2_latest.git
cd DCNv2_latest
python setup.py build develop

---

📂 Trained Models and Dataset Annotations

To facilitate reproduction of our results, we provide:

• Trained model weights for both SatVideoDT and SatMTB datasets.
• Dataset JSON annotations (COCO format) corresponding to benchmark splits.
• Raw tracking output results generated by FMA-Net.

All files can be downloaded from Google drive

🚀 Training

SatMTB

python train_satmtb.py 
    --model_name DLADCN --gpus 0,1 
    --lr 1.25e-4 --num_epochs 30 --batch_size 4 
    --seqLen 5 --num_classes 3 
    --datasetname SatMTB --data_dir ./data/SatMTB/ 
    --warmup_iters 19512

SatVideoDT

python train_satvideoDT.py 
    --model_name DLADCN --gpus 0,1 
    --lr 1.25e-4 --lr_step 14,19 
    --num_epochs 20 --batch_size 4 --seqLen 5 
    --datasetname ICPR --data_dir ./data/ICPR/

---

🔍 Inference

SatVideoDT

python test_satvideoDT.py 
    --model_name DLADCN --gpus 0 
    --load_model ./checkpoints/SatvideoDT.pth 
    --datasetname ICPR --data_dir ./data/ICPR/ 
    --seqLen 5 --conf_thres 0.2

SatMTB

python test_satmtb.py 
    --model_name DLADCN --gpus 4 
    --load_model ./checkpoints/Satmtb.pth 
    --datasetname SatMTB --data_dir ./data/SatMTB/ 
    --seqLen 5 --num_classes 3 --conf_thres 0.2

---

📈 Evaluation

pip install motmetrics

Then replace the default parser:

cp ./motmetrics_io/io.py <python_env>/site-packages/motmetrics/io.py

Run evaluation:

python eval_satvideoDT.py
python eval_satmtb.py

---

🙏 Acknowledgements

We sincerely appreciate the contributions of the following open-source projects:

• MP2Net
• CenterNet

---

📝 Citation

If our paper or code proves useful to your work, we would greatly appreciate your citation and a star for this repository.

@ARTICLE{11214536,
  author={Lu, Weiqing and Sun, Bin and Li, Shutao and Li, Xiliang},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={FMA-Net: Flow-Driven Motion-Aware Network for Multiobject Tracking in Satellite Videos}, 
  year={2025},
  volume={63},
  pages={1-15},
  doi={10.1109/TGRS.2025.3624205}}