[INFFUS 25]FS-Diff: Semantic Guidance and Clarity-Aware Simultaneous Multimodal Image Fusion and Super-Resolution

AVMS Dataset

📊 Preview

📋 Details

📷 Sensor

DJI M30T, Its infrared camera operates within a spectral range of 8-14µm. The optical centers of the infrared and visible cameras are 3 cm apart.

📍 Main Scene

• The aerial photography location is in Foshan, Guangdong, China.

📁 Dataset Information

• Size: 600 image pairs
• Format:
  • IR: 24-bit grayscale
  • Visible: 24-bit RGB
• Resolution: 512×512 pixels
• Alignment: All pairs are precisely registered
• Annotations: 8,891 labeled instances across 7 categories

  Note: We welcome feedback on annotation quality

Statistics

Category	Objects	Images	Coverage
Car	8,046	419	86.75%
Person	526	118	24.43%
Bus	113	79	16.36%
Truck	157	86	17.81%
E-Bike	4	2	0.41%
Tent	138	54	11.18%
Boat	7	7	1.45%
Total	8,991	483	-

📥 Download

🚀 Usage

Environment Setup

pip install -r requirement.txt

infer

# run the script
python infer.py -c [config file]

train

# run the script
python sr.py -p train -c [config file] -enable_wandb -log_eval

🎓 Citation

Please cite our paper:

@article{FS-Diff,
  title={FS-Diff: Semantic guidance and clarity-aware simultaneous multimodal image fusion and super-resolution},
  author={Jie, Yuchan and Xu, Yushen and Li, Xiaosong and Zhou, Fuqiang and Lv, Jianming and Li, Huafeng},
  journal={Information Fusion},
  pages={103146},
  year={2025},
  publisher={Elsevier}
}