Skip to content

Latest commit

 

History

History
84 lines (69 loc) · 2.92 KB

File metadata and controls

84 lines (69 loc) · 2.92 KB
G-MAD: An Open-Source Toolkit for Synthetic RGB-T Military Object Detection Data Generation

🛠️ An Open-Source Toolkit for Synthetic RGB-T Military Object Detection Data Generation

Python 3 Arma 3 Windows10-11 MIT


Preview

Welcome

Updates

  • (03/2026) Welcome!

Quick start

  • We assume that you have already finished setting up Arma 3 by our tutorial in advance.
  • Please install necessary python libraries in requirements.txt
    conda create -n g-mad python=3.8 -y
    conda activate g-mad
    pip install -r .\requirements.txt
    
  • For example, run the below command to generate 10 scenes of sunny day which covers all camera tilt cases between -60 and 60 degrees, from 9AM to 6PM, on the map named malden for training:
     python main.py  -weather 'sunny' -map_name 'malden' \
                     -arma_root_path 'C:/Users/{user_name}/Documents/Arma 3' \
                     -save_root_path 'C:/Users/{user_name}/Desktop' \
                     -start_hour 9 -end_hour 18 \
                     -n_times 10 -mode 'EO' \
                     -class_path 'classes/CLASSES.csv' \
                     -invalid_bbox_path 'classes/INVALID_BBOX.csv' \ 
                     -look_angle_min -60 -look_angle_max 60
    
  • To create only on-nadir view scenes, set both look_angle_min and look_angle_max to 0.
  • ⭐ Try to use our GUI tool for convenience!
     python main_GUI.py
    

Dataset structure

  • The directory structure of our dataset is as follows:
|—— 📁 {train or test}_{map_name}_{weather}_{start_hour}_{end_hour}_...
	|—— 📁 0000 (scene number)
		|—— 📁 20  (look angle)
			|—— 🖼️ EO_0000_-0.png  
			|—— 📄 ANNOTATION-EO_0000_20.csv (including bbox labels)
		|—— 📁 +20 
			|—— 🖼️ ...
	|—— 📁 0001
		|—— 📁 -20
		|—— 📁 ... 
	|—— 📁 0002
		|—— 📁 -20
		|—— 📁 ...
	...
	|—— 📄 meta_..._.csv (including in-game shooting time, weather, and error logs per each scene)
  • You may need to transform the above folder structure before training your own model.

Citation

  • Paper is coming soon!

Contribution

  • If you find any bugs for further improvements, please feel free to create issues on GitHub!
  • All contributions and suggestions are welcome. Of course, stars (🌟) are always welcome.