MSLabel

MSLabel is a tool for bounding box labeling of multi-spectral (multi-layer, and possibly time-series) imagery. This tool is particularly useful when visual inspection of two bands, simultaneously, is required to create labels.

For example, in the figure below, we can visually inspect the right-hand side image (image from band 2 that temporally corresponds/associates with the target image from band 1) to annotate the target image from band 1 (target band).

---

Installation and Setup

Clone the Repository

Clone the project repository to your local machine:

git clone https://github.com/yourusername/MSLabel.git
cd MSLabel

Requirements

The following libraries are required to run the application:

• Python 3.7+
• PyQt5
• OpenCV
• NumPy

Install them using pip:

pip install PyQt5 opencv-python numpy

Set Up the Directory Structure

Ensure the directory is structured following expected format:

images/
├── band1/
│   ├── band_1_image_1.png
│   ├── band_1_image_2.png
│   ├── band_1_image_3.png
│   └── ...
├── band2/
│   ├── band_2_image_1.png
│   ├── band_2_image_2.png
│   ├── band_2_image_3.png
│   └── ...
└── 

Run the Application

Launch the tool by executing:

python main.py

How to Use

• Launch the Tool
• Run the application using the command above. The graphical user interface (GUI) will open.
• Navigate Between Images: Use the "Next" and "Back" buttons to cycle through images in the dataset.
• Draw Bounding Boxes: Use your mouse to select regions of interest (ROIs) on the target image. The auxiliary band image is displayed alongside the target band for better labeling accuracy.
• Save Annotations: Aave your bounding boxes by clicking the "Save" button.
• Annotations are stored in the annotation_files/ directory in XML format.

Output

• Bounding box annotations for each image are saved as XML files.
• Saved annotations are stored in the annotation_files/ directory.
• Each XML file corresponds to a specific image, containing its bounding box data.

---

Citation

If you use this tool, please consider citing the following works:

• MLMT-CNN - Object Detection and Segmentation in Multi-layer and Multi-spectral images
• Active Region Detection in Multi-spectral Solar Images
• MSMT-CNN for Solar Active Region Detection with Multi-Spectral Analysis