Minimalistic Image Labelling Framework (MILF)

A tkinter-based, minimalistic utility for pixel-wise image annotation with support for SAM-based prompting.

Prerequisites

The file requirements.txt contains the necessary Python packages for this project. To install, run:

pip install -r requirements.txt

Next, download SAM vit_b model checkpoint from their GitHub repository, and place it in the assets directory.

assets/sam_vit_b_01ec64.pth

To start the application, run the main script from the project root:

python3 main.py

Alternatively, self-contained executables built via PyInstaller are available on the Releases page for Windows and macOS. These builds can be directly executed without a local Python environment or external dependencies.

Usage

1. Load the dataset

   Select the directory containing image files (.png, .jpg, .jpeg, .tiff, .npy), followed by a .csv file containing the color map. Ensure that the color map is properly formatted: Label,HexCode. For example,

   Target,ff0000
   Background,ffffff

2. Optionally load masks

   Select the directory containing existing masks for updating.

3. Select mode

   • Polygon: Draw polygons with left-click, and fill regions with right-click, or fill the entire image with return.
   • Brush: Paint freehand masks using adjustable brush size.
   • Prompt: Place positive (left-click) and negative (right-click) points to prompt SAM for automatic prompt generation.

4. Additional Features

   • Use Left / Right arrow keys to navigate between images in the directory .
   • Use Up / Down arrow keys to zoom in and out.
   • Ctrl+Click on a pixel to copy its class ID; useful for replacing or updating annotated segments.
   • Optionally enter textual comments/notes for each image.
   • A .log file is generated to track annotation progress.

5. Save annotations

   Masks are saved as .png files in a subdirectory called segmented/ within the image directory. Textual comments, if any, are stored in .txt files corresponding to each image.

Citation

If you find this repository useful, please consider giving us a star ⭐

@misc{rajani2025benthicat,
    title={BenthiCat: An opti-acoustic dataset for advancing benthic classification and habitat mapping},
    author={Hayat Rajani and Valerio Franchi and Borja Martinez-Clavel Valles and Raimon Ramos and Rafael Garcia and Nuno Gracias},
    year={2025},
    eprint={2510.04876},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}