Segment Anything 2.1 / Florence 2 Automasker

This Python script allows for interactive image masking using either AI-based segmentation with Segment Anything 2.1 or manual painting/lasso selection in a UI.

Features

• AI Mode (Segment Anything 2.1)
  • Hovering over objects previews a mask using the tiny model.
  • Clicking applies a mask using the large model.
• AI Mode (Florence 2 + Segment Anything 2.1)
  • Provide a prompt along with input and output folders to automatically mask every image in the folder
• Manual Masking Mode
  • Lasso Selection (Default Mode)
    • Press L to toggle lasso mode.
    • Left click: Lasso a white mask.
    • Middle click: Lasso a grey mask.
    • Right click: Erase mask.
  • Paintbrush Mode
    • Left click: Paint a white mask.
    • Middle click: Paint a grey mask.
    • Right click: Erase mask.
• Zooming
  • Use the mouse wheel to zoom in and out.
• Saving
  • Press S to save mask
• Reset
  • Press R to reset masks
• Fullscreen
  • Press esc to fullscreen
• Navigation
  • Press left and right keys to navigate between files.
• Toggle AI Mode
  • Press tab to toggle AI mode (off by default)

Installation

Ensure you have Git and Python installed, then run the following commands:

git clone https://github.com/Tophness/Segment-Anything-Mask-Generator
cd Segment-Anything-Mask-Generator
python -m venv venv
call venv\Scripts\activate
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu126
pip install -r requirements.txt

Usage

Run the script with:

python main.py

You can also provide input and output folders to automatically launch in manual / SAM 2.1 mask editing mode using those folders, choose a file in the folders to start from and use cpu if you don't want to use gpu:

python main.py --input_folder "folderpath" --output_folder "folderpath" --start_file "start_file.png" --cpu

Screenshots

Selecting a hand in AI mode

Drawing a white mask in lasso mode

Removing part of a mask in lasso mode

Drawing a grey mask in paintbrush mode

License

This project is released under the MIT License.