The Segmentation Tool is a graphical user interface (GUI) tool developed using PyQt5 and based on the Facebook Segment Anything Model. It uses token points to semi-automate the process of image segmentation.
Screenshot of the segmentation tool in action
Input image of bare landscape
Segmentation mask for the bare landscape
Input image of roads
Segmentation mask for the roads
Input image of urban area
Segmentation mask for the urban area
- Open an image using the tool. Note that it may take some time to load.
- Add and delete classes as needed.
- Select the class that you want to segment and click "Start".
- In the opened window:
- Right-click to add + points
- Left-click to add - points
- Middle-click to delete the nearest point
- Once segmentation is complete, close the figure.
- Click on the class name in the list again to view the results.
- The results can be found in the "masks" folder.
The Segmentation Tool requires the following Python libraries to be installed:
- matplotlib
- pytorch
- torchvision
- PyQt5
- segment-anything
The Facebook Segment Anything Model and its pre-trained checkpoint can be downloaded from the following repository: https://github.com/facebookresearch/segment-anything/
Make sure to use the "sam_vit_b_01ec64.pth" checkpoint for this tool.
- Facebook Segment Anything Model repository: https://github.com/facebookresearch/segment-anything/
- Pre-trained checkpoint: models/sam_vit_b_01ec64.pth