VisiCut

Interactive Image Segmentation and Cutout with SAM-2

VisiCut is a Google Colab-friendly project that leverages the powerful Segment Anything Model 2 (SAM-2) for interactive image segmentation and transparent background cutouts. Simply upload an image, click points on the objects you want to segment, and VisiCut will generate a precise mask and a ready-to-use transparent PNG.

This repository holds the core Python utility functions, designed for clean integration into a Google Colab notebook workflow.

Features

• Point-Based Interaction: Define segmentation regions by simply clicking points on your target object.
• SAM-2 Powered: Utilizes the state-of-the-art Segment Anything Model 2 for highly accurate masking.
• Transparent Cutouts: Generates clean PNG images with the segmented object and a transparent background.
• Google Colab Ready: Optimized for easy setup and execution within Google Colab environments.
• Modular Code: Core logic is separated into Python modules for reusability and clarity.

How to Use in Google Colab

To use VisiCut, you'll run a Google Colab notebook that clones this repository and executes the interactive segmentation.

1. Open the Colab Notebook:

   • [Link to your VisiCut Colab Notebook here] (You will create this notebook later, after setting up your GitHub repo and putting the .py files in it. Once created, paste its public share link here.)

2. Run Cell 1 (Setup and Clone Repository):

   • This cell will install necessary libraries (SAM-2, ipycanvas, etc.) and clone this VisiCut GitHub repository into your Colab environment.
   • It will also add the repository path to Python's sys.path so you can import functions from my_image_utils.py.

3. Run Cell 2 (Upload Image):

   • Execute this cell to upload the image you wish to segment.

4. Run Cell 3 (Interactive Pointing):

   • An interactive canvas will appear. Click directly on the object you want to segment. You will see a small lime green dot appear for each click.
   • The canvas is clickable even if the image doesn't immediately load (due to widget timing). Ensure you see the image before clicking for accurate point placement. If the image doesn't appear, try running the cell again or re-running Cell 1 and Cell 3.

5. Run Cell 4 (Generate Cutout):

   • This cell will send your clicked points to SAM-2, generate the mask, visualize the result, and finally produce the transparent PNG cutout.
   • The cutout image will be displayed, and a .png file will be saved to your Colab environment (visible in the file browser on the left sidebar).

---

Project Structure

• README.md: This file.
• my_image_utils.py: Contains core Python functions for image processing, SAM-2 interaction helpers, and result visualization.
  • process_image_for_sam: Prepares images for SAM-2.
  • display_mask_on_image: Visualizes the segmentation mask.
  • create_rgba_cutout: Generates the final transparent cutout.
  • (Add any other .py files and their main functions here)

Installation (for local development, not required for Colab)

If you wish to run parts of this project locally (outside of Colab), you would typically: