Brightsun10/instance-segmentation-demo

🖼️ Instance Segmentation with Mask2Former

This demo performs advanced instance segmentation using Mask2Former from Facebook AI. It identifies and highlights individual objects in an image with:

• —Colored masks
• —Bounding boxes
• —Class labels and confidence scores

🚀 How It Works

• —Input an image via upload or example selection.
• —The app uses the facebook/mask2former-swin-large-coco-instance model to detect objects.
• —Only the following classes are visualized:
• —cat, dog, car, truck, bus, person
• —Results are drawn on the image and displayed along with a status message.

🧠 Model

• —Architecture: Mask2Former with Swin-Large backbone
• —Dataset: COCO Instance
• —Framework: Hugging Face Transformers + PyTorch

💻 Technologies Used

• —Python 🐍
• —Gradio for UI
• —Hugging Face Transformers
• —PIL & NumPy for image manipulation

📷 Example Images

Try out with example images like:

• —Cats vs. Dogs
• —Street scenes with vehicles and people

You can also upload your own images!

📌 Notes

• —Detection is limited to high-confidence predictions (score > 0.9)
• —This demo is optimized for CPU; inference may take up to 30 seconds.

🛠️ Developer Notes

This app uses the following Gradio configuration:

sdk: gradio
sdk_version: "4.24.0"
app_file: app.py