Instance Segmentation and Detection using a Tkinter-based interface

This project is a sophisticated computer vision application that performs real-time instance segmentation and object detection using a user-friendly Tkinter interface. The project identifies and isolates individual objects within images, providing both bounding box visualization and segmentation masks.

Key Features

• Interactive GUI built with Tkinter for easy image processing
• Detects and highlights the top 3 most confident object instances
• Dual visualization modes:
  • Bounding box detection with class labels
  • Precise segmentation masks with custom color overlays
• Supports multiple image transformations:
  • Blur
  • Flip (horizontal/vertical)
  • Rotate
  • Crop (center/random)
  • Rescale

Technical Highlights

• Built on PyTorch's powerful deep learning framework
• Efficient image processing using NumPy and PIL
• Custom transformation pipeline for image preprocessing
• Confidence-based filtering for optimal detection results
• Clean architecture with modular design for easy extensibility

Use Cases

• Object detection in natural scenes
• Instance segmentation for image analysis
• Educational tool for computer vision concepts
• Rapid prototyping of image processing workflows

Dependencies

1. PyTorch pip3 install torch torchvision torchaudio
2. Numpy pip3 install numpy
3. Matplotlib pip3 install matplotlib
4. Pillow pip3 install pillow
5. mypackage-hs094 pip3 install ./my_package_hs094-0.0.1-py3-none-any.whl

Notes

1. This software determines only the top 3 entities in the given image based on a confidence score. Other detected entities are not covered by segmentation masks or bounding boxes.
2. Temporary image files (1 for bounding box and 1 for segmentation mask) are saved to disk in the working directory during the runtime of the script. It is destroyed automatically afterwards.

Screenshots