MVANet Image Segmentation

A modern web application for image segmentation using the MVANet deep learning model. Features a beautiful web dashboard and FastAPI-based backend for batch processing of images with real-time progress monitoring.

Features

Web Dashboard

• Modern UI: Clean, responsive interface with real-time updates
• Job Submission: Easy folder path input for batch image processing
• Test-Time Augmentation: Toggle TTA for improved segmentation accuracy
• Real-time Console: Live processing logs with timestamps and color-coded messages
• Latest Image Viewer: Preview the most recently processed image with overlays
• Task Persistence: Processing continues across page refreshes - monitor from anywhere

Processing Capabilities

• Recursive Folder Processing: Automatically finds and processes all 'images' folders
• Dual Output: Generates both segmentation masks and overlays
• GPU Acceleration: Supports CUDA for fast processing
• Background Processing: Non-blocking job execution with task tracking
• Error Handling: Comprehensive logging and error reporting

API Endpoints

• Task submission and status tracking
• Real-time log streaming
• System status monitoring
• Image serving and retrieval

Installation

1. Clone the repository:

   git clone https://github.com/OpsiClear/MVANet.git
   cd MVANet

2. Install dependencies using uv (recommended) or pip:

   # Using uv (faster)
   uv sync
   
   # Or using pip
   pip install -r requirements.txt

3. Ensure the model files are in the models/ directory:

   • models/MVANet.pth - Main segmentation model
   • models/swin_base_patch4_window12_384_22kto1k.pth - Backbone model

Usage

Starting the Application

Run the FastAPI server directly:

python api_app.py

The application will start on http://localhost:8001 by default.

Web Interface

1. Open your browser and navigate to http://localhost:8001
2. Enter the full path to your input folder (must contain folders named 'images')
3. Toggle Test-Time Augmentation if desired (improves accuracy but slower)
4. Click "Submit Job" to start processing
5. Monitor progress in real-time via the console output
6. View the latest processed image using the "Latest Image" button

Input Folder Structure:

your-input-folder/
├── images/           # Must be named 'images'
│   ├── image1.jpg
│   ├── image2.png
│   └── ...

Output Structure:

your-input-folder_overlay/  # Segmentation overlays
your-input-folder_masks/    # Binary masks

API Endpoints

Submit Processing Job

curl -X POST "http://localhost:8001/api/process" \
    -H "Content-Type: application/json" \
    -d '{
        "input_folder": "/path/to/folder",
        "use_tta": true
    }'

Check Task Status

curl "http://localhost:8001/api/status/{request_id}"

Get System Status

curl "http://localhost:8001/api/system/status"

Get Task Logs

curl "http://localhost:8001/api/logs/{task_id}"

Get Latest Processed Image

curl "http://localhost:8001/api/latest-image"

Technology Stack

• Backend: FastAPI with async/await support
• Frontend: Bootstrap 5 with vanilla JavaScript
• Deep Learning: PyTorch with MVANet architecture
• Model Backbone: Swin Transformer
• Image Processing: PIL/Pillow, NumPy

License

MIT License