🌱 AgriSegment

Multi-modal Plant Segmentation Suite for Agricultural Research

A web-based platform that makes AI-powered plant segmentation accessible to agronomists and researchers without programming expertise.

📦 Suite Components

AgriSegment provides four specialized tools for different segmentation workflows:

🎨 `hybrid/` - Port 8000

Hybrid SegFormer + SAM Workflow

Best for: Research projects requiring highest accuracy
Method: Automatic detection with interactive refinement
Key feature: Feedback loop where corrections improve AI performance
Output: High-quality masks with editable point annotations

✏️ `interactive/` - Port 8001

Interactive SAM Segmentation

Best for: Quick single-image tasks
Method: Point-based interactive segmentation
Key feature: Real-time mask generation from user clicks
Output: Binary masks, overlays, transparent PNGs

⚡ `semantic/` - Port 8002

Automated SegFormer Processing

Best for: Large-scale batch processing
Method: Fully automatic plant detection
Key feature: Multi-image upload with ZIP download
Output: 2x3 visualization grid with statistics

🔬 `panoptic/` - Port 8003

Advanced Mask2Former Analysis

Best for: Detailed multi-modal segmentation analysis
Method: Semantic, instance, and panoptic segmentation modes
Key feature: Comprehensive statistics and confidence thresholds
Output: Labeled segments with coverage percentages

🎯 Quick Selection Guide

Your Need	Recommended Tool	Why
Highest accuracy for research	hybrid/	Manual refinement + feedback loop
Quick single-image segmentation	interactive/	Fast, interactive, no setup
Process 100+ images	semantic/	Automated bulk processing
Instance-level detection	panoptic/	Multiple segmentation modes
Learning/experimenting	interactive/ → hybrid/	Start simple, then advanced
Production pipeline	semantic/ → hybrid/	Auto-process, refine when needed

🚀 Quick Start

Method 1: Run Individual Tool

# Navigate to desired tool
cd hybrid/        # Port 8000
cd interactive/   # Port 8001
cd semantic/      # Port 8002
cd panoptic/      # Port 8003

# Install and run
bash installer.sh
python server.py

Access: http://localhost:PORT

hybrid/: 8000
interactive/: 8001
semantic/: 8002
panoptic/: 8003

Method 2: Docker (All Tools)

# Start all four tools simultaneously
docker-compose up

# Or run in detached mode
docker-compose up -d

Access all tools:

💡 The Hybrid Workflow Innovation (`hybrid/`)

AgriSegment hybrid/ implements a continuous improvement cycle:

┌─────────────────────────────────────────────┐
│  1. SegFormer detects plant regions         │
│            ↓                                 │
│  2. User refines with SAM (click points)    │
│            ↓                                 │
│  3. Refined masks = better training data    │
│            ↓                                 │
│  4. Retrain SegFormer with improved data    │
│            ↓                                 │
│  5. Better automatic detection              │
│            ↓                                 │
│        (cycle repeats)                       │
└─────────────────────────────────────────────┘

This feedback loop enables continuous dataset improvement and model refinement.

🛠️ System Requirements

Minimum

OS: Linux, macOS, Windows 10+
Python: 3.8 or higher
RAM: 8GB
Storage: 10GB free space
Internet: Required for model downloads

📥 Installation

Prerequisites

# Verify Python version
python --version  # Should be 3.8+

# Verify pip
pip --version

Install Single Tool

# Clone repository
git clone https://github.com/mehran-tarif/AgriSegment.git
cd AgriSegment

# Install specific tool
cd [tool-folder]  # hybrid, interactive, panoptic, or semantic
bash installer.sh

Install All Tools

# Navigate to each folder and install
cd hybrid && bash installer.sh && cd ..
cd interactive && bash installer.sh && cd ..
cd panoptic && bash installer.sh && cd ..
cd semantic && bash installer.sh && cd ..

First Run

Models download automatically on first startup (5-10 minutes depending on internet speed).

📖 Usage

1. Upload Images

Drag & drop files into the upload area
Or click "Upload" button to browse
Supported formats: JPG, PNG, JPEG, BMP
Max size: 50MB per image (recommended: under 10MB)

2. Configure Settings

hybrid/: Adjust point generation parameters
panoptic/: Select segmentation mode (semantic/instance/panoptic)
interactive/: Choose model size (base/large/huge)
semantic/: Set number of concurrent processes

3. Process Images

Click "Generate" or "Segment" button
Wait for processing (1-30 seconds depending on image size and tool)
View results in real-time

4. Download Results

Individual files: Click download icon on each result
Batch download: Click "Download All as ZIP"
Available formats:
- Binary masks (PNG)
- Colored overlays (PNG)
- Transparent backgrounds (PNG)
- Statistics (JSON/CSV)

📊 Output Formats

All Tools Provide

✅ Original image
✅ Segmentation mask
✅ Overlay visualization
✅ Downloadable files

`hybrid/` & `panoptic/` Include

✅ Confidence scores
✅ Area measurements (pixels & percentage)
✅ Class labels
✅ Point coordinates (hybrid/)

File Structure

results/
├── session_[id]/
│   ├── original_image.jpg
│   ├── mask.png
│   ├── overlay.png
│   ├── transparent.png
│   └── statistics.json

🔧 Configuration

Change Port

Edit server.py in any tool folder:

if __name__ == "__main__":
    uvicorn.run(
        app, 
        host="0.0.0.0", 
        port=8000  # Change this number
    )

Adjust Memory Usage

For low-memory systems, reduce batch size:

# In server.py
BATCH_SIZE = 1  # Default: 4

Enable GPU

GPU automatically detected if CUDA available. To force CPU mode:

# In server.py
device = "cpu"  # Default: "cuda" if available else "cpu"

🐛 Troubleshooting

Port Already in Use

# Find and kill process using port
lsof -ti:8000 | xargs kill -9

# Or change port in server.py

Out of Memory Error

Solutions:

Reduce image size (resize to 1024x1024 or smaller)
Close other applications
Set BATCH_SIZE=1 in configuration
Use CPU mode instead of GPU

Models Not Downloading

Causes:

Internet connection interrupted
Firewall blocking Hugging Face
Insufficient disk space

Solutions:

# Manually download models
python -c "from transformers import AutoModel; AutoModel.from_pretrained('nvidia/segformer-b4-finetuned-ade-512-512')"

# Check available disk space
df -h

Import Errors

# Reinstall dependencies
pip install -r requirements.txt --force-reinstall

# Or use installer
bash installer.sh

📚 Documentation

Detailed guides for each tool:

Additional resources:

🤝 Contributing

We welcome contributions! Here's how you can help:

Report Bugs

Open an issue with:

Tool name and version
Steps to reproduce
Expected vs actual behavior
Screenshots if applicable

Suggest Features

Open an issue with:

Clear feature description
Use case explanation
Potential implementation approach

Submit Pull Requests

Fork the repository
Create feature branch (git checkout -b feature/AmazingFeature)
Commit changes (git commit -m 'Add AmazingFeature')
Push to branch (git push origin feature/AmazingFeature)
Open Pull Request

📄 Citation

Tarif, M., & Quaglia, D. (2025). AgriSegment: A Deep Learning–Based Multi-Modal Plant Segmentation Suite for Agricultural Research. University of Verona, Department of Computer Science. https://doi.org/10.5281/zenodo.17237438

@software{tarif_agrisegment,
  author       = {Tarif, Mehran and
                  Quaglia, Davide},
  title        = {AgriSegment: A Deep Learning–Based Multi-Modal
                   Plant Segmentation Suite for Agricultural Research},
  year         = 2025,
  publisher    = {University of Verona, Department of Computer Science},
  doi          = {10.5281/zenodo.17237438},   % Concept DOI (always resolves to the latest version)
  url          = {https://doi.org/10.5281/zenodo.17237438}
}

👥 Authors

Tarif Mehran
Department of Computer Science
University of Verona, Italy
📧 mehran.tarifhokmabadi@univr.it

Davide Quaglia
Department of Computer Science
University of Verona, Italy

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

Funding:

This work was supported by:

Italian Space Agency (ASI) - Project: "An Open, Efficient, and Customizable Pipeline for the Automated Processing of Remote Sensed Data for Computational Agro-Ecology"
Regione del Veneto - PR FESR 2021-2027, Action 1.1.1 - Project: "AGRIFUTURE: Il Futuro della Sostenibilità per le Sfide Competitive delle Aziende Agroalimentari Venete" (Project ID: 24279_001587_04499230235)

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of this work.

Institutional Support:
University of Verona, Department of Computer Science

Built with open-source technologies:

SegFormer by NVIDIA
Segment Anything (SAM) by Meta AI
Mask2Former by Meta AI
FastAPI web framework
PyTorch deep learning framework
Transformers by Hugging Face

📞 Support

Need help?

📖 Check Documentation
🐛 Report Issues
💬 Join Discussions
📧 Email: mehran.tarifhokmabadi@univr.it

⭐ Quick Tips

First time? Start with interactive/ - it's the simplest
Need accuracy? Use hybrid/ with manual refinement points
Processing many images? Use semantic/ for automated workflows
Research paper? Use panoptic/ for detailed statistics
Production deployment? Combine semantic/ + hybrid/ pipeline

🔗 Links

Project Homepage (coming soon)
GitHub Repository
ICROPM 2026 Paper
University of Verona

Made with 🌱 for agricultural research

⭐ Star on GitHub | 🐛 Report Bug | ✨ Request Feature

Name		Name	Last commit message	Last commit date
Latest commit History 7 Commits
docs		docs
hybrid		hybrid
interactive		interactive
panoptic		panoptic
semantic		semantic
.gitignore		.gitignore
CITATION.cff		CITATION.cff
LICENSE		LICENSE
README.md		README.md

License

mehran-tarif/AgriSegment

Folders and files

Latest commit

History

Repository files navigation