ChemAI-Lab/vIR-OLO-10FG

vIR-OLO

Model Description

vIR-OLO (vision model for Infrared spectroscopy using YOLO) is a specialized YOLO-based model designed for automated peak detection and annotation in infrared (IR) spectroscopy analysis. This model was developed to assist researchers and technicians in identifying spectroscopic features, significantly accelerating the analysis workflow.

Key Features

• Optimized for IR Spectra: Trained on infrared spectroscopy data
• Multiple Class Variants: Available in 10-class (10FG) and 12-class (12FG) configurations
• High Accuracy: Delivers reliable peak detection for common IR functional groups
• Integration Ready: Seamless integration with the vIR-OLO annotation tool
• YOLO Format: Compatible with YOLOv11l and Ultralytics ecosystem

---

Model Variants

spectrai_ultralytics_IR_10FG.pt

• Classes: 10 functional groups
• Use Case: General IR spectroscopy analysis
• Performance: Optimized for common chemical compounds

spectrai_ultralytics_IR_12FG.pt

• Classes: 12 functional groups
• Use Case: Extended IR spectroscopy analysis
• Performance: Enhanced coverage for specialized analysis

---

Quick Start

Using with vIR-OLO (Recommended)

The recommended way to use these models is through the vIR-OLO application, a comprehensive tool for IR spectroscopy annotation and analysis.

For complete setup and usage instructions, please visit the vIR-OLO GitHub Repository, which includes:

• ✅ Installation guide (Python environment setup)
• ✅ Project creation and workflow
• ✅ Model downloading and loading
• ✅ Annotation and prediction features
• ✅ Troubleshooting and FAQs

Direct Model Usage

If you want to use the models directly with Python:

pip install ultralytics huggingface_hub

from ultralytics import YOLO

# Load model from Hugging Face
model = YOLO('UrielGC/spectrai-IR-YOLO-10FG')

# Run inference on an IR spectrum image
results = model.predict(source='path/to/ir_spectrum.png')

# Visualize results
results[0].show()

---

Training Data

• Domain: Infrared Spectroscopy
• Format: YOLO (normalized bounding box coordinates)
• Resolution: Optimized for typical IR spectra images
• Annotation Format: .txt files with class ID and normalized coordinates

---

Model Architecture

• Base Model: YOLOv11l
• Framework: Ultralytics
• Input Size: Configurable (typically 640x640)
• Output: Bounding boxes with class predictions and confidence scores

---

Dataset Format

The model expects input images in the following format:

dataset/
├── images/
│   ├── train/
│   │   └── *.png, *.jpg
│   └── val/
│       └── *.png, *.jpg
└── labels/
    ├── train/
    │   └── *.txt
    └── val/
        └── *.txt

Label Format

<class_id> <x_center> <y_center> <width> <height>

Where coordinates are normalized to [0, 1] relative to image dimensions.

---

Integration with vIR-OLO

These models are primary components of the vIR-OLO annotation and inference platform:

• Automatic Download: Models can be downloaded directly from the application
• Hybrid Workflow: Combine AI predictions with manual corrections
• Label Management: Automatic label merging for multiple models
• YOLO Export: Native support for YOLO format annotations

For more information about vIR-OLO, visit: GitHub Repository

---

Usage License

These models are available for research and educational purposes. Commercial use inquiries should be directed to the model authors.

---

Contact & Support

For issues, questions, or feature requests:

• GitHub Issues: vIR-OLO Issues

---

Made with ❤️ for the spectroscopy research community