👁️ Trinetra: Advanced Image Classification Pipeline

Trinetra, named after the "third eye" in Hindu mythology, is a comprehensive image classification pipeline that streamlines the process from data labelling to model training and evaluation. It provides clear vision into your image data using state-of-the-art CLIP models.

🌟 Features

• 🏷️ Automated image labelling for efficient dataset preparation
• 🧠 Fine-tuning of CLIP models for custom image classification tasks
• 🔢 Support for multiple precision formats (FP16, FP32)
• 📊 Integration with Weights & Biases for experiment tracking
• 💾 Automatic saving of best-performing models
• 🧮 Confusion matrix computation for performance analysis
• 📁 Flexible model saving in different formats (PTH, PT, ONNX)

🛠️ Installation

Automated Setup

1. Ensure you have Anaconda or Miniconda installed.
2. Clone the repository:

   git clone https://github.com/your-username/Trinetra.git
   cd Trinetra
   

3. Run the setup script generator:

   python generate_setup.py
   

4. Follow the instructions to run the generated setup script for your operating system.

Manual Setup

If you prefer to set up manually:

1. Create and activate a new Conda environment:

   conda create --name trinetra python=3.8
   conda activate trinetra
   

2. Install the required packages:

   pip install -r requirements.txt
   pip install git+https://github.com/openai/CLIP.git
   

🚀 Usage

Image Labelling

To label your images using CLIP:

python src/classifier_labelling.py --data_dir "path/to/unlabelled/images" --categories "Human" "Vehicle" --output_dir "path/to/labelled/output"

In the categories argument, you can specify the categories you want to label.

Model Training

To fine-tune a CLIP model on your labelled dataset:

python src/finetine_classifier.py --data_dir "path/to/labelled/dataset" --epochs 10 --batch_size 32 --learning_rate 0.001 --loss_function focal --precision FP16 --save_format pt

Command-line Arguments

• --data_dir: Path to the dataset directory
• --epochs: Number of training epochs
• --batch_size: Batch size for training
• --learning_rate: Learning rate for the optimizer
• --loss_function: Loss function to use (e.g., "focal", "cce")
• --precision: Precision format for training ("FP16" or "FP32")
• --save_format: Format to save the model ("pth", "pt", or "onnx")
• --clip_model: CLIP model to use (default: "ViT-B/32")

For a full list of options, run:

python src/finetine_classifier.py --help

📁 Project Structure

Trinetra/
│
├── src/
│   ├── classifier_labelling.py
│   ├── finetine_classifier.py
│   ├── loss_functions.py
│   ├── precision_formats.py
│   └── model_saver.py
│
├── requirements.txt
├── generate_setup.py
├── README.md
└── .gitignore

🔮 Upcoming Features

• Object Detection capabilities
• Semantic and instance segmentation support
• Advanced data augmentation techniques
• Multi-GPU and distributed training optimization
• Model interpretability tools

🤝 Contributing

Contributions to Trinetra are welcome! Please feel free to submit a Pull Request.

Development Setup

1. Fork the repository on GitHub.
2. Clone your forked repository locally.
3. Create a new branch for your feature or bug fix.
4. Make your changes and commit them with clear, descriptive messages.
5. Push your changes to your fork on GitHub.
6. Submit a Pull Request to the main Trinetra repository.

📊 Performance

Trinetra has been tested on various datasets and consistently achieves high accuracy in image classification tasks. Specific performance metrics will vary based on the dataset and chosen model architecture.

📄 License

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

📞 Contact

For any queries or suggestions, please open an issue on GitHub or contact the maintainer at [[email protected]].

---

Empower Your Vision with Trinetra! 👁️🚀

         _____
     _.-'     `'-._
   ,'               '.
  /                   \
 |    ___________     |
 |   |           |    |
 |   |    ( )    |    |
 |   |___________| 👁️  |
  \                   /
   '.               ,'
     '-.._______.-'

 Trinetra watches everything