Empowering Scene Classification with Advanced Transfer Learning
TransferVision is a robust deep learning project designed to classify images across six distinct scene categories with remarkable precision. Leveraging pre-trained models and cutting-edge transfer learning techniques, this project demonstrates the potential of AI when applied to relatively small datasets, achieving state-of-the-art results through fine-tuning and optimization.
- Advanced Transfer Learning: Utilizes pre-trained models (ResNet50, ResNet100, EfficientNetB0, VGG16) to extract meaningful image features.
- Comprehensive Image Augmentation: Empirical regularization techniques including rotation, zoom, flip, contrast, and translation to enhance generalization.
- High-Performance Metrics: Achieved 95%+ accuracy with Precision: 96%, Recall: 94%, AUC: 98%, and F1 Score: 95%.
- Optimized Training Process: Implemented techniques such as early stopping, batch normalization, dropout, and ADAM optimizer for robust performance.
- Programming Language: Python
- Deep Learning Framework: Keras, TensorFlow
- Pre-trained Models: ResNet50, ResNet100, EfficientNetB0, VGG16
- Tools: OpenCV (for image processing and augmentation)
-
Clone the Repository:
git clone https://github.com/himanshumahajan138/TransferVision.git cd transfervision -
Set Up the Environment:
- Create a virtual environment and activate it:
python3 -m venv env source env/bin/activate - Install the required dependencies:
pip install -r requirements.txt
- Create a virtual environment and activate it:
-
Prepare the Dataset:
- Place the training and testing images in their respective folders (organized by class).
- Ensure images are preprocessed (resized or zero-padded).
-
Run the Training Script:
python train.py
- Automatically performs data augmentation and trains models with early stopping.
-
Evaluate the Model:
python evaluate.py
- Reports metrics: Precision, Recall, AUC, and F1 Score.
-
Make Predictions:
python predict.py --image path/to/image.jpg
- Training and Validation Loss: Consistently reduced over 50-100 epochs.
- Metrics: Precision, Recall, AUC, and F1 Score metrics highlight the reliability and accuracy of the models.
- Transfer Learning Efficiency: Demonstrated how pre-trained models excel with small datasets.
- Data Augmentation Impact: Showcased the value of image augmentation in enhancing generalization.
We welcome contributions! If you have ideas or improvements, please open an issue or submit a pull request.
This project is licensed under the MIT License - see the LICENSE file for details.
For questions or collaborations, feel free to reach out:
- Email: himanshumahajan138@gmail.com
- LinkedIn: Himanshu Mahajan
"Fine-Tuning Excellence with TransferVision" 🚀