This project involves developing an image classification system using the VGG16 model, a well-known architecture in the field of deep learning. The objective is to accurately classify images into four predefined categories, showcasing the model's ability to learn and predict from visual data.
Customized VGG16 Model: Leveraged the VGG16 model with modifications to the classifier to suit my specific classification needs. Data Augmentation: Implemented several image transformation techniques to enhance model generalization. Dynamic Learning Rate Adjustment: Utilized a learning rate scheduler to optimize the training process. Performance Evaluation: Employed accuracy metrics and a confusion matrix for comprehensive model evaluation.
Python PyTorch torchvision PIL (Python Imaging Library) NumPy Seaborn Matplotlib scikit-learn
- Environment Setup:
Ensure Python 3.x is installed. Install required libraries: pip install -r requirements.txt.
- Training the Model:
Run python train.py to start the training process. Model checkpoints and logs will be saved in trained.pt
- Running Predictions:
Use python test.py to run predictions on new images.
Train Accuracy= 100%, Validation Accuracy=90.77%