This project demonstrates a deep learning pipeline to classify images of flowers into 17 different categories using a pre-trained EfficientNet-B0 model. It utilizes PyTorch and TIMM (PyTorch Image Models) for training, validation, and testing, implementing mixed precision training to optimize performance. The dataset used is the 17 Flower Classes dataset.
The goal of this project is to classify flowers into 17 different classes using deep learning. We leverage transfer learning with the EfficientNet-B0 model pre-trained on ImageNet and fine-tune it on our flower dataset. The model training and evaluation are conducted using mixed precision training to reduce memory usage and improve training speed.
- Transfer Learning: Utilizes the EfficientNet-B0 model pre-trained on ImageNet.
- Mixed Precision Training: Reduces memory consumption and accelerates training.
- Dataset Augmentation: Resizing and normalization of images to improve model generalization.
- Gradient Accumulation: Handles larger batch sizes without exceeding GPU memory.
- Training, Validation, and Test Phases: Clearly separated phases for training, validation, and testing to monitor performance.
Ensure you have the following installed:
- Python 3.7+
- PyTorch
- TIMM (PyTorch Image Models)
- Other dependencies:
numpy,PIL,matplotlib,torchvision,tqdm,gc
The dataset contains images of 17 different flower classes. It is divided into training, validation, and test sets. You can download the dataset from Kaggle using the following link: 17 Flower Classes Dataset.
- Train Directory: Contains the images used for training the model.
- Validation: A portion (20%) of the training dataset is set aside for validation to monitor model performance and avoid overfitting.
- Test Directory: Contains the images used for testing the model.
The model is trained using a PyTorch script that includes functions for:
- Loading and Transforming Data: Resize to 299x299, normalization using ImageNet mean and std.
- Training Loop: Implements gradient accumulation for better memory management.
- Validation: A separate validation set, comprising 20% of the training data, is used to evaluate the model after each epoch to monitor validation accuracy and loss.
- Saving Best Model: The model with the best validation accuracy is saved.
- The best model weights based on validation accuracy are saved during training.
- After training, the model was evaluated on the test set to determine its classification accuracy.
- Test Accuracy: The model achieved an accuracy of 96.47% on the test set.
This project is licensed under the MIT License. See the LICENSE file for more details.