🌼 Flower Image Classifier 🌸
This project aims to build an image classifier that can identify different species of flowers. The trained model can be used in various applications, such as a mobile app that can recognize and name flowers when provided with images.
🚀 Getting Started
To get started with this project, you'll need to follow these steps:
- Load and preprocess the image dataset.
- Train the image classifier using a pre-trained model.
- Use the trained classifier to predict the content of images.
🔧 Prerequisites
- Python
- PyTorch
- torchvision
- Jupyter Notebook (for running the project)
- GPU (recommended for faster training)
📊 Data Description
The dataset used for this project is split into three parts: training, validation, and testing. Each dataset requires specific preprocessing steps. The means and standard deviations of the images should be normalized to [0.485, 0.456, 0.406] and [0.229, 0.224, 0.225], respectively.
🏗️ Building and Training the Classifier
To build and train the image classifier, follow these steps:
- 🔄 Load a pre-trained network, such as VGG.
- 📊 Define a new, untrained feed-forward network as a classifier with ReLU activations and dropout.
- ⚙️ Train the classifier layers using backpropagation with the pre-trained network as a feature extractor.
- 📈 Monitor loss and accuracy on the validation set to optimize hyperparameters.
🖼️ Class Prediction
You can use the trained model to make predictions for a given image. The predict function takes a path to an image and a model checkpoint and returns the top-k most probable classes along with their probabilities.
probs, classes = predict(image_path, model)
print(probs)
print(classes)💾 Model Checkpoint
The trained model checkpoint, named checkpoint.pth, can be found Here. You can use this checkpoint to load the trained model for making predictions.
📝 Authors
Ankit Malik
🙏 Acknowledgments
This project is part of certification Program. Special thanks to the Udacity team for providing the project guidelines.