README.md

metadata

license: mit
title: 🌸 Fulla 🌸
sdk: gradio
emoji: 👀
colorFrom: pink
colorTo: purple
pinned: true
short_description: A flower classifier built with PyTorch and Gradio.

🌸 Fulla 🌸

Fulla (فلة) is a deep learning project that classifies flowers from images using a ResNet-based neural network and transfer learning. Built with PyTorch and deployed with Gradio, this app blends the elegance of nature with the power of machine learning.

Upload a picture of a flower. Watch the model guess. Let it bloom!

🖼️ Live Demo

You are looking at the live demo! For more details, check out the GitHub repository.

✨ Features

• 🌼 102 Flower Classes: Trained on the comprehensive Flowers102 dataset.
• 🧠 Transfer Learning: Built on a pre-trained ResNet model for powerful feature extraction.
• 🧪 High Accuracy: Achieves strong performance on the test set.
• 🖼️ Interactive UI: A simple, beautiful interface built with Gradio.
• 🚀 Deployed: Live and accessible on Hugging Face Spaces.

📊 Results

The model was evaluated on a held-out test set, achieving the following performance:

• Final Test Accuracy: 79.38%
• Weighted F1-Score: 0.7886

Confusion Matrix

The confusion matrix below shows the model's high performance, with a strong diagonal indicating correct predictions across most classes.

🛠️ How to Run Locally

1. Clone the repository:

   git clone [https://github.com/salihelfatih/fulla](https://github.com/salihelfatih/fulla)
   cd Fulla
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Launch the app:

   python -m app.interface
   

🧠 Model Architecture

• Backbone: Pre-trained ResNet (ImageNet)
• Strategy: Freeze the feature extractor and train a new classifier head with 102 outputs.
• Loss: CrossEntropyLoss
• Optimizer: Adam
• Framework: PyTorch

🙌 Credits

Developed by Salih Elfatih as a capstone project on deep learning and computer vision. Flowers bloom. So should code!