Flower Classification using MobileNetV2

Overview

This project is a deep learning application that classifies flower images into five categories using transfer learning with MobileNetV2.

The model can classify:

• Daisy
• Dandelion
• Rose
• Sunflower
• Tulip

Features

• Transfer Learning with MobileNetV2
• TensorFlow/Keras
• Gradio Web Interface
• Hugging Face Deployment
• Real-time Flower Prediction

Model Performance

• Validation Accuracy: 89.54%

Technologies Used

• Python
• TensorFlow
• MobileNetV2
• NumPy
• Gradio
• Hugging Face Spaces

Files

• Flower_Classification.ipynb — Model training notebook
• flower_classifier_fixed.keras — Trained model
• app.py — Gradio application
• README.md — Project documentation

Challenges Faced

While deploying the model on Hugging Face, the model incorrectly predicted "dandelion" for almost every image.

Root Cause

The images were being normalized twice:

• First in app.py
• Again inside the model through a Rescaling layer

Solution

Removed the extra normalization step from app.py, which fixed the predictions and restored expected model performance.

Future Improvements

• Support more flower species
• Add confidence visualization
• Improve model accuracy
• Deploy a mobile-friendly version

Application Preview

Home Page

Prediction Example

Live Demo

Try the application here: https://huggingface.co/spaces/Babu06/flower_classifier