🔍 Logo Detection – Flask + Deep Learning

A modern deep learning–powered web application that identifies brand logos from images using a trained Xception CNN model.
Built with Flask, TensorFlow, and a clean, interactive UI.

✨ Deep Learning–Powered Brand Logo Recognition

Upload an image → Predicts the brand logo with confidence.
Trained on the Flickr Logos 27 Dataset, deployed via a clean Flask interface.

---

📋 Table of Contents

• 📋 Table of Contents
• ✨ Introduction
• ⚙️ Tech Stack
  • 🧠 Machine Learning
  • 🌐 Backend
  • 🎨 Frontend
• 🔋 Features
  • 🔍 Logo Detection
  • 🖼️ Modern UI
  • 🧠 Deep Learning
  • 🗂️ Flask Integration
• 📁 Dataset
• 🤖 Model Architecture
• 🚀 Project Workflow
• 🤸 Quick Start
  • 🔧 Prerequisites
  • 1️⃣ Clone Repository
  • 2️⃣ Create Virtual Environment
  • 3️⃣ Install Dependencies
  • 4️⃣ Run the App
• 🧱 Project Structure
• 🖥️ App Flow
• 🧠 Architecture Overview
  • 🧩 Backend (Flask)
  • 🎨 Frontend
  • 🤖 Deep Learning Model
• 🚀 Future Enhancements
• 🤝 Contribution
• 🔗 Contacts
• 📄 License
• 🙏 Acknowledgements
  • ⭐ Show Your Support

---

✨ Introduction

AI Logo Detection is an end-to-end deep learning project that identifies brand logos in images.
This project consists of:

• A deep learning model (Xception + custom CNN head) trained via transfer learning.
• A Flask web app where users upload an image.
• The model predicts:
  • Logo Name
  • Confidence Score
  • Displays the uploaded image.

Built for accuracy, performance, and clean UI.

---

⚙️ Tech Stack

🧠 Machine Learning

• TensorFlow 2.x
• Keras
• Xception (pretrained on ImageNet)
• OpenCV
• NumPy
• Pandas

🌐 Backend

• Flask
• Werkzeug

🎨 Frontend

• HTML5
• CSS3
• JavaScript (for image preview)

---

🔋 Features

🔍 Logo Detection

• Upload any image (JPG/PNG)
• Model returns:
  • Predicted brand
  • Confidence %
  • Raw probability vector

🖼️ Modern UI

• Drag-and-drop upload
• Live image preview
• Clean dark theme
• Result preview panel

🧠 Deep Learning

• 27 logo classes
• Transfer learning using Xception
• High accuracy model stored in logo.h5

🗂️ Flask Integration

• Handles uploads
• Saves files to static/uploads/
• Renders predictions dynamically

---

📁 Dataset

This project uses the Flickr Logos 27 Dataset, which contains:

• 27 brand classes
• Training + validation annotations
• Bounding box coordinates
• Distractor images

Dataset includes brands such as:

Adidas, Apple, BMW, CocaCola, Ferrari, Ford, Google, Intel, Nike,
Pepsi, Porsche, Puma, RedBull, Starbucks, Yahoo, Vodafone, McDonalds, etc.

---

🤖 Model Architecture

Built using Transfer Learning:

• Backbone: Xception (pretrained on ImageNet)
• Custom classification head:
  • AveragePooling
  • Flatten
  • Dense (128)
  • Dropout (0.5)
  • Dense (27 softmax)

Loss: categorical_crossentropy Optimizer: Adam

The model is exported as:

logo.h5

---

🚀 Project Workflow

1️⃣ Dataset extraction → bounding box cropping 2️⃣ Data augmentation using ImageDataGenerator 3️⃣ Transfer learning with Xception 4️⃣ Model training on 224×224 images 5️⃣ Evaluation & classification report 6️⃣ Model exported to project 7️⃣ Flask app built for real-time inference 8️⃣ User uploads an image 9️⃣ Prediction displayed in browser

---

🤸 Quick Start

🔧 Prerequisites

• Python 3.8+
• pip
• Virtual environment recommended

---

1️⃣ Clone Repository

git clone https://github.com/itssanthoshhere/Logo-Detection.git
cd Logo-Detection

2️⃣ Create Virtual Environment

python3 -m venv venv
source venv/bin/activate

3️⃣ Install Dependencies

pip install -r requirements.txt

(or manually)

pip install flask tensorflow pillow numpy

4️⃣ Run the App

python app.py

Navigate to:

👉 http://127.0.0.1:5000

Upload an image → get prediction 🎉

---

🧱 Project Structure

logo_flask_full/
├── app.py
├── classes.txt
├── logo.h5
├── logo_detection_flickr27.ipynb
├── logos/
│   ├── Ferrari.jpg
│   ├── McDonald's.jpg
│   └── ford.jpg
├── static/
│   ├── style.css
│   └── uploads/
├── templates/
│   └── index.html
└── venv/

---

🖥️ App Flow

1. User opens web page

2. Uploads image (drag & drop or file picker)

3. Flask receives file

4. Model processes image

5. Prediction returned

6. UI displays:

   • Logo name
   • Confidence
   • Image preview

---

🧠 Architecture Overview

🧩 Backend (Flask)

• Loads TensorFlow model at startup
• Handles /predict endpoint
• Saves uploaded images
• Returns prediction + confidence

🎨 Frontend

• Styled HTML/CSS
• Image preview before upload
• Displays prediction results

🤖 Deep Learning Model

• Transfer learning
• 27-class softmax classifier
• Preprocessed with normalization

---

🚀 Future Enhancements

• 🔲 Bounding box logo localization
• 🔲 Support for multiple logos in one image
• 🔲 Convert model to TensorFlow Lite (TFLite)
• 🔲 Deploy on Render / Railway / AWS
• 🔲 Add API-only mode (REST endpoints)
• 🔲 Add history of predictions
• 🔲 Add brand logo icons in UI

---

🤝 Contribution

Contributions are welcome!

1. Fork repo
2. Create a feature branch
3. Commit changes
4. Open a PR 🎉

---

🔗 Contacts

• GitHub: Itssanthoshhere
• LinkedIn: Santhosh VS

---

📄 License

For educational and research purposes only. Logos belong to their respective brands.

---

🙏 Acknowledgements

• Flickr Logos 27 Dataset
• TensorFlow / Keras
• Flask
• Kaggle
• OpenCV

---

⭐ Show Your Support

If you like this project, give it a star ⭐ on GitHub — it motivates me to build more awesome ML apps!

---