Jackson or Not Jackson?

Project Overview

This project was developed as part of the Master's program in Computer Science and Big Data, option Recherche Big Data, in the course Recherche en Big Data with M. Touati.

The goal is to build a Convolutional Neural Network (CNN) to analyze and classify images of Jackson Wang (Wang Jia Er), a Hong Kong rapper, singer, and performer, member of the K-pop group GOT7. Jackson Wang has a successful international solo career and is also active in fashion and entertainment. This project aims to help identify him correctly, as people sometimes mistake him for other idols.

---

Architecture

The CNN architecture used in this project is as follows:

Base Model (MyCNN)

• Input: (3, 128, 128) RGB image
• Conv Layer 1: 3 → 32 filters, 3x3, ReLU, MaxPool2d(2x2) → 64x64
• Conv Layer 2: 32 → 64 filters, 3x3, ReLU, MaxPool2d(2x2) → 32x32
• Conv Layer 3: 64 → 128 filters, 3x3, ReLU, MaxPool2d(2x2) → 16x16
• Classifier: Flatten → Linear(1281616 → 256) → ReLU → Dropout → Linear(256 → 2)

Optimized Model (MyCNNOptimized)

• Input: (3, 128, 128) RGB image
• Conv Layer 1: 3 → 32 filters, 3x3, ReLU, MaxPool2d(2x2) → 64x64
• Conv Layer 2: 32 → 64 filters, 3x3, ReLU, MaxPool2d(2x2) → 32x32
• Classifier: Flatten → Linear(643232 → 256) → ReLU → Dropout → Linear(256 → 2)

The optimized model is lighter and faster to train while keeping good accuracy.

---

Dataset

The dataset contains images of Jackson Wang and "other" images (labeled divers), split into:

dataset/
├─ train/
│ ├─ jackson/
│ └─ divers/
├─ test/
│ ├─ jackson/
│ └─ divers/

• Training set: 160 Jackson, 160 divers
• Test set: 40 Jackson, 40 divers

---

Installation

1. Clone this repository:

git clone <repo-url>
cd CNN-Jackson-or-not

pip install -r requirements.txt

---

Usage

Training the model Run the main training script:

python or python3 src/main.py

This will train the CNN model on the dataset and save the trained weights as model.pth

---

Streamlit Application

Launch the interactive app to classify images:

streamlit run app.py

• Upload an image of Jackson or someone else.
• Click Analyse to see the prediction and confidence.
• Provide feedback if the prediction is correct or not.

Author

Chloé Makoundou / Cholorsplash This project was created as part of a Master's program in Computer Science and Big Data, option Recherche Big Data.