🧠 Who is that PokeML? - Image classifier

A deep learning model that classifies Pokémon images using transfers learning with Mobilenetv2 + Keras. Using its own custom-related model predicts more than 150 Pokémon from files or URLs.

📦 Project Overview

The project trains an image classifier that can identify 150+ Pokémon characters from images. It uses Mobilenetv2 as a pretrand base and adds a custom head for Pokémon classification.

Trained using tensorflow + keras, with data growth, fine-tuning, and Early stopping for best performance.

Test Dataset to train model to identify pokemon: https://www.kaggle.com/datasets/lantian773030/pokemonclassification

🚀 features

• ✅ Transfer Learning (Mobilentv2 + Custom Head)
• 🎨 data growth (rotation, zoom, flips)
• 🧠 fine tuning for high accuracy
• 🔁 Erustoping + Model Czechpointing
• Predic from 🌐 image url or local file
• 📁 saved model + class label
• 🧪 60%+ verification accuracy on 150 classes

📁Directory Structure

pokemon-classifier/
├── data/                    # Pokémon images (organized by class folders)
│   ├── pikachu/
│   ├── charmander/
│   └── ...
├── model/                   # Saved model and class index map
│   ├── pokemon_classifier.h5
│   └── class_indices.json
├── main.py                  # Training script (uses transfer learning with MobileNetV2)
├── search.py                # Prediction script (predicts from image URL or file)
├── README.md                # Project documentation (this file)

🧪 Model Info

Property	Value
Base Model	MobileNetV2
Classes	~150 Pokémon
Accuracy	~74% train, ~60% val
Epochs	30 (with EarlyStopping)
Save Format	.h5 (can convert to .keras)

🏁 How to Run

🔧 1. Train the Model

Make sure your Pokémon images are inside data/ and organized like:

data/
├── bulbasaur/
├── charmander/
├── pikachu/
....

Then run: python main.py

After training: The model is saved to model/pokemon_classifier.h5 Class mappings saved to model/class_indices.json

🔍 2. Predict Pokémon from Image URL

python search.py

🙌 Credits

🧩 Pretrained model: MobileNetV2 🎨 Dataset: PokeAPI sprites 🛠 Frameworks: TensorFlow, Keras, Pillow, NumPy