Overview

This project focuses on identifying Indonesian batik motifs from images using deep learning. The core objective was to systematically compare multiple optimizers and hyperparameter tuning methods, then deploy the best-performing model as a usable identification system. The final model was trained in two phases (transfer learning + fine-tuning) and deployed as an interactive web application.

What I Built

• A batik image classification model using MobileNetV2 with transfer learning
• Comparison of 4 hyperparameter tuning methods (Grid Search, Random Search, Bayesian Optimization, PSO)
• Evaluation of multiple optimizers to find the most stable and accurate configuration
• A two-phase training strategy to improve generalization and performance
• An end-to-end batik motif identification system deployed for real-world usage

Tools & Technologies

• TensorFlow / Keras — model development and training
• MobileNetV2 — transfer learning backbone
• Python — data processing and modeling
• Streamlit — interactive web application for deployment

Live Demo:

SIBIMO - Try ur batik!