Reproducible CIFAR-100 image classification. CNN baseline and EfficientNetV2B0 transfer learning with proper resize→preprocess, staged fine-tuning, and saved metrics/artifacts (78.2% top-1). Keras/TensorFlow CIFAR-100 pipeline: augmentation, 224×224 resize, EfficientNetV2B0 backbone, staged fine-tuning, and evaluation (top-1 0.782, top-5 0.959).

Project assessment for IKT466 - Introduction to Machine Learning , Autumn 2025

This project builds a high accuracy image classification system for the CIFAR100 dataset using the WideResNet28x10 architecture. The goal is to achieve strong top 1 accuracy through advanced augmentation, stable training strategy, and detailed model analysis.