cifar100-image-classification

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.

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).

Part 3 of Visual Scrambling — four ResNet-style architectures reconstructed in PyTorch, with residual blocks, custom BatchNorm, theory notebooks, and CIFAR-100 experiments.

Building a Convolutional Neural Network (CNN) from scratch using PyTorch for CIFAR-10 image classification while understanding every concept from data preprocessing to model evaluation.

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