Skip to main content
Back to Projects
AI/MLApr 2026 – Present

CIFAR-10 Image Classification

An applied machine learning project that trains, evaluates, and compares three CIFAR-10 classifiers — a Custom CNN, MobileNetV2, and ResNet-18. Built with Python, PyTorch, and scikit-learn, it includes Grad-CAM for visual interpretability, INT8 quantisation for model efficiency, CLI inference tools, and a live interactive demo hosted on Hugging Face Spaces via Gradio. ResNet-18 achieved 87.48% test accuracy.

Live Demo View Source

Tech Stack

PythonPyTorchscikit-learnGradioHugging Face SpacesGrad-CAM

Key Highlights

  • ▸Built an end-to-end Python pipeline for training, evaluating, and comparing CIFAR-10 classifiers — ResNet-18 achieved 87.48% test accuracy
  • ▸Applied data augmentation, cosine annealing, and progressive unfreezing across all three model architectures
  • ▸Added Grad-CAM visual interpretability, INT8 quantisation, and CLI inference tools to the evaluation workflow
  • ▸Deployed a live interactive demo on Hugging Face Spaces using Gradio