Computer Vision with CNNs

An introduction to Convolutional Neural Networks (CNNs) for image processing and computer vision tasks using PyTorch.

Learning Path

Main Track

Work through these notebooks in order:

#	Notebook	Description
1	Convolutional Networks Theory	Core concepts: convolution operations, filters (kernels), padding, stride, and pooling layers (nn.Conv2d, Max Pooling vs Avg Pooling)
2	MNIST with CNNs	Application of CNNs to the MNIST dataset, including data transformations, image normalization, and PyTorch dataloaders
3	CIFAR-10 Classification	Advanced image classification on color images: handling 3-channel data, deeper architectures, baseline models, optimization techniques, and evaluation metrics

Production Examples

The model created in the CIFAR-10 Classification notebook has been deployed as production-ready applications on Hugging Face Spaces:

• CIFAR-10 FastAPI & Docker: A REST API built with FastAPI and containerized using Docker.

  • GitHub Repository: avidaldo/cifar-10-fastapi
  • Live Deployment: Hugging Face Space

• CIFAR-10 Gradio: An interactive web interface directly deployed using Gradio.

  • Live Deployment: Hugging Face Space

Prerequisites

• Completion of the Deep Learning Fundamentals module
• Solid understanding of PyTorch basic constructs (tensors, nn.Module, dataloaders, and the standard training loop)
• Basic knowledge of multi-dimensional arrays and image representation