Introduction to Computer Vision

Course outline and learning expectations

This is a tutor-led course. The outline shows what your tutor will cover; teaching materials and examinations are provided directly to enrolled students.

What you will learn

• Explain how digital images are represented as pixels, channels and NumPy arrays
• Load, inspect, resize, normalise and augment image data
• Explain CNN concepts including convolution, filters, pooling and feature maps
• Build and train a CNN using TensorFlow/Keras
• Evaluate an image classifier using validation data, a confusion matrix and error analysis
• Apply transfer learning with a pretrained MobileNet model
• Describe the basic object-detection workflow and distinguish it from image classification
• Develop and document a prototype classifier for a locally relevant image problem

Modules

1. 1

   How computers see images

   Represent images as pixels, channels and arrays while extending prior NumPy skills

   Pixels and colour channelsImage dimensions and shapesNumPy image arraysLoading, displaying and saving images
2. 2

   Image preprocessing basics

   Prepare image data for model training and inference

   Resizing and croppingNormalisation and scalingColour-space conversionData augmentationDataset organisation and labels
3. 3

   Convolutional Neural Networks explained intuitively

   Understand how CNNs learn useful visual features

   Convolution and filtersFeature mapsActivation functionsPoolingFrom edges to higher-level features
4. 4

   Building a CNN with TensorFlow/Keras

   Create, train and troubleshoot an image-classification model

   Keras model workflowInput pipelinesCNN architectureTraining and validationOverfitting and regularisationSaving and loading models
5. 5

   Image classification project

   Build and evaluate a small end-to-end classifier

   Problem definitionDataset preparationModel trainingAccuracy and confusion matrixError analysisProject documentation
6. 6

   Transfer learning with pretrained models

   Use MobileNet to build a practical classifier for modest computing resources

   Pretrained ImageNet modelsMobileNet input preprocessingFeature extractionFine-tuning basicsModel size and inference considerations
7. 7

   Object detection basics

   Learn how models locate and label multiple objects in images

   Classification versus detectionBounding boxesConfidence scoresIntersection over UnionDetection datasets and annotationsResponsible use and limitations
8. 8

   Capstone - image classifier for a local problem

   Prototype a classifier using phone photos of healthy and visibly affected coffee or maize leaves

   Problem scopingEthical image collection and consentData labelling and quality checksTraining and evaluationTesting on new photosPresenting limitations and next steps