Computer Vision

What you'll learn

• Represent images as arrays, convert between colour spaces and apply filters, edge detection and thresholding.
• Detect and match keypoints and apply geometric transforms and data augmentation.
• Classify images with pretrained CNNs and adapt them with transfer learning.
• Run object detection and segmentation models and evaluate them with IoU, NMS, average precision and Dice.
• Test robustness and avoid preprocessing mistakes when deploying vision models.
• Discuss modern vision foundation models and the privacy and fairness issues of computer vision.

Syllabus

Images and Vision Tasks

1. What Computer Vision Does
2. Images as Arrays of Pixels
3. Colour Spaces

Classical Image Processing

1. Filtering and Denoising
2. Edge Detection
3. Thresholding and Connected Components
4. Keypoints and Feature Matching

Geometry, Augmentation and Data

1. Geometric Transforms
2. Data Augmentation
3. Datasets and Labelling

Image Classification With Deep Learning

1. CNN Architectures in Brief
2. Classifying With a Pretrained Model
3. Transfer Learning: Linear Probes and Fine-Tuning

Object Detection

1. Bounding Boxes and IoU
2. Non-Maximum Suppression
3. Running a Pretrained Detector
4. Evaluating Detectors: Precision, Recall and AP

Segmentation

1. Semantic, Instance and Panoptic Segmentation
2. Running a Pretrained Segmentation Model
3. Segmentation Metrics: IoU and Dice

Robustness and Deployment

1. Robustness to Real-World Conditions
2. Preprocessing Consistency
3. Deploying Vision Models

Foundation Models, Ethics and a Checklist

1. Vision Foundation Models
2. Privacy, Fairness and Responsible Use
3. A Computer Vision Project Checklist