Computer Vision
Computer Vision is the field of AI that trains computers to interpret and understand visual information from images and videos. The human visual system is ...
Making Machines See
Computer Vision is the field of AI that trains computers to interpret and understand visual information from images and videos. The human visual system is so effortless that people rarely appreciate how computationally demanding vision actually is. An image is simply a grid of pixel values. Extracting from that grid the understanding that it shows a specific dog breed playing in a park requires solving problems of feature detection, object recognition, scene understanding, and spatial reasoning simultaneously.
Early computer vision relied heavily on hand-crafted feature detectors such as SIFT and HOG, which identified corners, edges, and gradient patterns that human engineers believed to be diagnostically useful. These features were then fed into classical classifiers like support vector machines. Performance plateaued because hand-crafted features failed to capture the full complexity of visual variation caused by lighting, viewpoint, occlusion, and deformation.
The 2012 ImageNet competition marked a turning point when AlexNet, a deep convolutional neural network, reduced the top-five error rate by more than ten percentage points compared to the best hand-crafted approach. Convolutional layers automatically learn hierarchical features, detecting edges in early layers, shapes in middle layers, and high-level object parts in deeper layers. Today, vision transformers and diffusion models push performance even further across tasks including image classification, object detection, semantic segmentation, and image generation.
Object Detection in Practice
Object detection goes beyond classifying an entire image by also localising each object with a bounding box. The YOLO family of models achieves real-time detection by dividing the image into a grid and predicting bounding boxes and class probabilities for each cell in a single forward pass, making it practical for applications like autonomous vehicle pedestrian detection.
Exam Tip
Exam questions on computer vision frequently ask you to explain convolution. A convolutional layer applies a small learnable filter by sliding it across the input image, computing the dot product at each position to produce a feature map. This operation detects local patterns while being translationally invariant. Sketching a simple three-by-three filter applied to a five-by-five input and showing the resulting three-by-three output is an effective way to demonstrate understanding.