Computer Vision (CV)
In short
The field of AI that teaches machines to “see” — to understand and make decisions based on images, video, and other visual data.
Think of a quality inspector at a factory who stares at thousands of products sliding by on a conveyor belt, spotting tiny scratches or defects. Computer Vision is like cloning that inspector — except the clone never gets tired, never blinks, and can check a thousand items per second.
Computer Vision is all about getting machines to extract useful information from visual stuff — photos, video feeds, medical scans, satellite images, you name it. When you unlock your phone with your face, that’s CV. When a Tesla detects a stop sign, that’s CV. When a radiologist’s software highlights a suspicious spot on a lung X-ray, that’s CV too.
For years, the go-to architecture for CV was the Convolutional Neural Network (CNN). CNNs work by scanning an image in small patches, detecting low-level features first (edges, textures) and building up to high-level ones (faces, objects). They were kind of the breakthrough that made CV actually useful — before CNNs, computer vision was pretty rough. The real-world applications are everywhere: self-driving cars use CV to detect lanes, pedestrians, and other vehicles in real time. Manufacturing plants use it for automated quality control, catching defects that a human eye would miss. In healthcare, Deep Learning-based CV models can detect cancers, retinal diseases, and fractures from medical images, sometimes more accurately than trained specialists. And facial recognition — love it or hate it — powers everything from airport security to unlocking your phone.
More recently, the field has shifted toward Vision Transformers (ViTs), which apply the same Transformer architecture that powers LLMs but for images. Instead of scanning in patches like a CNN, a Vision Transformer chops an image into a grid of patches and processes them kind of the same way a language model processes words. This turned out to work really well, especially for large-scale tasks.
The biggest shift right now is Multimodal models — systems that bridge vision and language. Models like GPT-4o, Gemini, and Claude can look at an image and answer questions about it, describe what’s happening, or even generate images from text descriptions. This blurs the line between CV and Natural Language Processing in a way that would have seemed like science fiction just a few years ago. You’re no longer building a separate “vision system” and a separate “language system” — it’s all one model that can see and talk.
Related
- AI - CV is a subfield of AI
- Deep Learning - most modern CV runs on deep learning
- Multimodal - modern models combine vision with language
- Natural Language Processing - the “other side” — text instead of images
- Neural Network - CNNs are a type of neural network
- Transformer - Vision Transformers brought the transformer revolution to CV