A field of AI that enables machines to interpret and understand visual information from images and video — detecting objects, recognizing faces, reading scenes, and extracting actionable insights from pixels.
In Depth
Computer Vision is the discipline of enabling machines to interpret visual information the way humans do — and in many cases, far more precisely and quickly. It sits at the intersection of image processing, deep learning, and geometry, using neural networks (primarily CNNs and Vision Transformers) to extract meaningful features from raw pixels. A computer vision system doesn't 'see' in the human sense; it transforms arrays of pixel values into structured representations — bounding boxes, class labels, segmentation masks, depth maps — that downstream systems can act upon.
The field encompasses multiple levels of visual understanding. Image classification assigns a category to an entire image. Object Detection finds and localizes multiple objects within an image. Semantic Segmentation labels every pixel with a class. Instance Segmentation goes further, distinguishing individual instances of the same class. Pose Estimation identifies the positions of human body keypoints. Optical Character Recognition (OCR) extracts text from images. Each level requires progressively more detailed spatial understanding.
Computer Vision has benefited enormously from the deep learning revolution. The ImageNet moment in 2012 — when AlexNet's CNN reduced the image classification error rate from 26% to 16% — marked the beginning of an era of rapid progress. Today's systems achieve superhuman performance on many standard benchmarks. Multi-modal models that jointly process images and text (CLIP, GPT-4V, Gemini) are expanding computer vision beyond pure visual tasks toward broader scene understanding and visual reasoning.
Computer Vision gives machines the ability to extract structured, actionable information from visual data — transforming pixels into meaning and enabling AI to operate in the physical, visual world.