Computer Vision Overview

Principles

Definition

Computer vision is a field of artificial intelligence and computer science dedicated to enabling computers to interpret and understand the content of digital images and videos, effectively automating tasks that the human visual system performs. This research area investigates methods for acquiring, processing, analyzing, and understanding visual data to extract meaningful information and inform decision-making.

Ontological type

Core Methods

Foundational Tasks

Datasets and Benchmarks

Key developments

Key Figures

Feature and Model Vision era

Jitendra Malik [1] was a prominent figure at Stanford University [3] and the University of California, Berkeley [4] during the Feature and Model Vision era. His key contributions in this era include the introduction of 'Normalized cuts and image segmentation' [7], which provided a graph-based formulation for image segmentation that linked local cues with global interpretation. Martha J. Farah [2] was associated with Harvard University [5] and the Institute of Cognitive and Brain Sciences [6] during this era. Her notable contribution in this era is the 1993 paper 'Parts and Wholes in Face Recognition' [8], which argued for decomposing recognition into parts and wholes and shaped modular, feature-based approaches.

Discriminative Detection and Learning era

Ross Girshick [1], active across University of Washington [3] and University of California, Berkeley [4] during this era, helped shape discriminative object detection and learned representations. His key contributions include Object Detection with Discriminatively Trained Part-Based Models [9], Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation [8], and You Only Look Once: Unified, Real-Time Object Detection [7], which collectively defined the discriminative, end-to-end learning trajectory for robust detection in this era. Trevor Darrell [2], affiliated with Massachusetts Institute of Technology [5] and University of Pennsylvania [6] during this era, helped drive end-to-end discriminative architectures in this field. His contributions include Fully Convolutional Networks for Semantic Segmentation [10] and Long-term Recurrent Convolutional Networks for Visual Recognition and Description [11], enabling end-to-end dense prediction and temporal context, which were pivotal for scalable recognition in the era.

Contextual Deep Vision Synthesis era

Ross Girshick [1], a central figure in contextual deep vision, is associated with the University of Washington [3] and the University of California, Berkeley [4] in this era. His key contributions in this era include Faster R-CNN [7], which brought Region Proposal Networks for real-time object detection, and Feature Pyramid Networks [8], which enabled robust multi-scale feature fusion. Kaiming He [2] is another pivotal figure linked to the era through Menlo School [5] and Shanghai University [6]. His contributions across Faster R-CNN [7] and Feature Pyramid Networks [8] helped shape backbone design and scalable architectures that improved accuracy and deployment efficiency.