Computer vision

An artificial neural network is a computing system of many simple connected units whose behaviour is set by connection weights. Its roots lie in 1943 models of neurons, and its learning versions began with Frank Rosenblatt's perceptron in 1958.

Geoffrey Hinton is a computer scientist born in London whose research on neural networks, including back-propagation and deep convolutional networks, is central to deep learning. He shared the 2018 ACM A.M. Turing Award and the 2024 Nobel Prize in Physics.

In 1958 Frank Rosenblatt described the perceptron, a model whose connection strengths change with training so that it learns to classify stimuli by trial and error. It became the starting point for trainable neural networks.

The Mark I Perceptron was the first hardware implementation of the perceptron algorithm: a custom analog computer built at Cornell Aeronautical Laboratory for image recognition, with photocells as input and motor-driven potentiometers as learnable weights.

Yann LeCun is a French computer scientist known for convolutional neural networks, which learn to recognise images directly from pixels. He co-authored the 1998 paper that introduced LeNet-5 and shared the 2018 ACM A.M. Turing Award.

Shakey was a mobile robot built at SRI International between 1966 and 1972, the first AI-based mobile robot that could perceive and reason about its surroundings and act without step-by-step instructions. Its project produced the A* search algorithm and the STRIPS planning system.

ALVINN was a 1988 Carnegie Mellon neural network that learned to steer the Navlab van by watching a human driver, an early demonstration of learned end-to-end driving.

A convolutional neural network applies the same small pattern detectors across a whole image, so a feature is found wherever it appears. Trained with backpropagation in 1989 and refined as LeNet-5 in 1998, it became standard in computer vision.

CIFAR-10 is a labelled image dataset of 60000 32×32 colour images in 10 mutually exclusive classes, split into 50000 training and 10000 test images; it was created by Krizhevsky, Nair and Hinton for training and comparing image classifiers.

The 2009 paper introduced ImageNet, a large, cleanly labelled image database organised by the WordNet hierarchy. It later became the basis of an annual benchmark, the ImageNet Large Scale Visual Recognition Challenge.

A CVPR 2011 Best Paper that recovers 3D body joint positions from a single depth image without temporal information, by classifying pixels into body parts, and runs at 200 frames per second on consumer hardware.

The 2012 NeurIPS paper that introduced AlexNet, a large convolutional network trained on GPUs. It won the ILSVRC-2012 image classification competition by a wide margin and is widely seen as a turning point for deep learning in computer vision.

A 2014 study showing that stacking many small 3×3 convolution filters, up to 16–19 weight layers, improves large-scale image recognition; its models took first place in localisation and second in classification at ImageNet Challenge 2014 and were released publicly.

The 2014 paper that introduced the Inception architecture and its 22-layer GoogLeNet network, which took first place in the ImageNet 2014 classification and detection challenge with a 6.67% top-5 error.

Mask R-CNN is a 2017 framework for object instance segmentation. It detects objects and produces a high-quality mask for each instance by adding a parallel mask branch to Faster R-CNN.

This 2018 paper introduces a style-based GAN generator that separates high-level attributes from stochastic variation and allows scale-specific control, improving distribution quality, interpolation, and disentanglement.

A 2020 paper by Mildenhall, Srinivasan, Tancik, Barron, Ramamoorthi and Ng that represents a scene as a neural radiance field: a network mapping 5D coordinates to density and colour, optimized from a sparse set of images with known camera poses.

Tesla's most advanced driver-assistance package, sold as a $99 monthly subscription. It steers, changes lanes, follows routes and parks, while the driver stays responsible and the car is not autonomous.

A 2020 paper showing that a plain Transformer applied directly to sequences of image patches, with no convolution at all, can match state-of-the-art convolutional networks on image classification.

A 2023 paper that introduces the Segment Anything project: a promptable segmentation task, a model, and a dataset of over 1 billion masks on 11 million images, all released to advance foundation models for computer vision.

Claude 3 Opus, announced on 4 March 2024, was the most capable model in the Claude 3 family. It added vision and launched with a 200K-token context window and a Needle In A Haystack recall above 99%.

Claude 3.5 Sonnet, released on 20 June 2024, was the first model of the Claude 3.5 family. It ran twice as fast as Claude 3 Opus at the mid-tier price and powered the new Artifacts and Projects features.

Google's self-driving car prototype was a modified production car that used cameras, radar and a roof-mounted laser range finder to sense traffic, drove routes that had been mapped in advance by manually driven cars, and always carried a trained safety driver and a software operator.

A multimodal model handles more than one kind of input or output, such as images together with text, inside a single model. Work on CLIP and Flamingo showed how such models are trained and how they can be steered toward new tasks.

ImageNet is a large, hierarchical image database built on the WordNet conceptual model and released at Princeton University in 2009. It holds more than 14 million images in 21,841 categories, with bounding-box annotations, and a 1,000-class subset used by the ILSVRC challenge.

A public research university in Toronto where Geoffrey Hinton has been affiliated since 1987 and where AlexNet, the 2012 deep convolutional network, was developed.

Bell Labs was the research arm of the Bell System, where the transistor, Unix and the C language were created. Yann LeCun's convolutional networks for handwritten digits were developed there, which made it an early home of neural-network research.

The Stanford Artificial Intelligence Laboratory, started by John McCarthy in 1963, is associated with the Stanford arm robot design, the Stanley self-driving car and the SAIL programming language.

Stanford computer scientist Fei-Fei Li created ImageNet, the large-scale image dataset and challenge that gave data-driven computer vision a common benchmark, and has led AI research in both academia and industry.

CVPR is the annual Conference on Computer Vision and Pattern Recognition, first held in 1983 and usually held in June in North America. It is known for selective, double-blind review and open-access papers.

AI researcher and educator, born in Bratislava and raised partly in Toronto, who co-founded OpenAI, led AI at Tesla, taught Stanford's CS231n course, and later joined Anthropic's pre-training team.

A 3-layer back-propagation network that takes camera and laser range finder images and outputs the direction to steer, trained on simulated road images and tested on Carnegie Mellon's autonomous navigation test vehicle.

In 1989 LeCun and colleagues at AT&T Bell Laboratories trained a convolutional network with backpropagation to read handwritten zip-code digits from U.S. mail, reaching 5.0% test error. The work showed that the convolutional design could be learned end to end on a real task.

Microsoft's corporate research division, founded in 1991, whose researchers produced work such as the ResNet image-recognition architecture and helped build features used in Kinect, Bing, HoloLens, Cortana and Microsoft Translator.

MNIST is a database of 60,000 training and 10,000 test images of handwritten digits, normalized to 28×28 pixels and described in the 1998 paper by LeCun and colleagues as a shared benchmark for document recognition.

LeCun, Bottou, Bengio and Haffner’s 1998 review argued that learning systems should replace hand-designed features. It showed that convolutional networks such as LeNet-5 outperformed other methods on handwritten digits, and it introduced the MNIST dataset.

CIFAR-10 and CIFAR-100 are labeled image datasets drawn from the 80 million tiny images collection, documented in a 2009 technical report by Alex Krizhevsky, Vinod Nair and Geoffrey Hinton. CIFAR-10 holds 60,000 32×32 colour images in 10 classes.

A 2009 technical report by Alex Krizhevsky that introduces CIFAR-10 and CIFAR-100 and studies multi-layer generative models for learning features from tiny colour images.

A consumer graphics card NVIDIA released in November 2010 on its Fermi architecture. Two of them, each with 3 GB of memory, later trained AlexNet, the network that won ILSVRC-2012 and which NVIDIA's timeline credits with sparking the era of modern AI.

A free beta software development kit released by Microsoft on 16 June 2011 that let developers use the Kinect sensor's depth sensing, motion tracking and voice recognition on Windows PCs, initially for non-commercial work.

AlexNet is the eight-layer convolutional network that won the ILSVRC2012 image classification task by a wide margin, showing that a large network trained on a large labelled dataset with GPUs could beat hand-designed features.

In 2012, Quoc Le, Andrew Ng and colleagues trained a large unsupervised neural network on 10 million internet images and showed it could detect faces without a single labelled face.

R-CNN, posted to arXiv by Girshick, Donahue, Darrell and Malik in November 2013, applied convolutional neural networks to region proposals for object detection and lifted mean average precision on PASCAL VOC 2012 to 53.3%, more than 30% above the previous best[^1].

Zeiler and Fergus ask why large convolutional networks perform so well on ImageNet, introduce a visualization technique for intermediate layers, and use ablation to find an architecture that outperforms Krizhevsky et al. The model also transfers to Caltech-101 and Caltech-256.

DeepFace is a 2014 face verification pipeline that rebuilt the alignment and representation stages of the standard face pipeline, reaching 97.35% accuracy on Labeled Faces in the Wild with a nine-layer network of over 120 million parameters.

A 2014 paper introducing Microsoft COCO, a large dataset of everyday scenes in which common objects are labeled with per-instance segmentations, built to study object recognition as part of scene understanding.

Caffe is a BSD-licensed C++ library with Python and MATLAB bindings for training and deploying convolutional neural networks, described in a 2014 paper by Yangqing Jia and colleagues.

A 2014 paper from Google researchers that introduced the Inception architecture, whose GoogLeNet model set a new state of the art in ILSVRC 2014 by widening and deepening a network within a fixed computational budget.

SenseTime is a Chinese AI company founded in October 2014 and listed in Hong Kong at the end of 2021. It develops the SenseNova large models and, in 2026, reported its first IFRS net profit since listing.

A 2014 paper by Vinyals and colleagues that trained a single deep network to turn a photograph into a sentence, reporting a BLEU-1 score of 59 on Pascal against 25 for the previous state of the art.

A convolutional network for biomedical image segmentation, posted to arXiv on May 18, 2015, that trains end to end from very few annotated images and won the ISBI cell tracking challenge 2015 in its categories.

A 2015 paper that introduced a Region Proposal Network sharing full-image convolutional features with the detection network, making proposals nearly cost-free and enabling end-to-end training, 5 fps GPU detection, and state-of-the-art accuracy.

The 2015 paper introduced YOLO, which treats object detection as a single regression from image pixels to bounding boxes and class probabilities. One neural network makes predictions in one evaluation, reaching real-time speed.

The 2015 YOLO paper recast object detection as a single neural network's regression over bounding boxes and class probabilities, achieving real-time detection while reducing false positives on empty background.

The 2015 paper introduced residual networks, which add each layer's input back to its output through a shortcut connection. This made networks of over a hundred layers trainable and won first place in several ILSVRC and COCO tasks in 2015.

CycleGAN learns image-to-image translation between two domains without aligned pairs, combining an adversarial loss with a cycle-consistency constraint to constrain the mapping.

A 2017 paper by Lin, Goyal, Girshick, He and Dollar that argues dense one-stage detectors are held back by extreme foreground-background imbalance and proposes focal loss, a reshaped cross-entropy that down-weights well-classified examples.

A 2017 paper by Karras, Aila, Laine and Lehtinen that trains generative adversarial networks by growing the generator and discriminator from low to high resolution, improving stability, quality and variation.

A 2018 paper by Andrew Brock, Jeff Donahue and Karen Simonyan that trained generative adversarial networks at the largest scale attempted at the time, introduced BigGANs and set a new state of the art in class-conditional image synthesis.

A 2020 paper by Alexey Dosovitskiy and colleagues showing that a pure Transformer applied directly to sequences of image patches — the Vision Transformer, or ViT — matches state-of-the-art convolutional networks on image classification when pre-trained on large amounts of data.

An image-text model from OpenAI that learns to match pictures with their captions. Trained on about 400 million image-caption pairs, it can classify images by comparing them against written prompts, with no task-specific training.

CLIP is an OpenAI model trained on 400 million image-text pairs to map pictures and sentences into one shared space, so it can classify images zero-shot from a written description.

The Segment Anything Model (SAM) is a promptable image segmentation model Meta AI released in April 2023. Given a point, a box or a mask, it outlines the object the prompt refers to.

GPT-4V is a version of GPT-4 that accepts image input alongside text, documented in OpenAI's system card of 25 September 2023 and released to developers through preview API models; image understanding later became native in GPT-4o.

Fuyu-8B is a decoder-only multimodal model that Adept open-sourced in October 2023; it drops the image encoder, accepts arbitrary resolutions, and reads screens and charts.

Claude 3 Sonnet was the middle model of Anthropic's Claude 3 family, announced on 4 March 2024. It powered the free tier of claude.ai, offered a 200K-token context window, and was retired on 21 July 2025.

Claude 3 Haiku is the smallest and fastest model in Anthropic's Claude 3 family, announced in March 2024 with low per-token pricing, a 200K-token context window and image input; retired in April 2026.

SAM 2 is a Meta FAIR foundation model that segments objects in images and videos from simple prompts such as points, boxes or masks, released together with a large video segmentation dataset and training code.

Llama 3.2, announced on 25 September 2024, added vision models at 11 billion and 90 billion parameters and small text-only models at 1 billion and 3 billion for edge and mobile devices.

Computer use is an Anthropic capability, released in public beta in October 2024, that lets an AI model operate a computer the way a person does — reading the screen, moving a cursor, clicking and typing — so it can drive ordinary software built for humans.

Gemini Robotics is a vision-language-action model from Google DeepMind, built on Gemini 2.0, that outputs physical actions to control robots. It was announced on 12 March 2025 alongside Gemini Robotics-ER.

Tesla Robotaxi is Tesla's ride-hailing service, in which passengers hail autonomous vehicles through the company's app; its dedicated two-seat Cybercab has no steering wheel, pedals or mirrors. Limited service began in Austin in June 2025.

Qwen3-VL is Alibaba's open vision-language model family, released in September 2025 under Apache 2.0. Its 235B flagship handles long video, GUI operation and 2D and 3D grounding.

Reka Edge is a 7-billion-parameter vision-language model from Reka AI, released on 11 March 2026, built to take images, video and text and to run locally on laptops, phones and small edge devices.

EXAONE 4.5 is LG AI Research's first open-weight vision language model: a 33-billion-parameter model with a 1.2-billion-parameter vision encoder, trained on documents and text, with a 256K-token context window, released in April 2026.

Released on 16 April 2026 at the same price as Opus 4.6, Claude Opus 4.7 added an xhigh effort level and accepted much larger images. It was the first model to test automatic blocking of high-risk cybersecurity requests.

Visual odometry lets a rover measure how far it actually moved by tracking features in camera images taken before and after a short drive, instead of trusting wheel turns. NASA's Mars Exploration Rovers were the first to demonstrate it.

AutoNav is the autonomous driving software used by NASA's Spirit and Opportunity Mars rovers. It combines stereo vision, GESTALT terrain assessment, Visual Odometry, and Field D* path planning to let rovers choose and drive safe paths without step-by-step direction from Earth.

Harry Shum is a computer scientist who spent most of his career at Microsoft, helping found Microsoft Research Asia in Beijing and later leading the company's AI and Research group. He is an IEEE Fellow and ACM Fellow recognised for computer vision and graphics.

Runway builds generative video, image and audio tools used by more than 60 million creative users, and researches general-purpose world models that simulate video, audio and interfaces in real time.

Reka AI is an AI research lab building models and infrastructure for the physical AI era. Its releases include reka-flash-3, a 21-billion-parameter reasoning model, and Reka Edge, a 7-billion-parameter vision-language model designed to run on devices at the edge.

SRI International is a non-profit research institute in Menlo Park, California, founded in 1946 as the Stanford Research Institute. Its AI Center built Shakey the robot, and its later work led to Siri.

Quoc Le is a machine-learning researcher whose 2013 Stanford Ph.D. thesis was on scalable feature learning. He co-authored papers on large-scale unsupervised learning, sequence-to-sequence models, neural architecture search and EfficientNet; the sequence-to-sequence paper won a NeurIPS 2024 Test of Time award.

Kaiming He is a computer-vision and deep-learning researcher, first author of the 2015 ResNet paper that introduced residual connections, now a tenured associate professor at MIT.

Princeton University, chartered in 1746 as the College of New Jersey, is where Turing, McCarthy and Minsky took their doctorates and where Fei-Fei Li began ImageNet; seventeen Turing Award laureates have been affiliated with it.

New York University, chartered in 1831 and centred on Washington Square Park, is a private research university whose Courant Institute and Center for Data Science made it a home for machine learning research and data science training.

Tang Xiaoou (1968–2023) was a Chinese computer vision researcher at the Chinese University of Hong Kong and co-founder and chairman of SenseTime, one of China's best-known AI companies.

Ali Farhadi (born 1982) is a computer scientist and professor at the University of Washington who served as chief executive of the Allen Institute for AI from 2023 to 2026, a period that included fully open releases of the OLMo, Molmo and Tülu models.

Microsoft Research Asia is Microsoft's research laboratory in the Asia-Pacific region, founded in Beijing in November 1998. It works on artificial intelligence, human-computer interaction and related fields, and its researchers have included the first author of the ResNet paper.

Megvii is a Chinese computer vision company founded in 2011 by three Tsinghua graduates. Its Face++ cloud platform opened face recognition to developers, and it was the earliest founded of the "CV four little dragons."

Yitu Technology is a Chinese computer vision company founded in September 2012 by Zhu Long and Lin Chenxi. It is counted among the "CV four little dragons", alongside Megvii, SenseTime and CloudWalk, that grew out of computer vision in China's early AI industry.

A 2015 arXiv paper by Kelvin Xu, Jimmy Ba, Ryan Kiros, Kyunghyun Cho and co-authors that introduced an image-captioning model with visual attention, able to show where it is looking while writing each word.

CloudWalk Technology is a Chinese AI company built on academic face recognition research. Founded in April 2015, it was the last of the four Chinese computer vision firms known as the "CV four little dragons" to be established.

Flamingo is a DeepMind visual language model announced in April 2022. It takes prompts that mix images, videos and text and replies in language, learning new tasks from only a few examples.

RT-2 is a vision-language-action model from Google DeepMind, announced on 28 July 2023, that turns images and text instructions into robot actions written as text tokens. In over 6,000 trials it reached 62% on unseen scenarios, up from 32% for RT-1.

Qwen-VL is a family of vision-language models released in August 2023 that builds on the Qwen language model to caption images, answer questions about them, locate objects and read text, with open Base and Chat versions.

Yi-VL-34B and Yi-VL-6B are open vision-language models from 01.AI, released on 23 January 2024. They read images and text together and ranked first among open models on the MMMU and CMMMU benchmarks.

Step-1V is a roughly 100-billion-parameter multimodal model from StepFun, unveiled in March 2024, that reads text, data and charts in images and supports video understanding; it ranked first on OpenCompass at launch and was succeeded by Step-1.5V.

Grok-1.5 is xAI's upgrade to Grok-1, with better reasoning and a 128,000-token context window. Its companion Grok-1.5V was xAI's first multimodal model. Neither model's weights were released publicly.

Project Astra is a Google DeepMind research prototype of a universal AI assistant that watches and listens while it responds, shown at Google I/O in May 2024 and later rebuilt on Gemini 2.0.

SAM 2 is Meta's 2024 successor to the Segment Anything Model. It segments objects in both images and videos in real time, and its release included the SA-V video dataset.

Qwen2-VL is Alibaba's vision-language model released on 29 August 2024. It understands videos longer than 20 minutes, reads text in many languages, and can act as an agent on phones and robots.

Pixtral 12B is Mistral AI's first multimodal model, pairing a 12B language decoder with a 400M vision encoder. Released openly under Apache 2.0 in September 2024, it reads images at native resolution and was retired at the end of 2025.

Molmo is a family of open vision-language models released by Ai2 on 25 September 2024 that can point at objects in images, with weights, code, data and evaluations published openly.

Pixtral Large is Mistral AI's open-weight multimodal model announced on 18 November 2024, built on Mistral Large 2, combining a 123B-parameter decoder with a 1B vision encoder and a 128K-token context window. It was retired in May 2026.

A family of open mixture-of-experts vision-language models from DeepSeek, released in three sizes from 1.0B to 4.5B activated parameters for document, chart and OCR understanding, succeeding DeepSeek-VL.

Qwen2.5-VL is Alibaba's vision-language model family released in January 2025 in 3B, 7B, 32B and 72B sizes. It reads charts and documents, localises objects, works as a visual agent and understands videos over an hour long.

Hunyuan3D 2.0 is Tencent's large-scale system for generating high-resolution textured 3D assets: a shape diffusion transformer plus a texture model, released with code and pretrained models on 21 January 2025 alongside its technical report.

UI-TARS is a GUI agent model from ByteDance Seed: it takes only screenshots as input and outputs keyboard and mouse actions, reported state-of-the-art results on more than 10 benchmarks, and was released with open weights under Apache-2.0.

Cohere's first vision-language model, released in March 2025 as open-weight 8B and 32B versions. It handles images and text in 23 languages and, in Cohere's comparisons, beats several much larger models, under a non-commercial licence.

Kimi-VL is an open-source vision-language model from Moonshot AI, released in April 2025. It pairs a 16B-parameter mixture-of-experts language model with a native-resolution vision encoder, so only about 3B parameters are active at a time.

ByteDance Seed's GUI agent model from April 2025 and the successor to UI-TARS. Its 7B version is open-source under Apache-2.0, and the family reports state-of-the-art results on several screen-control benchmarks.

Command A Vision is Cohere's first vision-language model in the Command family, released in July 2025. It pairs a 112-billion-parameter Command A language model with a SigLIP2 image encoder to read images, charts and documents and answer in text.

A vision-language model released by Zhipu on 11 August 2025, built on GLM-4.5-Air and covering image, video, long-document and GUI tasks; Zhipu reports state-of-the-art results at its scale on 42 benchmarks.

UI-TARS-2 is a GUI agent trained with multi-turn reinforcement learning and released in September 2025 as an all-in-one upgrade to UI-TARS-1.5, with benchmark scores across web, desktop, Android, games and software-engineering tasks.

DeepSeek-OCR is an open-source model of about 3B parameters that renders long text as 2D images and reads it back from a small number of vision tokens, testing whether optical compression can stand in for very long text contexts.

Baidu's open-weight multimodal reasoning model from November 2025: 28B total parameters with 3B active per token, "Thinking with Images" zoom-and-tool reasoning, and Apache 2.0 weights.

SAM 3 is Meta's segmentation model from November 2025. It finds, segments and tracks every instance of a concept named by a short text prompt or an example image, and Meta reported it doubles the accuracy of existing systems on this task.

Molmo 2 is the second generation of Ai2's open vision-language models, released on 11 December 2025, which extends the Molmo line from still images to video with pointing, tracking and timestamped answers.

Phi-4-reasoning-vision-15B is Microsoft's 15-billion-parameter open-weight vision-language model, released on 4 March 2026 under the MIT licence, built to switch between step-by-step reasoning for math and science and direct answers for image perception tasks.

Zhu Songchun is a computer vision researcher who worked at the University of California, Los Angeles from 2002 and was named in 2020 to lead Peking University's Institute for Artificial Intelligence.

Yin Qi is a Chinese AI entrepreneur who co-founded the facial-recognition company Megvii with Tang Wenbin and Yang Mu after graduating from Tsinghua, and later became chairman of Qianli Technology, turning to smart automotive technology.

Sun Jian was a computer-vision scientist who spent 13 years at Microsoft, co-created ResNet, and later became chief scientist at Megvii. He died aged 45 after a sudden illness in June 2022.

XRCE's computer-vision team competed in the ImageNet challenge and won the 2011 classification task using high-dimensional image signatures, product quantization and linear SVMs; its Fisher-vector methods performed strongly across the 2010–2014 challenges.

Xu Li is a co-founder, chairman and chief executive officer of SenseTime, the Chinese artificial intelligence company he helped start in 2014. He studied computer science at Shanghai Jiao Tong University and the Chinese University of Hong Kong.

NVLM is a family of multimodal large language models from NVIDIA, described in a paper posted to arXiv in September 2024; its flagship NVLM-D-72B was later released as open weights for non-commercial use.

Mistral OCR is a document-understanding API released by Mistral AI on 6 March 2025 that reads text, tables, equations and embedded media across thousands of scripts and languages, and was later superseded by newer OCR versions.

Robostral Navigate is Mistral AI's first embodied navigation model: an 8-billion-parameter system that steers wheeled, legged or flying robots from RGB images and a plain-language instruction, using a single ordinary camera and no depth sensor.