Computer vision

Publishing authors

Base scenario: 32k publishing authors. Paper count × 3.96 known authors per paper ÷ 3 papers per author per year. The range assumes 1–5 papers per author; the base and both ends of the range are at least the 3,911 distinct Author IDs already observed in the sample.

The publication rate is an assumption, not a measured rate for this field. The range is a scenario, not a confidence interval.

Random sample: 1,000 of 24,491 papers (4.1%). 11.8% of authorship records lack an Author ID; 24.2% of retrieved works lack an abstract. Retrieved 2026-09-26.

Mean known authors per paper: 3.96; 95% bootstrap interval 3.79–4.15. This describes sampling variability, not search accuracy or uncertainty in the number of researchers.

Applied coauthors are included. Missing or incorrectly linked author records affect these figures. They do not measure research jobs or everyone working in a field.

Private investment

An assumed allocation of broader investment segments. The range is 0.5–2× the base case. Shares are editorial assumptions, not measured deals or research spending.

Base case: $5.54B.

Retail: $4.10B × 10.00%
Internet of things: $14.6B × 15.00%
Medical and healthcare: $11.8B × 25.00%

Company investment, not revenue or research spending. Ranges are scenarios, not confidence intervals.

Scope and related research

Medical and industrial uses also appear in the applications view.

Organizations

Selected examples

Meta Models
Segment Anything models perform promptable image segmentation.
Roboflow Product
Computer vision development.
Cognex Product
Deep-learning tools for industrial image inspection and recognition.

More organizations (5)

Mobileye Product
Driving perception and autonomy.
OpenAI Research
CLIP supports open-vocabulary image classification using text descriptions.
Paige Product
AI pathology.
Part of Tempus
PathAI Product
AI pathology.
Qure.ai Product
Medical imaging AI.

Video understanding and tracking

Action and event recognition, temporal localization, tracking, optical flow, and video question answering.

Video understanding Tracking

4,819 papers

1.3× vs. 2023

Authors · scenario: 4.0k–20k
Investment · scenario: $198M–$794M

Sources and methodology

Publication activity

2019: 2,192
2020: 2,498
2021: 2,885
2022: 3,252
2023: 3,754
2024: 4,317
2025: 4,819

Matches include papers applying these methods. Search precision and recall have not been systematically measured.

Title and abstract matches for articles, preprints and reviews; retracted work is excluded. Versions and overlapping areas may be counted more than once.

("action recognition" OR "video understanding" OR "video question answering" OR "temporal action localization" OR "object tracking" OR "optical flow")

Retrieved 2026-09-26.

Publishing authors

Base scenario: 6.7k publishing authors. Paper count × 4.18 known authors per paper ÷ 3 papers per author per year. The range assumes 1–5 papers per author; the base and both ends of the range are at least the 4,038 distinct Author IDs already observed in the sample.

The publication rate is an assumption, not a measured rate for this field. The range is a scenario, not a confidence interval.

Random sample: 1,000 of 4,819 papers (20.8%). 12.6% of authorship records lack an Author ID; 21.3% of retrieved works lack an abstract. Retrieved 2026-09-26.

Mean known authors per paper: 4.18; 95% bootstrap interval 4.02–4.34. This describes sampling variability, not search accuracy or uncertainty in the number of researchers.

Applied coauthors are included. Missing or incorrectly linked author records affect these figures. They do not measure research jobs or everyone working in a field.

Private investment

An assumed allocation of broader investment segments. The range is 0.5–2× the base case. Shares are editorial assumptions, not measured deals or research spending.

Base case: $397M.

Autonomous vehicles: $7.94B × 5.00%

Company investment, not revenue or research spending. Ranges are scenarios, not confidence intervals.

Scope and related research

Video generation is a separate area. Vision-language models can support these tasks.

Organizations

Selected examples

TwelveLabs Models
Marengo and Pegasus model visual, audio and temporal information in video.
Google DeepMind Models
Gemini models process video for temporal understanding and question answering.
Meta Models
SAM 2 extends promptable segmentation and object tracking to video.

More organizations (2)

Oosto Product
Visual intelligence.
Reka AI Models
Multimodal models.

3D vision and spatial perception

Depth, stereo, structure from motion, reconstruction, pose estimation, and scene understanding.

Spatial AI 3D reconstruction Depth estimation

6,743 papers

1.5× vs. 2023

Authors · scenario: 6.3k–32k
Investment · scenario: $198M–$794M

Sources and methodology

Publication activity

2019: 2,636
2020: 3,045
2021: 3,361
2022: 3,669
2023: 4,372
2024: 5,220
2025: 6,743

Matches include papers applying these methods. Search precision and recall have not been systematically measured.

Title and abstract matches for articles, preprints and reviews; retracted work is excluded. Versions and overlapping areas may be counted more than once.

("depth estimation" OR "stereo matching" OR "structure from motion" OR "3D reconstruction" OR "point cloud understanding" OR "scene understanding")

Retrieved 2026-09-26.

Publishing authors

Base scenario: 11k publishing authors. Paper count × 4.70 known authors per paper ÷ 3 papers per author per year. The range assumes 1–5 papers per author; the base and both ends of the range are at least the 4,532 distinct Author IDs already observed in the sample.

The publication rate is an assumption, not a measured rate for this field. The range is a scenario, not a confidence interval.

Random sample: 1,000 of 6,743 papers (14.8%). 10.5% of authorship records lack an Author ID; 22.0% of retrieved works lack an abstract. Retrieved 2026-09-26.

Mean known authors per paper: 4.70; 95% bootstrap interval 4.51–4.88. This describes sampling variability, not search accuracy or uncertainty in the number of researchers.

Applied coauthors are included. Missing or incorrectly linked author records affect these figures. They do not measure research jobs or everyone working in a field.

Private investment

An assumed allocation of broader investment segments. The range is 0.5–2× the base case. Shares are editorial assumptions, not measured deals or research spending.

Base case: $397M.

Autonomous vehicles: $7.94B × 5.00%

Company investment, not revenue or research spending. Ranges are scenarios, not confidence intervals.

CVPR 2026 — research topics AAAI 2026 — AI taxonomy

Scope and related research

Reconstructs observed scenes. Generating new 3D content and robot navigation are related areas.

Organizations

Selected examples

NVIDIA Models
FoundationPose and FoundationStereo estimate object pose and stereo depth.
Wayve Research
Research on driving world models, 3D perception and reconstruction.
Skydio Product
Autonomous drones.

More organizations (2)

Exyn Technologies Product
Autonomous aerial mapping.
Tencent / Hunyuan Research
HY-World reconstructs, generates and simulates 3D environments.

Image restoration

Denoising, deblurring, super-resolution, inverse imaging, and physics-based vision.

Super-resolution Computational imaging

1,386 papers

1.1× vs. 2023

Authors · scenario: 3.8k–5.7k
Investment · scenario: $588M–$2.35B

Sources and methodology

Publication activity

2019: 521
2020: 748
2021: 919
2022: 1,070
2023: 1,235
2024: 1,297
2025: 1,386

Matches include papers applying these methods. Search precision and recall have not been systematically measured.

Title and abstract matches for articles, preprints and reviews; retracted work is excluded. Versions and overlapping areas may be counted more than once.

("image restoration" OR "image denoising" OR "image deblurring" OR "image super resolution" OR "computational imaging") AND (("machine learning" OR "deep learning" OR "neural network" OR "artificial intelligence"))

Retrieved 2026-09-26.

Publishing authors

Base scenario: 3.8k publishing authors. Paper count × 4.13 known authors per paper ÷ 3 papers per author per year. The range assumes 1–5 papers per author; the base and both ends of the range are at least the 3,810 distinct Author IDs already observed in the sample.

The publication rate is an assumption, not a measured rate for this field. The range is a scenario, not a confidence interval.

Random sample: 1,000 of 1,386 papers (72.2%). 10.5% of authorship records lack an Author ID; 27.7% of retrieved works lack an abstract. Retrieved 2026-09-26.

Mean known authors per paper: 4.13; 95% bootstrap interval 3.97–4.32. This describes sampling variability, not search accuracy or uncertainty in the number of researchers.

Applied coauthors are included. Missing or incorrectly linked author records affect these figures. They do not measure research jobs or everyone working in a field.

Private investment

An assumed allocation of broader investment segments. The range is 0.5–2× the base case. Shares are editorial assumptions, not measured deals or research spending.

Base case: $1.18B.

Medical and healthcare: $11.8B × 10.00%

Company investment, not revenue or research spending. Ranges are scenarios, not confidence intervals.

Scope and related research

Recovers a signal from measurements. Unconstrained image generation is listed separately.

Organizations

Selected examples

Topaz Labs Product
Image and video enhancement.
DxO Product
DeepPRIME uses neural networks for RAW-image denoising and demosaicing.
Adobe / Firefly Product
Camera Raw provides AI denoising and Super Resolution image enhancement.

More organizations (1)

Photoroom Product
Product imagery and editing.

OCR and document understanding

Text recognition, document layout, tables, formulas, document question answering, and structured extraction.

Document AI OCR

1,397 papers

1.9× vs. 2023

Authors · scenario: 3.1k–4.6k
Investment · scenario: $1.00B–$4.00B

Sources and methodology

Publication activity

2019: 421
2020: 474
2021: 548
2022: 562
2023: 749
2024: 865
2025: 1,397

Matches include papers applying these methods. Search precision and recall have not been systematically measured.

Title and abstract matches for articles, preprints and reviews; retracted work is excluded. Versions and overlapping areas may be counted more than once.

("optical character recognition" OR "document understanding" OR "document layout analysis" OR "document question answering" OR "table recognition" OR "scene text recognition")

Retrieved 2026-09-26.

Publishing authors

Base scenario: 3.1k publishing authors. Paper count × 3.29 known authors per paper ÷ 3 papers per author per year. The range assumes 1–5 papers per author; the base and both ends of the range are at least the 3,131 distinct Author IDs already observed in the sample.

The publication rate is an assumption, not a measured rate for this field. The range is a scenario, not a confidence interval.

Random sample: 1,000 of 1,397 papers (71.6%). 14.9% of authorship records lack an Author ID; 16.0% of retrieved works lack an abstract. Retrieved 2026-09-26.

Mean known authors per paper: 3.29; 95% bootstrap interval 3.07–3.57. This describes sampling variability, not search accuracy or uncertainty in the number of researchers.

Applied coauthors are included. Missing or incorrectly linked author records affect these figures. They do not measure research jobs or everyone working in a field.

Private investment

An assumed allocation of broader investment segments. The range is 0.5–2× the base case. Shares are editorial assumptions, not measured deals or research spending.

Base case: $2.00B.

Medical and healthcare: $11.8B × 5.00%
Accounting/finance: $2.59B × 40.00%
Legal tech: $3.79B × 10.00%

Company investment, not revenue or research spending. Ranges are scenarios, not confidence intervals.

Scope and related research

Combines vision and language; vision-language models are one implementation approach.

Organizations

Selected examples

Google Cloud Tools & infrastructure
Document AI recognizes text, layout and tables in documents.
Microsoft Tools & infrastructure
Document Intelligence recognizes printed and handwritten text and document layout.
ABBYY Product
Document AI.

More organizations (7)

Hyperscience Product
Document processing.
Instabase Product
Enterprise document AI.
LandingAI Product
Document extraction and document automation.
LlamaIndex Product
Document parsing and agents over enterprise information.
Rossum Product
Transactional document AI.
Upstage Models
Document AI and language models.
V7 Product
Document-centered workflow automation.

Human perception and biometrics

Face, body, hand and pose analysis, gesture recognition, motion analysis, and identification.

Pose estimation Biometrics

4,626 papers

1.1× vs. 2023

Authors · scenario: 3.5k–17k
Investment · scenario: $196M–$784M

Sources and methodology

Publication activity

2019: 2,823
2020: 3,198
2021: 3,439
2022: 3,657
2023: 4,039
2024: 4,181
2025: 4,626

Matches include papers applying these methods. Search precision and recall have not been systematically measured.

Title and abstract matches for articles, preprints and reviews; retracted work is excluded. Versions and overlapping areas may be counted more than once.

("human pose estimation" OR "face recognition" OR "gesture recognition" OR "human mesh recovery" OR "hand pose estimation" OR "person re identification")

Retrieved 2026-09-26.

Publishing authors

Base scenario: 5.6k publishing authors. Paper count × 3.62 known authors per paper ÷ 3 papers per author per year. The range assumes 1–5 papers per author; the base and both ends of the range are at least the 3,517 distinct Author IDs already observed in the sample.

The publication rate is an assumption, not a measured rate for this field. The range is a scenario, not a confidence interval.

Random sample: 1,000 of 4,626 papers (21.6%). 10.9% of authorship records lack an Author ID; 17.5% of retrieved works lack an abstract. Retrieved 2026-09-26.

Mean known authors per paper: 3.62; 95% bootstrap interval 3.48–3.77. This describes sampling variability, not search accuracy or uncertainty in the number of researchers.

Applied coauthors are included. Missing or incorrectly linked author records affect these figures. They do not measure research jobs or everyone working in a field.

Private investment

An assumed allocation of broader investment segments. The range is 0.5–2× the base case. Shares are editorial assumptions, not measured deals or research spending.

Base case: $392M.

Robotics: $7.84B × 5.00%

Company investment, not revenue or research spending. Ranges are scenarios, not confidence intervals.

Scope and related research

Analyzes observed people. Creating motion and avatars is covered under generation.