Scene and Document Text Detection and Recognition (OCR)

Finding text regions in images and reading the characters, from street signs to scanned pages.

Tier Grades

Tier A: Frontier AI labs

feasible

Baidu and big labs lead OCR systems.

Tier B: Indonesian AI labs

feasible

Models of 1.5M to 34.5M parameters are easy to retrain on Indonesian scripts and fonts.

Tier C: Institution units such as ITB

feasible

University units publish detectors and recognisers often.

Tier D: Organisations such as SEMAI (PoT, PoC, PoV)

feasible

Small OCR models fit eKYC and form pilots on modest hardware.

Where this comes from

Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.

Anchors

ComputePP-OCRv6 spans 1.5M to 34.5M parameters per its title; training compute is not stated in the abstract.
DataIn-house benchmarks: PP-OCRv6_medium reaches 83.2% recognition accuracy and 86.2% detection Hmean.
Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 16 authors on the PP-OCRv6 paper (arXiv 2606.13108).
Time horizonMonths to a year.
Shelf-life signalsDocument VLMs overtake pipeline OCR, Small OCR models match billion-scale VLMs
Pipeline roleTier A releases OCR systems; Tier B and Tier C add Indonesian text data; Tier D builds document capture pilots.

Where this comes from

Team size is a proxy (author counts of exemplar papers), not a headcount.

Who and What

Exemplars

Observed actors

Huazhong University of Science and Technology Jaided AI

Jaided AI

Microsoft AI

Linked models

PP-OCRv6OCR and documentexemplar

PP-OCROCR and documentexemplar

TrOCROCR and documentexemplar

EasyOCROCR and documentexemplar

DBNetOCR and documentexemplar

Nemotron OCR v2OCR and documentproduced by

PP-OCRv5 serverOCR and documentproduced by

PP-OCRv5 mobileOCR and documentproduced by

MGP-STROCR and documentproduced by

PARSeqOCR and documentproduced by

SVTROCR and documentproduced by

oCLIPOCR and documentproduced by

SPTSOCR and documentproduced by

TrOCR small handwrittenOCR and documentproduced by

TrOCR large printedOCR and documentproduced by

FCENetOCR and documentproduced by

University of Science and Technology of China (USTC)

ABINetOCR and documentproduced by

RobustScannerOCR and documentproduced by

University of Science and Technology Beijing

DRRGOCR and documentproduced by

ABCNetOCR and documentproduced by

SATRNOCR and documentproduced by

MASTEROCR and documentproduced by

PANet (text)OCR and documentproduced by

CRAFT text detectorOCR and documentproduced by

+17

PSENetOCR and documentproduced by

SAR (Show, Attend and Read)OCR and documentproduced by

Intel OCR models (OpenVINO)OCR and documentproduced by

TextSnakeOCR and documentproduced by

Institute of Automation, Chinese Academy of Sciences

NRTROCR and documentproduced by

EAST text detectorOCR and documentproduced by

Attention OCROCR and documentproduced by

CRNNOCR and documentproduced by

Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences

CTPNOCR and documentproduced by ASTEROCR and documentproduced by docTROCR and documentproduced by

DAMO OCR models (ModelScope)OCR and documentproduced by

Unified Detector (text and layout)OCR and documentproduced by

PDF-Extract-Kit 1.0OCR and documentproduced by

Surya and MarkerOCR and documentproduced by Text recognition ResNet-FC and ViTSTROCR and documentproduced by

OCDNet and OCRNet (TAO)OCR and documentproduced by

Links

R&D TopicScene and Document Text Detection and Recognition (OCR)

Models

DBNetHuazhong University of Science and Technology, OCR and document

EasyOCRJaided AI, OCR and document

PP-OCRBaidu, OCR and document

PP-OCRv6Baidu, OCR and document+37 more

Use Casesnone recorded

Who Runs Itnone recorded

Graph

Models 41

PP-OCRBaidu, OCR and documentexemplar

oCLIPNanyang Technological University, ByteDance, OCR and documentproduced by

PP-OCRv5 mobileBaidu, OCR and documentproduced by

PP-OCRv5 serverBaidu, OCR and documentproduced by

PP-OCRv6Baidu, OCR and documentexemplar

TrOCRMicrosoft AI, OCR and documentexemplar

ABCNetSouth China University of Technology, University of Adelaide, OCR and documentproduced by

FCENetSouth China University of Technology, SenseTime, OCR and documentproduced by

PDF-Extract-Kit 1.0OpenDataLab, OCR and documentproduced by

Surya and MarkerDatalab, OCR and documentproduced by

SVTRBeijing Jiaotong University, Baidu, OCR and documentproduced by

TextSnakePeking University, Megvii, OCR and documentproduced by

+29

TrOCR large printedMicrosoft AI, OCR and documentproduced by

TrOCR small handwrittenMicrosoft AI, OCR and documentproduced by

Unified Detector (text and layout)Google, OCR and documentproduced by

ABINetUniversity of Science and Technology of China (USTC), OCR and documentproduced by

Attention OCRGoogle, OCR and documentproduced by

CRAFT text detectorNaver, OCR and documentproduced by

CRNNHuazhong University of Science and Technology, OCR and documentproduced by

CTPNShenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, OCR and documentproduced by

DAMO OCR models (ModelScope)Alibaba DAMO, OCR and documentproduced by

DBNetHuazhong University of Science and Technology, OCR and documentexemplar

DRRGUniversity of Science and Technology Beijing, OCR and documentproduced by

EAST text detectorMegvii, OCR and documentproduced by

EasyOCRJaided AI, OCR and documentexemplar

Intel OCR models (OpenVINO)Intel, OCR and documentproduced by

MASTERPing An Insurance, OCR and documentproduced by

MGP-STRAlibaba DAMO, OCR and documentproduced by

Nemotron OCR v2NVIDIA, OCR and documentproduced by

NRTRInstitute of Automation, Chinese Academy of Sciences, OCR and documentproduced by

OCDNet and OCRNet (TAO)NVIDIA, OCR and documentproduced by

PANet (text)Nanjing University, OCR and documentproduced by

PARSeqUniversity of the Philippines Diliman, OCR and documentproduced by

PSENetNanjing University, OCR and documentproduced by

RobustScannerSenseTime, OCR and documentproduced by

SAR (Show, Attend and Read)University of Adelaide, OCR and documentproduced by

SATRNNaver, OCR and documentproduced by

SPTSSouth China University of Technology, OCR and documentproduced by ASTEROCR and documentproduced by docTROCR and documentproduced by Text recognition ResNet-FC and ViTSTROCR and documentproduced by

Papers

An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text RecognitionBaoguang Shi et al., arXiv preprint, 2015 founding
Scene Text Recognition with Permuted Autoregressive Sequence ModelsDarwin Bautista et al., ECCV 2022 (per arXiv comment), 2022 state of the art
TrOCR: Transformer-based Optical Character Recognition with Pre-trained ModelsMinghao Li et al., arXiv preprint, 2021 state of the art
Character Region Awareness for Text DetectionYoungmin Baek et al., CVPR 2019 (per arXiv comment), 2019 state of the art
Real-time Scene Text Detection with Differentiable BinarizationMinghui Liao et al., AAAI 2020 (per arXiv comment), 2019 state of the art

Where this comes from

The papers to read first: where the direction began, its current best, and a survey when one exists.

Sources

arxiv.org/abs/2606.13108 https://arxiv.org/abs/2606.13108

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Scene and Document Text Detection and Recognition (OCR)", https://aurion.semai.id/topics/scene-text-detection-and-recognition/, records of October 2026.