Optical Character Recognition2020en

icdar2015

Dataset from Papers With Code

Metrics:accuracy, cer, wer, f1
Current State of the Art

DTrOCR 105M

Unknown

93.5

accuracy

accuracy Progress Over Time

Showing 15 breakthroughs from Nov 2017 to Aug 2023

71.077.183.389.495.6Nov 2017Dec 2018Feb 2020Apr 2021Jun 2022Aug 2023accuracyDate

Key Milestones

Nov 2017
AON

From paper: AON: Towards Arbitrarily-Oriented Text Recognition

73.0
Jun 2018
ASTER

From paper: ASTER: An Attentional Scene Text Recognizer with Flexible Rectification

76.1
+4.2%
Oct 2019
SATRN

From paper: On Recognizing Texts of Arbitrary Shapes with 2D Self-Attention

79.0
+3.8%
Dec 2019
TextScanner

From paper: TextScanner: Reading Characters in Order for Robust Scene Text Recognition

79.4
+0.5%
May 2020
SEED

From paper: SEED: Semantics Enhanced Encoder-Decoder Framework for Scene Text Recognition

80.0
+0.8%
Feb 2021
CSTR

From paper: Revisiting Classification Perspective on Scene Text Recognition

81.6
+2.0%
Jun 2021
RCEED

From paper: Representation and Correlation Enhanced Encoder-Decoder Framework for Scene Text Recognition

82.2
+0.7%
Aug 2021
DPAN

From paper: Look Back Again: Dual Parallel Attention Network for Accurate and Robust Scene Text Recognition

85.5
+4.0%
Nov 2021
MATRN

From paper: Multi-modal Text Recognition Networks: Interactive Enhancements between Visual and Semantic Features

86.6
+1.3%
Dec 2021
S-GTR

From paper: Visual Semantics Allow for Textual Reasoning Better in Scene Text Recognition

87.3
+0.8%
Mar 2022
SIGA_S

From paper: Self-supervised Implicit Glyph Attention for Text Recognition

87.6
+0.3%
Sep 2022
MGP-STR

From paper: Multi-Granularity Prediction for Scene Text Recognition

90.9
+3.8%
May 2023
CLIP4STR-L (DataComp-1B)

From paper: CLIP4STR: A Simple Baseline for Scene Text Recognition with Pre-trained Vision-Language Model

91.4
+0.6%
Jul 2023
CPPD

From paper: Context Perception Parallel Decoder for Scene Text Recognition

91.7
+0.3%
Aug 2023
DTrOCR 105MCurrent SOTA

From paper: DTrOCR: Decoder-only Transformer for Optical Character Recognition

93.5
+2.0%
Total Improvement
28.1%
Time Span
5y 10m
Breakthroughs
15
Current SOTA
93.5

Top Models Performance Comparison

Top 10 models ranked by accuracy

accuracy1DTrOCR 105M93.5100.0%2CPPD91.798.1%3CLIP4STR-L (DataComp-1B)91.497.8%4MGP-STR90.997.2%5CLIP4STR-L90.897.1%6CLIP4STR-B*90.696.9%7SIGA_S87.693.7%8S-GTR87.393.4%9MATRN86.692.6%10CDistNet (Ours)86.392.2%0%25%50%75%100%% of best
Best Score
93.5
Top Model
DTrOCR 105M
Models Compared
10
Score Range
7.3

accuracyPrimary

#ModelScorePaper / CodeDate
1
DTrOCR 105M
93.5Aug 2023
2
CPPD
91.7Jul 2023
3
CLIP4STR-L (DataComp-1B)
91.4May 2023
4
MGP-STR
90.9Sep 2022
5
CLIP4STR-L
90.8May 2023
6
CLIP4STR-B*
90.6May 2023
7
SIGA_S
87.6Mar 2022
8
S-GTR
87.3Dec 2021
9
MATRN
86.6Nov 2021
10
CDistNet (Ours)
86.25Nov 2021
11
DiffusionSTR
86Jun 2023
12
DPAN
85.5
Look Back Again: Dual Parallel Attention Network for Accurate and Robust Scene Text RecognitionCode
Aug 2021
13
RCEED
82.2Jun 2021
14
CSTR
81.6Feb 2021
15
Yet Another Text Recognizer
80.2Jul 2021
16
SEED
80May 2020
17
TextScanner
79.4Dec 2019
18
SATRN
79Oct 2019
19
SAFL
77.5Jan 2022
20
ASTER
76.1
ASTER: An Attentional Scene Text Recognizer with Flexible RectificationCode
Jun 2018
21
DAN
74.5Dec 2019
22
AON
73Nov 2017
23
ViTSTR
72.6May 2021
24
Baek et al.
71.8Apr 2019
25
SAR
69.2Nov 2018

f-measure

Related Papers22

Other Optical Character Recognition Datasets

icdar2015 Benchmark - Optical Character Recognition | CodeSOTA