Codesota · Computer Vision · Handwriting Recognition · IAMTasks/Computer Vision/Handwriting Recognition
Handwriting Recognition · benchmark dataset · 1999 · EN

IAM Handwriting Database.

13,353 handwritten text lines from 657 writers. Standard handwriting benchmark.

Paper ↗Submit a result ↵
§ 01 · Leaderboard

Best published scores.

32 results indexed across 2 metrics. Shaded row marks current SOTA; ties broken by submission date.

Primary
cer · lower is better
All metrics
cer, wer
cer· primary
22 rows
#ModelOrgSubmittedPaper / codecer
01HTR-JAND—Dec 2024arxiv1.23
02GPT-4o miniOpenAIApr 2026arxiv1.71
03DRetHTR-base—Feb 2026arxiv2.26
04DTrOCR 105M—Aug 2023DTrOCR: Decoder-only Transformer for Optical Character R… · code2.38
05 Self-Attention + CTC + language model—Apr 2021Rethinking Text Line Recognition Models2.75
06TrOCR-large 558M—Sep 2021TrOCR: Transformer-based Optical Character Recognition w… · code2.89
07Transformer + CNN—Apr 2021Rethinking Text Line Recognition Models2.96
08MetaWriter—May 2025arxiv3.36
09TrOCR-base 334M—Sep 2021TrOCR: Transformer-based Optical Character Recognition w… · code3.42
10HTR-ConvTextDAIR-GroupDec 2024arxiv4.00
11TrOCR-small 62M—Sep 2021TrOCR: Transformer-based Optical Character Recognition w… · code4.22
12VAN—Dec 2020End-to-end Handwritten Paragraph Text Recognition Using … · code4.32
13Transformer w/ CNN (+synth)—May 2020Pay Attention to What You Read: Non-recurrent Handwritte…4.67
14HTR-VT(line-level)—Sep 2024HTR-VT: Handwritten Text Recognition with Vision Transfo… · code4.70
15Easter2.0—May 2022Easter2.0: Improving convolutional models for handwritte… · code6.21
16FPHR+Aug Paragraph Level (~145 dpi)—Mar 2021Full Page Handwriting Recognition via Image to Sequence … · code6.30
17Start, Follow, Read—Sep 2018papers-with-code · code6.40
18Decouple Attention Network—Dec 2019Decoupled Attention Network for Text Recognition · code6.40
19FPHR+Aug Line Level (~145 dpi)—Mar 2021Full Page Handwriting Recognition via Image to Sequence … · code6.50
20Leaky LP Cell—Feb 2019No Padding Please: Efficient Neural Handwriting Recognit… · code6.60
21FPHR Paragraph Level (~145 dpi)—Mar 2021Full Page Handwriting Recognition via Image to Sequence … · code6.70
22Transformer w/ CNN—May 2020Pay Attention to What You Read: Non-recurrent Handwritte…7.62
wer
10 rows
#ModelOrgSubmittedPaper / codewer
01GPT-4o miniOpenAIApr 2026arxiv3.34
02HTR-JAND—Dec 2024arxiv3.78
03DRetHTR-base—Feb 2026arxiv6.55
04MetaWriter—May 2025arxiv10.32
05HTR-ConvTextDAIR-GroupDec 2024arxiv12.90
06HTR-VT(line-level)—Sep 2024HTR-VT: Handwritten Text Recognition with Vision Transfo… · code14.90
07Leaky LP Cell—Feb 2019No Padding Please: Efficient Neural Handwriting Recognit… · code15.90
08VAN—Dec 2020End-to-end Handwritten Paragraph Text Recognition Using … · code16.24
09Decouple Attention Network—Dec 2019Decoupled Attention Network for Text Recognition · code19.60
10Start, Follow, Read—Sep 2018papers-with-code · code23.20
Fig 2 · Rows sorted by score within each metric. Shaded row marks SOTA. Dates reflect model or paper release where available, otherwise the date Codesota accessed the source.
§ 03 · Progress

6 steps
of state of the art.

Each row below marks a model that broke the previous record on cer. Intermediate submissions are kept in the leaderboard above; only SOTA-setting entries are re-listed here.

Lower scores win. Each subsequent entry improved upon the previous best.

SOTA line · cer
  1. Sep 1, 2018Start, Follow, Read6.40
  2. May 26, 2020Transformer w/ CNN (+synth)4.67
  3. Dec 7, 2020VAN4.32
  4. Apr 15, 2021 Self-Attention + CTC + language model2.75
  5. Aug 30, 2023DTrOCR 105M2.38
  6. Dec 24, 2024HTR-JAND1.23
Fig 3 · SOTA-setting models only. 6 entries span Sep 2018 → Dec 2024.
§ 04 · Literature

10 papers
tied to this benchmark.

Every paper below corresponds to at least one row in the leaderboard above. Click through for the arXiv preprint and, when available, the reference implementation.

§ 06 · Contribute

Have a score that beats
this table?

Submit a checkpoint and a reproduction script. We will run it, publish the score, and — if it takes the top — annotate the step on the progress chart with your name.

Submit a result ↵Read submission guide
What a submission needs
  • 01A public checkpoint or API endpoint
  • 02A reproduction script with frozen commit + seed
  • 03Declared evaluation environment (Python, deps)
  • 04One row per metric declared by this dataset
  • 05A contact so we can follow up on discrepancies