Project AURIONAI Use-Cases Research & Innovation
Microsoft AI
OCR and documentfamily

TrOCR

Microsoft transformer OCR that pairs a pretrained image encoder with a pretrained text decoder for text-line recognition.

Microsoft AIMicrosoft AIUSA
Figure from the TrOCR paper: The architecture of TrOCR, where an encoder-decoder model is designed with a pre-trained image Transformer as the encoder.
Levelfamily
ClassOCR and document
Parametersnot in the notes
Contextnot in the notes
Released21 Sep 2021
Opennessopen weights permissive
Runs onmobile or NPU
Statusactive

Lineage

Microsoft AITrOCR large printed belongs to this familyOCR and document21 Sep 2021608.12M
Microsoft AITrOCR small handwritten belongs to this familyOCR and document21 Sep 2021
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. Confidence and evidence are the notes' own.

R&D Topics

Scene and document text detection and recognition (OCR)exemplar

Identity

Developer
Microsoft AI
Family
not in the notes
Released
21 Sep 2021 sourcesannouncement page
Weights published
2 Mar 2022
Record checked
12 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
dense transformer
Total parameters
– sources
Active parameters
–
Checkpoint parameters
–
Note
Small, base and large for printed and handwritten text, plus stage-1 pretrained weights.
Context length
not in the notes
Input
image
Output
text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
open weights permissive
Licence
MIT License sources
Access routes
open weights download, API
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
mobile or NPU
Footprint
Edge or toolkit builds listed in: Qualcomm AI Hub.
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
not stated
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Benchmark Results

The notes record no benchmark result for this model.

Industry Use Cases

No industry use case links this model.

Sources

  1. arxiv.org/abs/2109.10282 https://arxiv.org/abs/2109.10282
  2. huggingface.co/microsoft/trocr-base-handwritten https://huggingface.co/microsoft/trocr-base-handwritten
  3. aihub.qualcomm.com/models/trocr https://aihub.qualcomm.com/models/trocr
  4. huggingface.co/microsoft/trocr-small-handwritten https://huggingface.co/microsoft/trocr-small-handwritten
  5. huggingface.co/api/models?pipeline_tag=image-to-text&sort=do... https://huggingface.co/api/models?pipeline_tag=image-to-text&sort=downloads&limit=50
  6. docs.roboflow.com/models/supported-models/trocr.md https://docs.roboflow.com/models/supported-models/trocr.md
  7. paperswithcode.com/method/trocr https://paperswithcode.com/method/trocr

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "TrOCR", https://aurion.semai.id/models/microsoft-trocr/, records of October 2026.