Overview of ATR/HTR systems

https://shorturl.at/vbvkW

https://leaflet.pub/8a9f8f63-8c33-4f6b-926c-c314417a0337

🌐 URL: https://transkribus.ai/

💶 Costs: 🆓

🗒️ Note: Use of large AI models by Transkribus. Suitable only for single pages.

🌐 URL: https://app.transkribus.org/models/public

💶 Costs: 🆓

🗒️ Note: Possibility to test any public model for free for single pages. Very limited export possibilities.

Example - Carolingian Minuscule Model

🌐 URL: https://app.transkribus.org/models/public/text/51210

🌐 URL: https://rescribe.xyz/

💶 Costs: 🆓

🌐 URL: https://pero.fit.vutbr.cz/

💶 Costs: 🆓

⚠️ No possibility of training custom models.⚠️

🌐 URL: https://www.transkribus.org/

💶 Costs: 50 credits/month for free, subscription model

Pros:

• very large community
• easy to sign-up and use
• a lot of public models
• part of an ecosystem
  1. ScanTent, Transkribus, Transkribus Sites

Cons:

• advanced features (field training, advanced export options. etc.) require subscription
• some solutions could be cheaper
• not open-source
• not possible to import/export models

🌐 URL: https://escriptorium.inria.fr/

💶 Costs: 🆓 but you need your own infrastructure

Pros:

• large community
• open-source
• suitable for non-Latin scripts
• possibility to import/export Kraken models (to be found on HTR-United and Zenodo)
• privacy of the data

Cons:

• need for own servers/infrastructure
• less user friendly (a new version should appear in 2025)

🌐 URL: https://www.ocr4all.org/

💶 Costs: 🆓 but you need your own infrastructure

Source of the image: https://www.ocr4all.org/about/ocr4all

🌐 URL: https://vision.calfa.fr/

💶 Costs: 3500 Euro per 3500 pages, following pages cheaper

🎯 Focus: non-Latin-alphabet based scripts

🌐 URL: https://scribblesense.cz/

💶 Costs: Currently 🆓

From the same developers as Projekt Pero.

🌐 URL: https://kraken.re/main/index.html

💶 Costs: 🆓 but you need your own infrastructure.

🗒️ Note: Used by eScriptorium.