You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Gerber, Mike 3eb33b01d8 Merge commit 'a8848c939b4abfca1582fb4da676a91361f88f00' 4 years ago
data@b3d65a11f5 🧹 Replace colons in file and directory names 5 years ago
LICENSE calamari-models/train-calamari-gt4histocr: Add license (Fixes qurator-spk/train-calamari-gt4histocr#1) 5 years ago
README.md Merge commit 'a8848c939b4abfca1582fb4da676a91361f88f00' 4 years ago
qurator_data_lib.sh ⬆ update */qurator_data_lib.sh 5 years ago
requirements.txt ⬆ calamari-models/train-calamari-gt4histocr: Update calamari-ocr to git version (https://github.com/Calamari-OCR/calamari/issues/146) 5 years ago
train.sh 🧹 Replace colons in file and directory names 5 years ago
upload-prepare.sh calamari-models/train-calamari-gt4histocr: Add a script to prepare upload 5 years ago

README.md

Train a GT4HistOCR Calamari model

train.sh trains a Calamari 1 model based on GT4HistOCR. Or rather 5 using cross-validation to use for confidence voting. This repository mainly serves as documentation of the provenance of the model published at https://qurator-data.de/calamari-models/, not as the definitive guide to training such a model.

Trained models

For a finished model have a look here: https://qurator-data.de/calamari-models/

Training your own model

If you really want to, you can use this script to train your own. It takes about 1 week on a Nvidia RTX 2080 GPU. Please use requirements.txt in that case to setup a virtualenv.

train.sh is able to download GT4HistOCR from the web if the data submodule is not available, that is if you're not a member of the Qurator team at SBB.