AI-Riksarkivet / htrflow

HTRflow is the underlying engine for our HTR-pipeline

Home Page:https://ai-riksarkivet.github.io/htrflow

Repository from Github https://github.comAI-Riksarkivet/htrflowRepository from Github https://github.comAI-Riksarkivet/htrflow

Error when trying to use SATRN model

fattynoparents opened this issue · comments

What happened?

How to use the Riksarkivet/satrn_htr model with htrflow?
When trying to add it to the pipeline config and run htrflow pipeline pipeline 1.jpg I get the following error:

NotImplementedError: Model satrn_htr is not supported. The available models are: DiT, LLavaNext, TrOCR, WordLevelTrOCR, YOLO.

The same happens if I try to use the Riksarkivet/rtmdet_lines model for segmentation.

The yolo and Riksarkivet/trocr-base-handwritten-hist-swe-2 models work fine.

Steps To Reproduce

No response

Relevant log output

No response

Hi,

Thanks for opening an issue.

HTRflow uses OpenMMLab's implementation of Satrn and RTMDet. It depends on the packages mmocr, mmdet and mmcv. These are not currently included in the HTRflow installation and need to be installed separately. Install them with:

pip install mmdet==3.1.0 mmocr

For mmcv, follow OpenMMLab's instructions on how to pick a version that suits your environment: https://mmcv.readthedocs.io/en/latest/get_started/installation.html#install-with-pip

We've yet only used and tested HTRflow in a Python 3.10 environment. If you still run into issues, try switching to 3.10.

Thanks I have managed to use the Satrn model in a Python 3.10 environment after having installed the mmocr, mmdet and mmcv packages.