Train (fine-tune) LayoutParser pre-trained model using custom dataset

[ammarsaf]

Motivation I want to fine-tune the pre-train model from LayoutParser by using my own dataset images. But I don't find any way in the documentation where I can train the pre-train model using my custom dataset, instead I only can train the Fast_RCNN or MaskRCNN model which means training from scratch. This might required million of dataset to get to get better performance metrics.

Is there any way I can fine-tune the model from LayoutParser? Is this even possible?

[epassaro]

I have the exact same question

[jfecunha]

Hi @Ammar-Azman and @epassaro,

You have this tutorial that explains how to fine-tune the models available on the model zoo: https://towardsdatascience.com/auto-parse-and-understand-any-document-5d72e81b0be9

The model weights link is within the config.yml under the Weights key. I hope that helps.

[epassaro]

Thank you, I already achieved that using the Detectron2 tutorial, but I will take a look anyway.