Finetuning Layout Analysis with Different Classes

[bely66]

Hi Everyone, I'm trying to finetune the layout analysis model on my dataset which is relatively small compared to the benchmarked dataset. So starting from scratch here doesn't make sense.

The issue I'm facing is I don't wanna train lists and figures for my dataset and here's the question

Would it be hard to get the pre-trained network fully except for the classification head? so that I have the knowledge without being forced to train for the same classes?

[bely66]

@tink2123 Would it be hard to do this fine-tuning?

[Tanmay98]

Hi, is there any update on this? I tried fine-training using 250 images but it is not giving good accuracy while inferencing.

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.