how to create the dataset for training??

[sumith5]

Hi Guys,
Can you guide me for the custom dataset preparation like which annotation tool need to be used to get the required format for training the text detection model.
Thanks in advance for help

[felixdittrich92]

Hi @sumith5 👋,

Currently we don't have a custom labeling tool for docTR so you can use any available of your choice (for example: LabelStudio and afterwards convert the labels to the required format:

docs

Reference datasets can be found here

Additional a great blog post about fine tuning docTR models: blog post

Hope this helps

Best,
Felix :)

[sumith5]

Hello @felixdittrich92,

Thank you for your prompt response.

I am working on building a custom OCR model for a new dataset. The dataset is annotated in COCO format, with a single class labeled as "text." My objective is to perform OCR on these images.
can you guide me for the conversion of the annotations in the required format and also please feel free to share your opinion

Please find below a sample image from my dataset for reference.