Training a Custom Model without using labeling tool

Question

I have this application that uses form recognizer .net sdk to validate a number of different documents. But now i want to add to a composed model another custom Model. I searched and found out that in order to train a custom model while using the training labels i needed the files <filename\>.pdf.labels.json in my blobcontainer. My question is how can i generate this files without using the labelling tool?

Answer 1

@Diogo Silva (1180611) It is possible to create a labels.json file and use it as part of your training without using the labeling tool but to get the file format that would work for your corresponding trainings you need to use the tool to run OCR on your files and generate labels from the tool that will create files(ocr & label files) in your blob container. Before you hit the training button you can copy the ocr.json and labels.json files and then edit them for re-use for subsequent trainings. But creating the label files for newer documents need the bounding box co-ordinates to be updated correctly using the tool or your own application that can create such files.

Here is an example of one such user who was able to follow this method to train models by creating their own label files.

There is not much detail about the same in the documentation but the guidance is to use the tool to generate these files even if you want to manually label the files later for training. I hope this helps!!

If an answer is helpful, please click on or upvote which might help other community members reading this thread.