Currently, I am using Form Recognizer version 2.1 preview to extract data from PDF file which contains scanned images. The text is fetched properly but mapping of values with labels is not as expected. I am using custom model Train with labels to extract data from the file and using FOTT website to label data which is giving the functionality to label data manually. I have trained model with manually labeled data for 5 pages which is not giving proper response. And also tried Train without labels is not giving proper result. However, the model is not giving exact result. Some labels are not fetched and some labels are having incomplete values in the result. I have read documentation from Microsoft Docs for learning about this cognitive service. It is not giving proper result as expected and I am not getting what is the issue with this. So, I want to know that what can be possible reason for not getting proper result.
When training a model are you using 5 documents from the same structure ? To improve the model you can also add a few additional documents so that your training data represents the documents you are trying to extract data from. When analyzing a document if a key value pair is missed you can add that document to the training data and train the model so that it can improve for that document also.