Hi, we are developing an invoice OCR, so we decided to use Textract to get info from the invoices and exepenses useing the AnalyzeExpense/StartExpenseAnalysis commands. In some cases there are fields that are not detected by them. We saw that there is an option to implement Textract with A2I to train our a model. But as we can see in the API documentantion it seems to only be available with AnalyzeDocument. Is there any option to integrate with the requests in which we are interested (AnalyzeExpense/StartExpenseAnalysis) ?
AWS 공식업데이트됨 2년 전
Thanks for your reponse. We will take a look.
Hi Ioan, We have been looking to use Custom Task Types, but we are not sure how we could use our model on Textract.
For a custom A2I task type, you would first call Textract (AnalyzeExpense), then call A2I StartHumanLoop with the raw output and build a custom UI to present that to the user for adjustment. Your custom UI could be simple (for e.g. see A2I sample UIs repo) or complex (For e.g. see this sample using a SageMaker post-processing model too)