Hi, I've been testing Textract for a few days and I found a very common behaviour when reading numbers/amount. I'm in Chile so money amount format is like this: (45.120.210,00) this is the equivalent to US Format for (45,120,210.00), in Chile is very rare to use decimals. Now, when Textract reads any amount, if this amount is smaller than 7 digits then it goes well but when is bigger than 7 digits then it fails.
Example 1: https://imgur.com/AnbbhgQ
Example 2: https://imgur.com/czRRPAP
Also, i found a problem that kind of duplicates the detected object. See the image below.
Duplicated: https://imgur.com/e907k3u
In this case the amount 1.703.355.983 was a good detection, but textract added a 83 which are the last 2 digits of it, so this number went from 1.703.355.983 to 1.703.355.98383.
This is what i've seen until today, i'll keep testing because i'm pretty sure that in the near future i'll be using Textract as my OCR solution. I hope this information/feedback became useful.