I need to use Vision API to OCR sensitive documents (passeports/IDs...). But Google Cloud doesn't seem to meet GDPR requirements and I'd prefer hosting the models on our servers.
Vision OCR On-prem exists (https://cloud.google.com/vision/on-prem) but I can't find any other information on how to install or use it. Only the pricing.
Has anyone ever done this before ? Is there any public documentation ?
Any help will be greatly appreciated, thanks.
If you don't want to jump through the hoops of working with Google's sales team for the on-prem vision solution, but you don't have an ML background to train/retrain your own model, you might consider Google AutoML. This would allow you to retrain a model with minimal effort - all in their web-based console - and the results would be downloadable and able to run on various platforms such as mobile (TFLITE) or in the browser (TFJS).
You can read about Vision API's data usage here: https://cloud.google.com/vision/docs/data-usage Note that Vision API doesn't retain your data.
If you still want to use Vision API OCR onprem anyway, you would have to contact sales: https://cloud.google.com/contact
