From here https://cloud.google.com/vision/docs/release-notes it says that there was an upgrade on OCR model for TEXT_DETECTION and DOCUMENT_TEXT_DETECTION. What is the recent model improvement compare to legacy model (is there some metrics used)?
As part of the best effort, can I make a suggestion for you on modifying the model field of Feature object into either “builtin/legacy” or “builtin/latest”. Though the OCR model will automatically be upgraded to the new model when there’s an update on “builtin/stable” (default). This means that google implemented the newest versions of TEXT_DETECTION and DOCUMENT_TEXT_DETECTION to work without a problem for future updates