Vision API OCR model upgrade release notes

[Tianli Yu]

We did a successful push to upgrade the Vision API's default OCR, please see the release note here: 

https://cloud.google.com/vision/docs/release-notes#May_20_2022

Please refer to this FAQ if you have any questions.

1. What if I want to use the old model after May 20’s rollout of the new model to builtin/stable (default)?

You can specify the model string “builtin/legacy” to continue to use the old model.

2. How long will the old model in “builtin/legacy” be in service?

The “builtin/legacy” model will be served for another 90 days. It will be phased out in August 2022.

3. Is there a way to still use the “builtin/legacy” model beyond 90 days?

Unfortunately, no. The “builtin/legacy” model is based on an old engine that is scheduled to be completely phased out by Google this summer. After 90 days, the “builtin/legacy” will be automatically pointed to the same model as “builtin/stable”.

4. Who should I contact if the new “builtin/stable” has regression on my data? 

Please contact your customer/support engineer to file bugs to the Cloud AI Document OCR team. If you don’t have a customer/support engineer, please post your question in this discussion group (https://groups.google.com/g/cloud-vision-discuss)

5. When will be the next update to the “builtin/stable”?

We intended to keep the same model for “builtin/stable” for at least a year after this roll out to provide a desired “freeze” of behavior. In certain rare events, there might be emergency bug fixes that are necessary which could change the model behavior. We will try our best to keep its impact minimal to our customers.