Facial Recognition via Azure Face API no longer available due to MSFT 'Responsible AI' policy
From Reddit: I'm running with docker, using the latest image. I've created an account on Azure and set up a face API. I get the endpoint and the API key and put them into the image recognition settings:
endpoint: https://xxxxxxxx.cognitiveservices.azure.com/
API key: 32d7a7089c394d7aaf0b112a2e123456 (obviously this is not real)
The logs only show one thing related to this failure:
[17:40:26.765-.NET ThreadPool Worker-ERR] Unable to start Azure service: Operation returned an invalid status code 'Forbidden'
If I change the API key to something bogus, I at least see that the message changes:
[17:40:13.751-.NET ThreadPool Worker-ERR] Unable to start Azure service: Operation returned an invalid status code 'Unauthorized'
So I know the login is working, it's just something else with the request that is forbidden. Any ideas on moving forward with this?
Quick bit of googling indicates that it may be this: https://stackoverflow.com/questions/73095624/unable-to-use-azure-face-api-forbidden-error
I didn't think I was using any of the gender/facial hair/etc properties, but I'll check and if I am, I'll remove them in a future release. It's probably not affecting me because I had my account for a while.
Removed the attributes that MSFT supports - you can try pulling webreaper/damselfly:dev and see if it solves the problem once the build finishes in about 15 mins.
Update: the problem appears to be the new policy for Azure Face, which means that only fully vetted Microsoft customers can use the face-recognition features within Azure.
https://azure.microsoft.com/en-us/blog/responsible-ai-investments-and-safeguards-for-facial-recognition/?WT.mc_id=AI-MVP-5003365
I have contacted Microsoft to see whether there's anything that can be done to resolve this issue, such as getting Damselfly vetted and approved for use somehow.
This also means that Facial Recognition via the Azure API will break for all existing users at the end of January 2023. :(
Closing this, as I've implemented a new face recognition solution that doesn't require Azure.