Benchmark of Face API on AffectNet dataset

J. Kibassa Maliba 1

Hi,

I have a question about Face API.
Do you have any documentation that gives the Face API benchmark on the AffectNet and MultiPie dataset ?
Here are the different SOTA models in the recognition of emoitions:
https://paperswithcode.com/sota/facial-expression-recognition-on-affectnet

I would like to know at which level Face API is placed.

Thank you in advance for your information.

1 answer

Ramr-msft 17,736 Reputation points

2021-06-07T13:40:27.5+00:00

@Anonymous Thanks for the question. Can you please add more details about the use case that you are trying.
Please follow the Face API latest docs and release notes for the new models that has been released. - https://learn.microsoft.com/en-us/azure/cognitive-services/face/releasenotes
Cognitive Services Face team have announced the private previewing a new Face SDK with Face Identification and Face Verification capabilities. The new Face SDK has a “hybrid” design, leveraging the best of the edge and the cloud. If you are interested in testing it out We can connect with the team for further details.

Internal model details can't be shared publicly, you can raise a support ticket to share the details securely and work with an engineer.
Please sign in to rate this answer.
J. Kibassa Maliba 1 Reputation point

2021-06-07T23:00:16.757+00:00

Thank you for your feedback.

Finally, I took the Face API subscription and tested on the accuracy of emotion detection on AffectNet datasets.
I got the score of 37% in accuracy while the SOTA model on https://paperswithcode.com/sota/facial-expression-recognition-on-affectnet are 61%.
Maybe I evaluated the model incorrectly or there are parameters that I adjusted incorrectly.
I tried to contact support but it requires a $29/month subscription to ask the question to support.
I would like to know if the 37% accuracy is really the score of the Face API emotion detection model on AffectNet or maybe I made a mistake in the evaluation?

I didn't manage to test with detection_03. I think that the emotion detection part is not yet integrated in it. I used detection_01.

Thanks
Sign in to comment

Use comments to ask for clarification, additional information, or improvements to the question.

Your answer