2nd FRCSyn: Face Recognition Challenge in the Era of Synthetic Data at CVPR 2024 Forum

Go back to competition Back to thread list Post in this thread

> Setting the Threshold

Dear organizers,

The Restriction section states that 'participants cannot optimize the threshold value using the evaluation datasets'. I wonder if we can use other datasets different from the training datasets to set the thresholds for evaluation datasets. Thanks!

Best,
Carina

Posted by: carina_zhao @ Feb. 20, 2024, 4:43 a.m.

Dear Carina,
Yes, you can select your threshold using other datasets.
Best regards

Posted by: FRCSyn @ Feb. 21, 2024, 4:23 p.m.

Dear Carina,
After further reviewing your question, we have decided that you can only use the training dataset indicated for each task to select your threshold.
We consider the data used to select the threshold as part of the training dataset. Specifically, we adhere to the rules outlined for training data, for sub-tasks 1.1 and 2.1 (maximum 500,000 synthetic face images), and for 1.3, and 2.3 (the proposed CASIA-WebFace dataset with around 500,000 real face images + maximum 500,000 synthetic face images). However, in sub-tasks 1.2, and 2.2 there are no restrictions on considering additional data, as long as they are synthetic.
Sorry for the inconvenience on our first answer,
FRCSyn team.

Posted by: FRCSyn @ Feb. 26, 2024, 2:52 p.m.
Post in this thread