FAQ

• Do I need to participate in the challenge in order to submit a paper to the workshop?

No, participation in the challenge is not a prerequisite for submitting a paper to the workshop . Researchers are encouraged to submit relevant work addressing the analysis of bias in facial attributes estimation even if they do not participate in the challenge. All relevant works will be considered.

• What is the exact use of BEFA challenge dataset?

The BEFA dataset is intended for test purposes only. The competitors should train the network independently of the test dataset

• What does the BEFA dataset contain?

The BEFA dataset is expected to contain less than 50K images, equally balanced across the following five attributes: age group, gender, ethnicity, pose and resolution. Each image will be provided with a unique identifier, url, bounding box coordinates of one face, and facial keypoints.

• What data can I train my models on?

Participants in the challenge can train their models on whichever training data they deem appropriate

• How do I evaluate the performance of my system on the BEFA dataset?

Participants in the challenge are required to submit the predictions of their systems for each image for the following three attributes: age group, gender, and ethnicity, each with an associated confidence score between zero and one. The evaluation will be performed by the organizers, not responsibility of the participants.