Hello all,
Please note our updated evaluation description. Instead of two steps (face location and attribute classification), the measurement of error only has one step (attribute classification), which will be given by the mean square error between your predictions and the ground-truth. The face location was previously introduced as a sub-challenge, but it was found that the face location labelling differs too much from one rater to another to be used as ground-truth for a Jaccard index.
You can find links to the new evaluation script in the corresponding section of the track.
Our apologies for any inconvenience caused.
Thank you for participating. Good luck!
Kind Regards,
The FotW team