MegaFace-mask about facex-zoo HOT 2 OPEN

About question 1, I think it is unnecessary to measure how the recognition quality degrades on the original MegaFace dataset for the same models, since the experiments illustrated here are used to demonstrate that for face recognition with masks, original datasets needs to change. So I think the question you actually want to ask is, whether the models trained on MegaFace-mask can be used to recognize the normal faces.
@wang21jun here. 😄
I guess the model trained on MegaFace-mask would perform worse on MegaFace, since the data distribution is quite different.

About question 2, I think it would be very easy to create MegaFace-mask dataset by using the files under /FaceX-Zoo/addition_module/face_mask_adding/FMA-3D/ .

Since you've got the face bounding box, then you can use the face alignment tools to get the corresponding 106 facial landmark detection results, after that, running add_mask_all.py or add_mask_one.py would both be OK.

from facex-zoo.

Whether the models trained on MegaFace-mask can be used to recognize the normal faces.

I'm curious about this. Have anyone done some experiments for this? Does the model trained on Megaface-mask can recognize faces as well as the model trained on MegaFace?

from facex-zoo.