Chalearn Single-modal (RGB) Cross-ethnicity Face anti-spoofing Recognition Challenge@CVPR2020 Forum

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

> Question: Can we use pretrained models to detect landmarks or generate augmented images?

1. Are we allowed to use pre-trained DL models to detect landmarks? e.g. https://github.com/1adrianb/face-alignment
2. Can we generate augmented images from training data using pre-trained DL models? e.g. GAN

Posted by: leo_hou @ Feb. 19, 2020, 8:41 a.m.

Hi,
Q1. Are we allowed to use pre-trained DL models to detect landmarks? e.g. https://github.com/1adrianb/face-alignment
A1: CeFA dataset is provide only the face region. I think the effect of landmarks is trivial. If you want to use face alignment, it's ok.

Q2. Can we generate augmented images from training data using pre-trained DL models? e.g. GAN
A1: We allow the participants to only used the traing data of CeFA, no extra dataset (such as Imagenet, Webface, ...). Because the augmented images is from the training data of CeFA, not other datasets, it is ok.
PLEASE Note: The development subset of CeFA can not be used in trainng process.

best wishes,

Posted by: gesture_challenge @ Feb. 19, 2020, 3:49 p.m.
Post in this thread