landmark detection in all_in_one folder

[choice17]

Thank you for your project, I can test image using the pretrained all_in_one models.
My question is that since the prepare data does not cover landmark training, may I know where is the pretrained weights come from?
I would like to know if the pretrained network is your implementation result. As It seems great with the landmark location.