Data Augmentation

Open mayankuxtc opened this issue 6 years ago • 0 comments

Hello everyone,

I am using 300 W-LP dataset containing 61225 images. I am getting good results on predicting 68 landmarks. But the model predicts open eyes even when the eyes are closed. I want to do the data augmentation as suggested in the paper to increase the dataset. Can someone help me, with the data augmentation? My concern is: If i rotate, translate, or rescale the target face, will the 3DMM parameters and ground truth landmark locations also change? If yes, how can I encounter this issue? Hope to hear from someone. Help is appreciated.

Thanks, Mayank [email protected]

Aug 08 '19 20:08 mayankuxtc