Reconstructing Children’s Faces from Cropped Lip and Nose Images: A Large Mask Inpainting Approach for Privacy-Conscious Cleft Data
The pipeline for Synthetically generating whole faces of children with Cleft condition is made of two submodules, Custom Keypoint Prediction Module and GAN-Inpainting Model.
Pretrained weights for LaFIn GAN-Inpainting module can be obtained here
Young Labelled Faces in Wild (YFLW) dataset for training Inpainting model for children can be obtained at project website
iBug-300 Dataset link for training Facial keypoint model can be obtained in project website
The images of pre-op cleft images are collected from flicker and included here along with references.