A facial keypoint extraction network demo, built with TensorFlow 2.
The network is a transfer learning of ResNet-50. We added an input convolution layer and several output dense layers.
This demo is trained on the Kaggle Face Images with Marked Landmark Points dataset, which predicts 15 keypoints given a grayscale facial image.
This repo is currently only a demo, and the network is designed purely for internal research purposes. Further instructions on usages, etc. might be available in the future.