search

FengZhenhua / Supervised-Descent-Method

Matlab implementation of the Supervised Descent Method (SDM) for facial landmark detection and face tracking
https://sites.google.com/view/fengzhenhua
MIT License
22 stars 14 forks source link

about the result #8

Open Yifan313 opened 6 years ago

Yifan313 commented 6 years ago

I found that if the image size of the training set is different from that of the testing set the result will be not good, so how to solve this problem?

FengZhenhua commented 6 years ago

In general, facial landmark detection is performed after face detection. If you have a face detector, you can run it for both training and test images. Then you can resize them using the face bounding boxes hence the training and test images have the similar resolution. Hope this is helpful.