Open tonybaigang opened 6 years ago
No. @davidsandberg already stated that his approach is to "just use the face bounding box and let the model generalize over different face poses." https://github.com/davidsandberg/facenet/issues/93. Take a look at https://github.com/davidsandberg/facenet/issues/329 for further discussion.
So, can I train a model using a pre-aligned dataset withou using this script? I mean, I have images only from faces and all ones are aligned... Is it possible or recommended?
Hi, Experts It seems the script align_dataset_mtcnn.py just do face detection, not did align as in align_dlib.py which use getAffineTransform to align. Why the align_dataset_mtcnn.py do not do real align?