Open kingdsz opened 5 years ago
Do you mean by aligning to rotate the face for example ? if it is the case, then you don't have to according to the OP, you better let the pictures as they are for better performance, so the network can learn from every position.
mtcnn doesn't generate aligned images. if you want to generate aligned face, then you can use align-dlib.py from openface
I used MTCNN, align_dataset_mtcnn.py to extract the face of the image, but found that he did not align the extracted faces.