Open bokey007 opened 4 years ago
align image first ^^
Is just not upside down faces but even the faces that are near 90 degree inclined. In order to align the images we need to know their orientation.
We may be able to find orientation after facial landmark localisation but we still need to detect faces before that.
The only option that I see is to retrain the model with augmented training data.
What you think ?
I think you can use face detection on both normal pictures and rotated pictures
I think you can use face detection on both normal pictures and rotated pictures
Unfortunately this does not work, here is an example with which you can try it out yourself: https://github.com/timesler/facenet-pytorch/issues/229
I aslo need to detect the faces that are upside down. But MTCNN is not able to pick them up. Any possible solution ?