search

wy1iu / sphereface

Implementation for <SphereFace: Deep Hypersphere Embedding for Face Recognition> in CVPR'17.
MIT License
1.58k stars 543 forks source link

How to do 3-patch ensemble? #66

Open hujun100 opened 6 years ago

hujun100 commented 6 years ago

Hi, Can you tell us the way to do 3-path ensemble? Thanks

wy1iu commented 6 years ago

Please see Appendix H in SphereFace paper (http://wyliu.com/papers/LiuCVPR17v3.pdf).

ghost commented 6 years ago

@wy1iu Are these 3-patch ensembles have the same network architecture, and different input as a different alignment applied to each input. Or each ensemble has different architecture.

When should we apply different alignment on the input at training, or just in inference, or both of them?

wy1iu commented 6 years ago

the network architectures are the same. The only thing that is different is the cropped patch.