Skip to content

SeesawFaceNets: sparse and robust face verification model for mobile platform

Notifications You must be signed in to change notification settings

pshashk/seesaw-facenet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 

Repository files navigation

PyTorch implementation of Seesaw-shuffleFaceNet

Differences from official implementation are minimal:

  • Li-ArcFace loss
  • Zero γ
  • standard SE Block (no bias, no prelu)
  • ReLU after first two convolutions
  • Hard versions for Swish and Sigmoid

Pretrained model (TorchScript)

Validation metrics can be computed by running python3 eval.py <path to model> <path to a folder containing InsightFace bin files>:

cfp_ff: 99.56%
lfw: 99.62%
agedb_30: 96.12%
vgg2_fp: 93.24%
cfp_fp: 94.49%
cplfw: 89.52%

For training code see foamliu/InsightFace-v3 and similar projects.

Tested with nightly builds of PyTorch (1.4.0.dev20191216).

About

SeesawFaceNets: sparse and robust face verification model for mobile platform

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages