/MobileFaceNet-Grayscale

MobileFaceNets trained with grayscale images

Primary LanguagePythonApache License 2.0Apache-2.0

MobileFaceNets

PyTorch implementation of MobileFaceNets: Efficient CNNs for Accurate Real-Time Face Verification on Mobile Devices. paper.

Features

  1. Black-and-white photos for training/validation.
  2. Retinaface & similarity transform for face alignment.
  3. Lightweight: Params size (MB): 0.95, FLOPs size (GB): 0.24.

Performance

Accuracy LFW Download
paper 99.55%
ours 99.45% Link

Dataset

Introduction

Refined MS-Celeb-1M dataset for training, 5,179,510 faces over 93,431 identities. LFW datasets for testing.

Dependencies

  • Python 3.6.8
  • PyTorch 1.3.0

Usage

Data preprocess

Extract images:

$ python extract.py
$ python pre_process.py

Train

$ python train.py

To visualize the training process：

$ tensorboard --logdir=runs