深度学习>> sfd.gluoncv>> 返回
项目作者： yangfly

项目描述 ：
Reproduce SFD face detector using gluon-cv
高级语言： Python
项目地址: git://github.com/yangfly/sfd.gluoncv.git
创建时间： 2018-10-19T01:43:31Z
项目社区：https://github.com/yangfly/sfd.gluoncv
开源协议：Apache License 2.0
下载


Description

This work is to reproduce S³FD, a real-time Single Shot Scale-invariant Face Detector.

Environment

Preparation

  1. # install MobulaOP following: https://github.com/wkcn/MobulaOP
  2. $$ git clone https://github.com/yangfly/sfd.gluoncv.git
  3. $$ cd sfd.gluoncv

Training your own model

  1. download wider face dataset into widerface/downloads/
    1.  $$ tree widerface/downloads
    2.  widerface/downloads
    3.  ├── eval_tools.zip
    4.  ├── Submission_example.zip
    5.  ├── wider_face_split.zip
    6.  ├── WIDER_test.zip
    7.  ├── WIDER_train.zip
    8.  └── WIDER_val.zip
  2. Parpare data: unzip data, annotations and eval_tools
    1.  $$ python tool/parpare.py
    2.  $$ tree widerface -L 1
    3.  widerface
    4.  ├── downloads
    5.  ├── eval_tools
    6.  ├── wider_face_split
    7.  ├── WIDER_train
    8.  └── WIDER_val
  3. Prepare custom val dataset for quick validation (crop and resize to 640)
    1.  $$ python tool/build_custom_val.py
    2.  $$ tree widerface -L 1
    3.  widerface
    4.  ├── downloads
    5.  ├── eval_tools
    6.  ├── WIDER_custom
    7.  ├── wider_face_split
    8.  ├── WIDER_train
    9.  └── WIDER_val
  4. train vgg16 based sfd with 4 gpus
    1.  $$ python sfd/train.py
    more supported base models in sfd/nn/sfd.py
  5. demo
    1.  $$ python sfd/demo.py --model models/vgg16/sfd_best.params
  6. eval on WIDER_val
    1.  $$ python sfd/eval.py --model models/vgg16/sfd_best.params