facedetection_ssd_train

Caffe SSD Face Detector Training Script

Requires

Caffe-SSD: caffe-ssd you can find at [here])

WiderFace Dataset: you need to download wider_face ,and make sure you place the dataset at correct folder,

├── WIDER
      ├── wider_face_split
             
      ├── WIDER_train
             
      └── WIDER_val     

Before you start to train, you need to modify some paths in Makefile

wider_dir: the path point to WIDER, e.g. /pathxxx/WIDER

caffe_dir: the path point to caffe-ssd directory, e.g. /pathxxx/caffe-ssd

Train

Make wider face lmdb data

make wider_lmdb

Train face detector

make train

Test

Test on images

python scripts/test_on_examples.py model/yufacedetectnet-open-v2.prototxt model/ssdfacedet_iter_9000.caffemodel image/

note: yufacedetectnet-open-v2.prototxt is design for detecting 35x35 min_face_size in 1080P

Ref

There are two respositories this project reference to

model: libfacedetection

script: MobilenetSSDFace