SSD is an unified framework for object detection with a single network.
we use SSD to face detect and locate base on widerface dataset.
This is a modification on original SSD which is based on mxnet. The official repository is available here. The arXiv paper is available here.
Evaluate model base on FDDB
# cd /path/to/face-detection-ssd-mxnet
sh scripts/test.sh
This example only covers training on Pascal VOC dataset. Other datasets should
be easily supported by adding subclass derived from class Imdb in dataset/imdb.py.
See example of dataset/pascal_voc.py for details.
- Download the converted pretrained
vgg16_reducedmodel here, unzip.paramand.jsonfiles intomodel/directory by default. - Download the widerface dataset into
data/, - modify
tools/widerfce2VOC.pyto your own parameters, and run to get the annotations. - modify
tools/prepare_widerface_pascal.shto your own parameters, run to get the .rec files for train. - run
srcips/train.sh