knwng / wider2pascal

A simple script for generating Pascal VOC devkit-style annotations for the WIDER faces dataset

##wider2pascal

wider2pascal is a simple matlab script for converting the publicly available WIDER faces database into a simplified Pascal VOC devkit style structure. It's a bit of a hack but it can be handy for getting up and running quickly with object detection libraries that are friendly to Pascal VOC style annotations (e.g. py-faster-rcnn).

To run the code, uncomment the file paths in demo.m.

The resulting folder structure will look like this:

WIDER_devkit/
	WIDER/
		Annotations/
			image_name1.xml
			...
		ImageSets/
			train.txt
			val.txt
		JPEGImages/
			image_name1.jpg
			....

NOTES:

• Rather than save space with symlinks, this code keeps things simple and will duplicate the entire dataset on disk in the new folder structure (Note: the dataset is a few of GB in size).
• Since annotations are only provided for the train and val partitions (the test annotations are stored on a private evaluation server), only those partitions are used in the new folder structure.
• Only the bounding box annotations are used transferred from the WIDER meta data (other attributes such occlusion level are not used)