Kitti format or PASCAL VOC format for my new object detection dataset(15,000 images)

[Sumanth Reddy]

I'm preparing a object detection data set with 15,000 images. And our university is planning to open source the data set. I'm confused which format should I use to annotate labels( Kitti or PASCAL VOC) since DIGITS uses kitti format. I have seen that most of the standard data-set's like Imagenet, MS COCO use PASCAL VOC format(XML file). But Kitti uses different format. There are many pretty good GUI tools for annotating images in PASCAL VOC format, are there any GUI tools to annotate images in Kitti format. If so, please tell me. And can we convert one format into other.

[Luke Yeager]

The object detection data extension in DIGITS only supports the KITTI format currently. And it puts the data into a format that DetectNet expects.

But it wouldn't be hard to write another data extension/plugin that reads VOC data and/or writes the data into a different format for the LMDBs.