Convert between visual object detection datasets
Switch branches/tags
Nothing to show
Clone or download
krosaen remove clipping of bbox values when reading in from KITTI format. The…
… max width and height were wrong, and this is not the best place to do this sort of cleaning in any case. Fixes issue #13
Latest commit 29e1691 Nov 10, 2017

Visual Object Dataset converter

Converts between object dataset formats. Requires Python 3.6.

Example: convert from data in KITTI format to Pascal VOC format:

$ python3.6 vod_converter/ --from kitti --from-path datasets/mydata-kitti --to voc --to-path datasets/mydata-voc

See for documentation on how to easily plug in additional data formats; you can define a function that can read in your data into a common format, and it will be then ready to convert to any supported format.

Similarly, you can implement a single function that takes the common format and outputs to the filesystem in your format and you will be ready to convert from e.g VOC to yours.

Currently support conversion from:


That 'train.txt' file for KITTI

When reading in KITTI, the script expects a train.txt file that isn't part of the original dataset. This is simply a file with the name of each datapoint you wish to capture. Here's an example with everything in the training set. You can also create it like so:

$ cd datasets/kitti && ls -1 training/image_2 | cut -d. -f1 > train.txt && cd -
$ head datasets/kitti/train.txt

Python2 support

This project is written using features requiring Python3.6+, but there is a fork that has been updated to work in Python2 if you need it.