.. _sec_automm_detection_voc_to_coco:

AutoMM Detection - Convert VOC Format Dataset to COCO Format
============================================================


`Pascal VOC <http://host.robots.ox.ac.uk/pascal/VOC/>`__ is a collection
of datasets for object detection. And VOC format refers to the specific
format (in ``.xml`` file) the Pascal VOC dataset is using.

In this tutorial, we will convert VOC2007 dataset from VOC format to
COCO format. See :ref:`sec_automm_detection_prepare_voc` for how to
download it. We will use our tool ``voc2coco``. This python script is in
our code:
`voc2coco.py <https://raw.githubusercontent.com/autogluon/autogluon/master/multimodal/src/autogluon/multimodal/cli/voc2coco.py>`__,
and you can also run it as a cli:
``python3 -m autogluon.multimodal.cli.voc2coco``.

**Note: In Autogluon MultiModalPredictor, we strongly recommend using
COCO as your data format.** However, for fast proof testing we also have
limit support for VOC format.

Convert Existing Splits
~~~~~~~~~~~~~~~~~~~~~~~

Under VOC format root path, we have the following folders:

::

   Annotations  ImageSets  JPEGImages

And normally there are some pre-defined split files under
``ImageSets/Main/``:

::

   train.txt
   val.txt
   test.txt
   ...

We can convert those splits into COCO format by simply running given the
root directory, e.g. \ ``./VOCdevkit/VOC2007``:

::

   python3 -m autogluon.multimodal.cli.voc2coco --root_dir ./VOCdevkit/VOC2007

The command line output will show the progress:

::

   Start converting !
    17%|█████████████████▍                                                                                  | 841/4952 [00:00<00:00, 15571.88it/s

Now those splits are converted to COCO format in ``Annotations`` folder
under the root directory:

::

   train_cocoformat.json
   val_cocoformat.json
   test_cocoformat.json
   ...

Convert Existing Splits
~~~~~~~~~~~~~~~~~~~~~~~

Instead of using predefined splits, you can also split the data with the
train/validation/test ratio you want. Note that this does not require
any pre-existing split files. To split train/validation/test by
0.6/0.2/0.2, run:

::

   python3 -m autogluon.multimodal.cli.voc2coco --root_dir ./VOCdevkit/VOC2007 --train_ratio 0.6 --val_ratio 0.2

The command line output will show the progress:

::

   Start converting !
    17%|█████████████████▍                                                                                  | 841/4952 [00:00<00:00, 15571.88it/s

And this will generate user splited COCO format in ``Annotations``
folder under the root directory:

::

   usersplit_train_cocoformat.json
   usersplit_val_cocoformat.json
   usersplit_test_cocoformat.json

Other Examples
~~~~~~~~~~~~~~

You may go to `AutoMM
Examples <https://github.com/autogluon/autogluon/tree/master/examples/automm>`__
to explore other examples about AutoMM.

Customization
~~~~~~~~~~~~~

To learn how to customize AutoMM, please refer to
:ref:`sec_automm_customization`.