Multiple Object Tracking (MOT)#

Format specification#

The Multiple Object Tracking (MOT) challenge dataset provides bounding box tracking data for multiple objects within image sequences.

Supported annotation types:

Bbox (object detection)

Supported annotation attributes:

track_id (int) - Unique ID assigned to an object within a trajectory
visibility (float) - Visibility ratio of each bounding box due to occlusion by another static or moving object, or due to image border cropping
occluded (boolean) - True if visibility < occlusion_threshold, otherwise False
ignored (boolean) - True if the confidence score of bounding box is zero, otherwise False
score (float) - Confidence score of bounding box

Import MOT dataset#

You can download the MOT challenge dataset here.

A Datumaro project with the MOT challange source can be created in the following way:

datum project create
datum project import --format mot <path/to/dataset>

It is possible to specify project name and project directory. Run datum project create --help for more information.

The MOT challenge dataset directory should have the following structure:

└─ Dataset/
  ├── gt
  │   ├── gt.txt
  │   └── labels.txt
  ├── img1
  │   ├── <name_1>.<img_ext>
  │   ├── <name_2>.<img_ext>
  │   └── ...
  └── seqinfo.ini (optional)

seqinfo.ini is provided by the MOT challange dataset but it is optional in Datumaro. It includes imdir field which is the name of directory having image files. If this file is given, Datumaro will find the image files from the directory written in the imdir field.

To make sure that the selected dataset has been added to the project, you can run datum project info, which will display the project information.

Export to other formats#

Datumaro can convert the MOT challange dataset into any other format Datumaro supports.

Such conversion will only be successful if the output format can represent the type of dataset you want to convert, e.g. object detection annotations can be saved in COCO instances format, but not as COCO keypoints.

There are several ways to convert the MOT dataset to other dataset formats:

datum project create
datum project import -f mot <path/to/mot>
datum project export -f coco_instances -o <output/dir>

datum convert -if mot -i <path/to/mot> -f coco_instances -o <output/dir>

Or, using Python API:

import datumaro as dm

dataset = dm.Dataset.import_from('<path/to/dataset>', 'mot')
dataset.export('save_dir', 'coco_instances', save_media=True)

Extra options for importing to the MOT format:

occlusion_threshold determines the occluded boolean attribution of a bounding box. If visibility < occlusion_threshold, the bounding box will have occluded=True, otherwise it will have occluded=False. The default value is occlusion_threshold=0.0.

Export to MOT#

There are several ways to convert a dataset to the MOT format:

# export dataset into MOT format from existing project
datum project export -p <path/to/project> -f mot -o <output/dir> \
    -- --save-media

# converting to MOT format from other format
datum convert -if coco_instances -i <path/to/dataset> \
    -f mot -o <output/dir> -- --save-media

Extra options for exporting to the MOT format:

--save-media allow to export dataset with saving media files (by default False)
--image-ext IMAGE_EXT allow to specify image extension for exporting dataset (by default - keep original or use .png, if none)
--save-dataset-meta - allow to export dataset with saving dataset meta file (by default False)

Examples#

Examples of using this format from the code can be found in the format tests