Can CVAT export annotations according to the MOTS dataset (with segmentation, .png or .txt format)?

[jong372]

My actions before raising this issue

• Read/searched the docs
• Searched past issues
  

Expected Behaviour

Try to export segmentations masks according to the MOTS format (see attachment)

Context

I'm trying to build a new Multiple object tracking and segmentation dataset. For this I'm eager to find a annotation tool that can provide me the right annotations formats used in the MOTS dataset (.png or .txt). Hence, supporting temporally
consistent object instance segmentations.

[zhiltsov-max]

Hi, could you provide a link to the format description? This format is not supported, but the mask in the question is very much like the Segmentation Mask format outputs (i.e. a PASCAL VOC png mask).

[jong372]

The following link provides the description of the annotation format; MOTS. Furthermore, this link also provides possible example annotation files you can download. Thanks in advance!

[nmanovic]

@zhiltsov-max , I will transfer the issue to Datumaro repository.

[jong372]

As an addition:
-The value is a combination of category and instance id as stated at https://www.vision.rwth-aachen.de/page/mots.-Because you need to have the instance identies across frames to be labeled, which is currently not available in open source annotation tools.
Please let me know if there is an update on this issue.