Format of ground-truth about VOTS2023-dev dataset

votchallenge / toolkit

The official VOT Challenge evaluation and analysis toolkit

http://www.votchallenge.net/

GNU General Public License v3.0

170 stars 46 forks source link

Format of ground-truth about VOTS2023-dev dataset #81

Closed davidyang180 closed 1 year ago

davidyang180 commented 1 year ago

Hi! as can be seen from the figure below, the shape information provided by the VOTS2023-dev dataset starting with 'm' can finally decode the region.mask of the corresponding frame. However, the ground-truth marked with 0 or 1 are decoded in the region.Special_object format, and it seems that there is no specific mask information. What is the difference between this? Does the corresponding frame have mask information? How to use it when evaluating?

lukacu commented 1 year ago

Please see #77 , it has the same question. Not all frames are annotated, performance analysis is only done on annotated frames.