VOT2020 Dataset_Groundtruth

102 views
Skip to first unread message

Mojtaba Marvasti

unread,
Apr 12, 2020, 11:21:21 PM4/12/20
to VOT Challenge technical support
Hello Luca, 

I hope you are well. 
As you mentioned on your participation page, "Trackers may report target location as a segmentation mask or a bounding box." Also, the dataset uses segmentation ground-truth. 
Please let me know about the annotations in the groundtruth.txt for the VOT 2020 dataset, and how I should save tracker's outputs if they're based on segmentation masks.

Thank you in advance.

Regards,
Mojtaba

alan lukežič

unread,
Apr 14, 2020, 4:02:21 AM4/14/20
to VOT Challenge technical support
Hi,

The annotations in VOT20 are in the form of segmentation masks. You can see this in the groundtruth.txt since each line starts with the 'm', which denotes 'mask'. The following numbers in each line are RLE-based encoding of the binary mask. 
To see how the tracker should report the segmentation mask you can check the VOT integration repository with example trackers: https://github.com/votchallenge/integration 
More specifically, you can check python_ncc_mask to see how a tracker reports a binary segmentation mask.

Best,
Alan

Dne ponedeljek, 13. april 2020 05.21.21 UTC+2 je oseba Mojtaba Marvasti napisala:

Mojtaba Marvasti

unread,
Apr 14, 2020, 1:47:38 PM4/14/20
to VOT Challenge technical support
Thanks a lot, Alan.

Best regards,
Mojtaba
Reply all
Reply to author
Forward
0 new messages