got-10k / toolkit

Official Python toolkit for generic object tracking benchmark GOT-10k and beyond
http://got-10k.aitestunion.com/
MIT License
555 stars 95 forks source link

cover.label file is inaccurate in some videos. #20

Open sherylwang opened 5 years ago

sherylwang commented 5 years ago

Hi, Really thanks for providing this dataset. During analysis of the bounding boxes in the videos, I find the visible levels in 'cover.label' may have problems in some videos like 'GOT-10k_Train_000385' and 'GOT-10k_Train_000602'. They are all 0 in the files. Please correct me if I have wrong understanding!

jhultman commented 4 years ago

I am also wondering about this. Would love some clarification about the correspondence between cover.label (0, 1, 2, 3, 4, 5, 6, 7, 8) and the visible ratio thresholds in Figure 5(e) of the paper.

temp

Is this table correct?

cover.label Visible ratio range
0 0.00
1 0.00 - 0.15
2 0.15 - 0.30
3 0.30 - 0.45
4 0.45 - 0.60
5 0.60 - 0.75
6 0.75 - 0.90
7 0.90 - 1.00
8 1.00