The 2nd Large-scale Video Object Segmentation Challenge - Track 1: Video Object Segmentation Forum

Go back to competition Back to thread list Post in this thread

> The given annotation of the test video `88b84fe107` is wrong

The tree is regarded as one leg of the giraffe. How to deal with such problem?

Thank you.

Posted by: theodoruszq @ Aug. 16, 2019, 3:12 a.m.

https://sm.ms/image/Ebso1kalI4OFd3x

Posted by: theodoruszq @ Aug. 16, 2019, 3:13 a.m.

https://sm.ms/image/hqsgKlNJTuF78fO

You could see my algorithm will ignore the un-annotated leg. So I am very confused how to deal with this.. Correct the annotation by myself? Maybe it is unfair..

Posted by: theodoruszq @ Aug. 16, 2019, 3:16 a.m.

May I ask which leg you are referring? For the left one, your results ignore the tail. For the right one, your results ignore one leg, which is annotated in the first frame.

Posted by: fyc0624 @ Aug. 16, 2019, 6:27 p.m.

https://sm.ms/image/ctIUjDsKMbeLrvE

Ok to say clearly, I annotated with bounding boxes to show out the wrong pixels. (This frame is the FIRST frame provided with annotation)

- The red box is the part of the tree, but the groundtruth thinks it is one leg.
- The blue box is the true leg of the giraffe, but the ground truth thinks it is not.

You could see it clearly from the adjacent/following frames.

Posted by: theodoruszq @ Aug. 17, 2019, 8 a.m.
Post in this thread