Open LarryDong opened 4 years ago
Hello, how did you solve this problem? Is it the result of manually adding event tracking on the second frame of the image in the .txt file?
I've had this problem too, and I'm just starting to get involved in this field, so I don't know if you say the change is what I think?
‘this code can be improved by adding the ground truth to tracks_obj from the beginning.’ How does this work? Can you see how you changed it?
‘this code can be improved by adding the ground truth to tracks_obj from the beginning.’ How does this work? Can you see how you changed it?
I forget what I was thinking, which is 4 years ago. However, the easiest way is check your tracking result and delete "extramely short" (only appeared once) feature, make this feature never appear.
For example, in my uploaded file, featue 5 only appears once (on first frame) so this feature lead to the program died. Just delete this line.
That's all I can remember.
Hi, thanks for this cool tool! But I met the problem:
It took me some time to locate this problem: the ground truth are generated by
calcOpticalFlowPyrLK
from the second frame. However, one of my features tracked by event is dead before the second frame, and all ground truth are deleted after feature died by:that leads
gt_t
to be empty and the index error.The problem can be solved by manually deleted short features (although the
filter_too_short
can filer short features but it depends on size not on time). Or this code can be improved by adding the ground truth totracks_obj
from the beginning.I hope this issue can help who have met the same problem, and help to improve this amazing tool.
(Example attached here is tracked from
outdoors_walking.bag
, this error happens for feature id 4) tracks_example.txt