Feedback / Discussion: correction tab
After discussing with Deniz, it seems like a good idea to open a discussion issue to collect feedback on the newly added "correction" tab in Petrack. So I will start here with my first comments on that and hope to inspire some others to join in this discussion!
I used this version of Petrack: Version: 0.10.0 Commit id: acb0f219 Commit date: Tue Jul 4 09:34:00 2023 Build from branch: HEAD Compile date: Fri Mar 24 01:03:10 2023 Build with: clang++ (14.0.3)
Velocity: In my opinion the current form of this test does not help during correction. It would be necessary to adapt the threshold, as the current value is (at least for the data I tested) far to low. I get tons of failed checks even though the trajectories look fine. If one could set the threshold higher (more distance allowed between tracked points) this feature could be used to detect tracking errors that are sometimes difficult to see. E.g. if one single point is strangely moved to the opposite side of the tracking area.
Length: This one is great! One can directly delete all the trajectories that are e.g. only present in one frame. That would have saved me quite some time during my corrections.
Inside: I think the idea of this test is good. But currently I am not sure if it does exactly what I think it is supposed to do. As far as I understood, it is checking whether the starting point of a trajectory is inside the recognition area (between the green lines). I expected the box "margin [px]" to define a region from the green lines towards the center (with the width of the value in "margin") in which the trajectory is allowed to start. I tried to visualise this in the screen shot attatched (I expected the starting points to now be allowed between the green and the yellow areas.) That would mean that all persons visible in the screenshot would be marked as "correct", while the person who's trajectory is visible only inside the bottleneck (red path) should be found as errorous. This however does not happen. Do I misinterpret this test or the parameters or is there still something strange happening? I think it would also help if the margin would be visible in the video as at least I am having some difficulties with choosing the values in pixels.