Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Correct Video Reward for KIS #496

Open
klschoef opened this issue Sep 19, 2024 · 4 comments
Open

Correct Video Reward for KIS #496

klschoef opened this issue Sep 19, 2024 · 4 comments
Labels
enhancement New feature or request

Comments

@klschoef
Copy link

klschoef commented Sep 19, 2024

For KIS tasks it would be good to reward finding the correct video, because often a video contains similar or even equal scenes (for example in news videos, where there is a preview of a scene). Such an example is given in V3C video 05579, which contains an almost identical scene at 02:52 and 16:23 (happened at IVR4B2024 at Task v7).

Screenshot 2024-09-19 at 10 38 39 Screenshot 2024-09-19 at 10 38 24

One possibility would be to give half the points for finding a scene in the correct video (e.g., 500 - ; instead of 1000 - ).

@klschoef klschoef added the enhancement New feature or request label Sep 19, 2024
@lucaro
Copy link
Collaborator

lucaro commented Sep 19, 2024

Adding further scoring functions is not a problem. What would you propose the actual function to look like @klschoef ?

@klschoef
Copy link
Author

As far as I know the KIS scoring currently is a follows:
int(max(0, r^i(t) - p^i(ws) ))
r^i(t) = (1000 - s_{min}) * ((t_{max} - t) / t_{max} + s_{min})
p^i(ws) = ws * 100

We could modify it for KIS tasks so that for correct videos, r could be used like this (thanks to Omar for this suggestion):
r^i(t) = (500 - s_{min}) * ((t_{max} - t) / t_{max} + s_{min})

If the correct segment in the video is found later, the original function could be used (and maybe the previously submitted segment in the same video could be considered as a wrong submission).

But I think this should be also discussed with the VBS teams.

@lucaro
Copy link
Collaborator

lucaro commented Sep 24, 2024

This would not consider how far away a submitted solution is from the target. Is that what we want, or would we want to consider that distance?

@klschoef
Copy link
Author

Yes, I think this is what we want (a reward for finding the correct video).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request
Projects
None yet
Development

No branches or pull requests

2 participants