Some question about "3.3. Testing Scheme"

[9p15p]

Dear Sir:
In "3.3. Testing Scheme",There is a saying that "we propagate the object mask until we reach a frame in which user annotations were given in any previous rounds."
I don't understand why the weight should get inverse when we reach the frame where our annotions were given before, just like the graph in Fig.4.
Are there some rofound meaning behind it？Or maybe I have a misunderstanding？
Looking forward to your reply！Thank you!

[seoungwugoh]

Hi 9p15p,
In this scheme, we linearly blend two results that propagated from solid and dashed arrow (current and the previous points of user interaction).
And the weights in this figure are for blending.
In general, the longer the propagation, the lower the quality.
And that is why the weights are getting lower with propagated distance.
We also tested with the constant weight (=0.5) for blending, the dynamic weights (current scheme) works better.

[9p15p]

Thank you^_^！I benefited a lot from your answer.
And I still have a question:
what does "the frame in which user annotation were given in any previous rounds" mean?
Does it mean that we get a bad result? Because we have given the annotation before , but the model make the same mistakes twice.（in my opinion, annotation is for correcting mistakes）
And what should we do after we reach this frame ?
Should we still continue to propagate the object mask until we don't make the same mistake again(i.e. the same annotation don't appear again) in a certain number of rounds?

best wishes!

[seoungwugoh]

It means previous round's results where user annotations were given (or frames close to them) will be also reliable.
Yes, annotations are for correcting mistakes but should be effective locally.
It is too risky to assume new annotations will be always helpful after a large number of propagation. And this local update is also computationally more friendly (as we don't need to estimate frames where weights are zero).

[9p15p]

ahh i got it. Thank you very much !^_^