Hi,
Good question.
The key point is that the option refers to the number of labeling iterations, not to the total amount of pixel-wise work.
Pixels with strength < T_low are rejected immediately, so they are not candidates for labeling. Therefore, their number may affect how many pixels need to be considered during a scan. However, it does not determine how many label-propagation iterations are needed. The number of labeling iterations is mainly determined by the spatial structure of the pixels with strength>=T_low, especially how weak pixels are connected to strong pixels.
So your intuition is correct that more pixels below T_low means fewer pixels remain as labeling candidates. But this affects the amount of work per scan, not necessarily the number of labeling iterations. Two images can have different numbers of pixels with strength below T_low, but still require the same number of labeling iterations if the relevant weak/strong edge structure is the same.
That said, I agree that this option could have been worded more clearly, for example by explicitly distinguishing between “number of candidate pixels” and “number of labeling iterations.”
Best,
Good question.
The key point is that the option refers to the number of labeling iterations, not to the total amount of pixel-wise work.
Pixels with strength < T_low are rejected immediately, so they are not candidates for labeling. Therefore, their number may affect how many pixels need to be considered during a scan. However, it does not determine how many label-propagation iterations are needed. The number of labeling iterations is mainly determined by the spatial structure of the pixels with strength>=T_low, especially how weak pixels are connected to strong pixels.
So your intuition is correct that more pixels below T_low means fewer pixels remain as labeling candidates. But this affects the amount of work per scan, not necessarily the number of labeling iterations. Two images can have different numbers of pixels with strength below T_low, but still require the same number of labeling iterations if the relevant weak/strong edge structure is the same.
That said, I agree that this option could have been worded more clearly, for example by explicitly distinguishing between “number of candidate pixels” and “number of labeling iterations.”
Best,
