I thought previously there was a little used option on when to do the masking pre- or post- module. I think that got removed. Maybe I am just making it up in my head; my memory is pretty bad lately.
I think the aim would be to āignoreā pixels not covered by the mask when calculating the blurred pixel, and then apply the mask (to pass through the original pixel or the blurred one, as we do now):
. . X X X
X X X X X
. X O X X
. . . X X
. . . . X
If āOā marks the input pixel (itās covered by the mask), āXā marks the neighbouring input pixels covered by the mask, ā.ā marks input pixels not covered by the mask, and the blur is a simple averaging, donāt calculate Oā (the pixel in the blurred image) by averaging all 25 pixels, but only from O and X, 3 + 5 + 4 + 2 + 1 = 15 pixels. (Of course, the mask is not binary, so instead of 0 or 1, use the appropriate weight according to the mask; and the blur is not just an average.)
Yes indeed, either to ignore those pixels outside, or inpaint border pixels into the unmasked area depth >=$radius of the blur as shown by Boris. Both probably hard to do as it may require some pipe architecture change.
1 Like
Once you get to know it, itās a lot more simple than it looks at first glance, but itās not nearly as simple as it could be (without sacrificing power).
Such an excellent demonstration. Also demonstrating the effort that would be required to make this work.
2 Likes
Yes, I think so. As @s7habo showed Iām not sure if it would require inpainting, or is the blur can simply be based off the unmasked pixels as you showed?
But yes, definitely what I was trying to explain.
1 Like
Yes, here: Difficulty with Blurs and Masks
Agree, ā or continue on that old thread.
1 Like
What an excellent result. I wonder if we could ever do this in DT or maybe this is were we just run off to GIMP. Now this is one video I would watch for sure.
I did.