Sorgelig wrote:I cannot change algo at runtime,
I know that.
Sorgelig wrote:but i can change coefficients.
Oh my, that's interesting...
Sorgelig wrote:Since you are familiar with scaling,
Don’t think I’m so expert… I know random stuff learned over the internet since I’m interested in gaming and I’m a software guy, but I’m not an expert in signal analysis at all. Anyway I can try to do my best and I can communicate with a wide range of other people.
Sorgelig wrote:may be you can explore the possibility of set specific coefficients for Lanczos to make it work as NN scaling.
And that’s brilliant, so you’re thinking instead of switching algorithm at runtime (impossible with Quartus VIP), using edge case parameters in order to make one algorithm to work like another one!!!
And that’s not limited to Lanczos3; i.e. a optiroc, a guy in the Classic Gaming Discord, proposed to changing bilinear biasing in order to alter its linearity… so it could work both as a Nearest Neighbor and as a “pixel edge interpolation”, basically the same result as integer scaling+bilinear approach, but in a single pass… in the end the interpolation needed in a non integer scaling scenario where we want something that works just like NN inside an enlarged pixel, but interpolates (possibly using a single final pixel) when it reaches it edges.
That’s an explanatory picture optiroc made.
Basically 0% is linear (bilinear if used on both axis), 100% is Nearest Neighbor and 50% (or another complex curve) is what we want for interpolating just on pixel edges.
Thank you in advance.
You do not have the required permissions to view the files attached to this post.