Moderators: Mug UK, Zorro 2, Greenious, spiny, Sorgelig, Moderator Team
Sorgelig wrote:If HDMI resolution is 1920x1080, it's already hard to process this resolution while you offer to magnify beyond this? You will climb to incredible clocks and hit the low-cost FPGA limits
Sorgelig wrote:Currently i only see the one possible adjustment - coefficients. If it will be possible to neutralize Lanczos function and act it as Near Neighbor, then it will be possible to choose between sharper and filtered looks.
Sorgelig wrote:I cannot control parameters of bilinear filter in run-time. Even in compile time on coefficient fractional bits can be changed.
Locutus73 wrote:Sorgelig wrote:I cannot control parameters of bilinear filter in run-time. Even in compile time on coefficient fractional bits can be changed.
How can we get a list of run-time variable coefficients for different algorithms?
Thank you in advance.
Best regards.
Locutus73
Sorgelig wrote:It seems Polyphase can be turned to Near Neighbor by coefficients.
Testing now - and it looks exactly the same as NN.
Sorgelig wrote:It seems Polyphase can be turned to Near Neighbor by coefficients.
Testing now - and it looks exactly the same as NN.
Sorgelig wrote:It seems Polyphase can be turned to Near Neighbor by coefficients.
Testing now - and it looks exactly the same as NN.
Locutus73 wrote:can Quartus VIP scaler use different sets of parameters for horizontal and vertical scaling?
So we could i.e. integer scale using NN the vertical axis and non integer scale the horizontal axis using "pixel edge interpolation"
Sorgelig wrote:Locutus73 wrote:can Quartus VIP scaler use different sets of parameters for horizontal and vertical scaling?
So we could i.e. integer scale using NN the vertical axis and non integer scale the horizontal axis using "pixel edge interpolation"
yes. Coefficients can be independent on X and Y. Coefficients can have any values in the range. They don't need to be exactly from Lanczos function.
Sorgelig wrote:So, i think with correct values it's possible to simulate bilinear. And of course it's possible to mix amount of polyphase making something between NN and polyphase.
Sorgelig wrote:Well. It still need to be tested. Currently i'm struggling with uploading coefficients correctly. Hope everything will be solved.
Locutus73 wrote:I knew that once we stimulate your curiosity you're unstoppable man!
Sorgelig wrote:Locutus73 wrote:I knew that once we stimulate your curiosity you're unstoppable man!
yeah.. like to tell the joke to entertain yourself...
Sorgelig wrote:By the way, is there an online calc of Lanczos function on given amount of taps and phases?
Sorgelig wrote:example of coeff.txt file for NN is here: https://github.com/MiSTer-devel/Main_Mi ... eff_nn.txt
Sorgelig wrote:It's not Altera specific thing. It's common known polyphase filtering.
some info is here: https://www.embedded.com/print/4017531
Locutus73 wrote:Good, and do you have the coefficients used by default when not using coeff.txt? Do they correspond to coeff_pp.txt?
Anyway as far I understand each row is a phase with its 4 coefficients right?
kitrinx wrote:I discovered today that if you build the cores with 1920x1200 as maximum input and output, you can use 1920x1200 resolution with mister, and the new integer scaling will correctly do 5x scaling to 1200. It might be worth building cores with 1200p support going forward.
Sorgelig wrote:I discovered if i grab 5 apples then there will be 5 apples in my hands. Magic!
Regards.
Locutus73
Users browsing this forum: Grabulosaure, SmokeMonster and 5 guests