What's with the 'Video Scaling Method' / 'High Quality Software Upscaling' thingy?

I've never really understood the different Video Scaling Methods that you can choose when you play a movie. What's the differense between this and the 'High Quality Software Upscaling' options under Settings -> Videos ->
Player?

High Quality Software Upscaling, Upscaling Method: [Bicubic], [Lanczos], [Sinc]

Video Scaling Method: [Nearest Neighbour], [Bilinear], [Bicubic]

There's also something unstable about the Video Scaling Method. If I try to select Bicubic I get a white video playing. And if I try to click 'Set as default for all movies' and restart the movie I will get no video, only audio in the back. When I push 'X' to stop the audio playing then xbmc crashes (hangs..) and I will need to kill it. I'm currently using Beta 2, but I've seen this for a long time.. atleast since Alpha 4. Oh.. If i start xbmc again and play the same video, then it appears the the 'Video Scaling Method' was not saved also. (However the cpu usage suggests otherwise.)

One more thing.. Why isn't the 'Interlace handling' option a general option under Settings -> Videos -> Player? I also don't understand why the 'Video Scaling Method' isn't there as well. You can even call me a complete idiot cos I don't know why I can select 'High Quality Software Upscaling, Upscaling Method: [Bicubic]' And still select 'Video Scaling Method: [Nearest Neighbour]' when the video is playing at the same time. However [Lanczos] can't be selected when you play a movie?? What's all this!!!


Still,

Keep up the good work guys!

well i think software upscaling is just supose to upscale your movies to your selected resolution altho software upscaling no longer works for me in beta2 makes my movies choppy and sound out of sync and i have a very powerful htpc. used to work in beta1 for me but not in beta2.

Odd that it doesn't work for you in beta2. Works way better for me. beta1 would make the cpu jump through the roof. Now it is good.

Eicar, sorry, didn't mean to hijack your thread. I wish I could answer you as I am also confused about all these different scalings!

Maybe due to the new ffmpeg libraries? They're now using multicore and the -DRUNTIME_CPUDETECT flag.

I discussed this with WiSo. I did not realize the -DRUNTIME_CPUDETECT flag was set on windows ffmpeg. That was removed on linux due to causing issues with high cpu with software upscaling (higher than it should be).

WiSo said he will build another ffmpeg without the -DRUNTIME_CPUDETECT flag so that you can test.

I just tested the new ffmpeg build and it solved the playback problem.
(Does not crash anymore.) Thanx!

However I would really like an answer on my original questions :o

-Eicar

Eicar Wrote:IHowever I would really like an answer on my original questions :o

Same here.

I googled "xbmc video scaling method" just to find the answer to these same very two questions of the OP. I just registered to the forum to bump this topic until someone comes up with a real answer for this.

Eicar Wrote:I've never really understood the different Video Scaling Methods that you can choose when you play a movie. What's the differense between this and the 'High Quality Software Upscaling' options under Settings -> Videos ->
Player?

High Quality Software Upscaling, Upscaling Method: [Bicubic], [Lanczos], [Sinc]

These methods use FFMPEG's software scaler (swscale) to resize the source video frames to the size of the screen (XBMC's resolution). The differences between them is the algorithm used and the number of "taps" performed for calculating the final pixel's color. Sinc uses 25 taps I think, Lanczos and Bicubic use 9 or 16. Sinc is generally the highest quality but also the most time consuming, followed by Lanczos and then Bicubic, but of course, this can be subjective. Note that all these methods are software methods, they are not hardware accelerated, so enabling them can result in frame drops and out of sync audio depending on your CPU, resolution etc.

Quote:Video Scaling Method: [Nearest Neighbour], [Bilinear], [Bicubic]

These methods are hardware accelerated. Nearest neigbour performs no upscaling at all, so it is the most "pixelated" (GL_NEAREST in OpenGL). Bilinear, as the name suggests, uses bilinear interpolation (GL_LINEAR in OpenGL) for texture scaling. Bicubic (currently broken on many GPUs) uses a pixel shader to upscale the textures using a 4x4 bicubic filter. Bilinear is the default and should get you good quality.

However, you are correct, these settings need to be consolidated and they currently are not, leading to confusion.

Many thanks for clearing that up because I always thought 'Nearest Neighbour' meant XBMC would pick one of the other two to use.