Downscaling

[alleycat]

Wondered if anyone knew - obviously upscaling results differ between programs and settings etc. When downscaling video, say 4k to 1080p, do different programs yield different results? For example is there any difference between using the resize filter in Virtualdub and the scale in Premiere? Or are they all as good as each other? I can't see any difference but wanted to check if there in theory should be. Thanks

[PDB]

In theory different methods should produce slightly different results but in practice at least using AP, I've never seen a major difference when testing. As long as you have certain settings checked.

AP uses the 3 major methods of upscaling/downscaling (plus nearest neighbor) depending on the video being previewed or using transform effects. From what I remember, by default it uses a mix of Bilinear and Bicubic while previewing and in the final rendering (depending on the set to frame or scale to frame settings). You do also have access to Lanczos if you use the transform effect. Obviously, this is a more important difference in upscaling but for downscaling when I tested years ago, I really didn't see much a difference. I have for a while been upscaling projects to 4K to try to maximize details and then scale them back to 1080p using the different methods. I have never found the Lanczo to be any better or worse then the Bi methods in downscaling. The big difference is the Lanczo/transform took a lot long to render.

The more important thing to know with AP is to have the render settings correct. You should always have Use Maximum Render Quality selected. Even though it is slower, when you don't have it checked, AP will throw out info to lower the resolution versus properly downscaling it with that check selected. That's most important thing to have when downscaling in AP.

(Also it never hurts to have 32bit render selected also even with the increased render time. It helps with those 8-bit colors)

As for AVI I vaguely remember a debate on the OT (which I can't find) that resampleHQ using spline36 was the best as it didn't have "gamma issues". Others methods had issues but I might be remembering that wrong. I've used vanilla spline36 or downscaled in SR with 36 and both seem fine to me but perhaps someone has more info then I do on that.

[alleycat]

Thanks @PDB this is all very useful.