Where should framerate conversion be put in a script?

[Chainmax]

I am in the process of converting a progressive movie from PAL to NTSC for playback on a standalone. I assume that framerate conversion goes right after the final resize and before sharpening or noise addition, is that correct? If so, what if I have two sharpening steps? If not, where would you put it?

[Alain2]

What do you call framerate convertion ? If it's something like assumfps, it can be anywhere, if it's something that inserts / removes frames, I would do it before filtering, and if it's by playing with fields / telecine (like for a 3:2 pulldown), this need to be the last step.

[Chainmax]

It's DePan or MVFlowFPS. About using it as the first step, what if the source is too noisy and/or blocky? Wouldn't those kinds of artifacts throw off the motion estimation?

[actionman133]

I admit I'm a bit of a newbie to this Chainmax, but this is my method based on some fuzzy logic.

For framerate conversion where the frames are analyzed (like MVFlowFPS...), I figured you denoise a test clip slightly (if the source is a little noisy) and use that for input analysis. You then apply the framerate change to the ORGINAL clip. Then denoise, then resize.

I generally believe that MVFlowFPS requires the cleanest possible signal, so any resizing or serious denoising beforehand should be avoided if at all possible. Denoising should also be used on as clean a source as possible, but I figure that its more important for MVFlowFPS to return good frames. After denoising, then resize.

That's my method, but I don't follow this as gospel... perhaps others might be able to offer another method.

[Alain2]

You should try both Chainmax and see what works best. My feeling is it's better to do motion estimation (and framerate convertion in your case) before denoising because the denoised clip is less detailed and can have artefacts introduced (if strong spatial denoise / low temporal denoise).

[Mug Funky]

going PAL to NTSC, it's beneficial to convert framerate after final resize (because it's working on smaller frames and therefore faster. and if the quality of the conversion is bad, having a larger frame wont make it look less bad).

going NTSC to PAL, framerate conversion before resize is quicker, obviously.

[actionman133]

@MugFunky

An obvious benefit of your layout is speed, but wouldn't a larger sample help with the conversion? Even if there are errors in the frames, wouldn't downsizing *afterward* reduce the intensity, rather than making the same errors on a smaller source (which will look larger in the final clip)?