[MEncoder-users] Tuning hqdn3d

[Andrew Berg]

Phil Ehrens wrote:
> This is highly subjective, and reasonable people will say I am
> on crack, but this is what I do:
>
> I avoid using hqdn3d on live material, and use nr instead.
> Something I might do with moderately bad source is nr=120 and
> unsharp=l3x3:0.1 to compensate a bit. When all I want to do is
> to hide some slight mosquito noise, nr=60 works very nicely.
>
> With animated material, start with 3:2:3:3 and increase the value
> of each component by "1", (for example, the next higher setting
> would be 4:3:4:4) until it becomes too blurry, then back off.
> With really awful multi-generation VHS stuff I have gone as high
> as 12:11:12:12. Pretty scary.
>   
If it works, I don't really care what you're on. :-D
I have plenty of live action and animated material, so I'll have to try
those. I'll also have to see if the unsharp filter will make certain
deint filters look better. yadif slows things down a lot, and they want
me to add mcdeint to it! Maybe when I have a CPU in the 6-7 GHz range.
> Related observations that are veering off-topic:
>
> Something that's a bit irritating is that when you use mplayer
> to feed ffmpeg via a FIFO, the nr is applied late, and
> it would be nicer to be able to apply it before the mplayer
> filters. But it may also be good, because it's probably the
> reason why my combination of nr and unsharp looks so good.
>   
I don't feed anything from MEncoder to FFmpeg, so no worries there...
> Note that the result of applying hqdn3d with Transcode is much
> better looking than with mencoder. I have no idea why this should
> be. The obvious difference is that with Transcode there are no
> "snail trails" even with very high values.
"Snail trails"? And what is Transcode (searching for such a generic term
would likely get me nowhere)?