I'm not even sure how video ends up like this. Crappy resizing??
Anyway, I've attached a very short Bluray sample (due to my slow internet connection) but it should be enough to show the problem. If anyone can explain how a video ends up like this and if there's a way to fix it, that'd be much appreciated.
Oh..... watch the lower part of the white keyboard. That's the jagged lines I'm referring to.
+ Reply to Thread
Results 1 to 12 of 12
I think it was originally 25i and someone deinterlaced it badly (discard field and resize?) to 25p.
Oh, and an imperfect fix:
SeparateFields() SelectOdd() nnedi3(dh=true)
Last edited by jagabo; 8th Jul 2015 at 08:57.
I'll give it a shot. Thanks.
It's somewhat disappointing stuff can make it to Bluray in that condition.
SelectOdd() seems to look a bit sharper for some reason.
Even though it's no doubt technically wrong, deinterlacing with QTGMC following nnedi3 seems to produce a slightly sharper result than QTGMC in progressive mode. Like this:
I need a faster computer......
How many cores do you have? Are you using the multi-thread version of QTGMC, especially since this is HD video? I use the MT version of QTGMC. Tweaking the parameters, I am able to fire all 8 threads of my machine at near 100% and the fps impact is significant. I can't imagine running QTGMC in single thread mode.
Usually I can get the CPU running fairly hard, but if not I run two encodes simultaneously. I had a few issues with the MT version of Avisynth (coming back to the PC expecting it to have spent hours encoding only to discover it'd spent hours displaying an error message) so I've stuck with the single threaded version.
I've experimented a bit with MP_Pipeline lately. It tends to speed things up a bit, probably because of the memory limit for a single 32 bit process. Although with MP_Pipeline I can usually only run one encode at a time due to the extra memory usage or it all goes to crap (I'm still running XP). I posted some initial results a while back, but it might be worth playing with even if you're using MT Avisynth.
This is single threaded Avisynth, running a single encode, re-encoding a 1280x720 video with the following script.
QTGMC(InputType=1, Preset="medium", EzDenoise=1.5)
### prefetch: 16, 0
crop(8, 4, -6, -4)
I'm still having a bizarre issue with the latest QTGMC and LSFMod in the same script. I run QTGMC in progressive mode for noise filtering quite a bit. Originally if I started encoding from the beginning it'd run at half the speed that it would when using Trim() to start encoding from not quite the beginning. MP_Pipeling seemed to fix it in that case, but the video I'm encoding at the moment (not the sample I posted in this thread) has exactly the opposite problem. If I start encoding from the beginning it runs at twice the speed it does if I use Trim() to not encode from the beginning, and even MP_Pipeline isn't helping.
It's only a problem with QTGMC 3.33d or 3.33s (the latest version from the doom9 QTGMC thread). QTGMC 3.33 or earlier are fine. Computers....
Last edited by hello_hello; 8th Jul 2015 at 21:10.
Wow, I am surprised you get 95% CPU usage from a single threaded version of QTGMC. I am not sure why my PC is different but I can never come even close to that in ST mode, more like 25% to 30%. My PC isn't exactly new either. It is an i7-3770K from 2012 so three years old. But like you, I am eager to rebuild when Skylake arrives. Although I am really hoping for a Skylake-E with 12 physical cores cuz I want my next machine to be 4K ready.
CPU usage can sometimes be low, and QTGMC in progressive mode doesn't slow things down quite as much as when it's in de-interlacing mode (from memory), but MP_Pipeline does help increase CPU usage if you're using additional filtering (I assume because it increases the amount of memory Avisynth can use) but depending on the resolution that can all change a bit. For standard definition sources I don't bother with MP_Pipeline and just run two encodes at a time (or split encoding jobs in half and run both halves together). That usually keeps the CPU sitting very close to 100%.
Like you I just tweak settings as appropriate for maximum CPU usage, only I guess I fiddle with different ones.
It appears the old man performing is Joe ****er (rest his soul)