Deinterlacing comparison between DGDecodeNV (Nvidia Purevideo) and QTGMC

[8-BaLL]

I thought It would make sense to create a new topic, since searching in google for comparisons between DGDecNV and QTGMC just returned some old threads, without real comparisons.

Well here you can check it out yourself NV Purevideo vs QTGMC:

http://screenshotcomparison.com/comparison/129017

Yadif vs NV Purevideo:

http://screenshotcomparison.com/comparison/129025

Script for DGDecodeNV:

For QTGMC:

Source: Purevideo Yadif: QTGMC:

Just thought Id share, because Ive seen many ppl ask about Deinterlacing comparison of Nvidia Pure Video compared to others, but noone actually provided real examples.
Most notable difference is on top of the Iphone Case and the text on top of the display.

Ofcourse QTGMC also removes the interlacing artifacts, which Purevideo doesnt, so I wasnt quite impressed with NV deinterlacing.

I think Yadif and some Yadif mods are even better quality than Nvidia Pure Video.

Edit: Yadif example added- Yadif has slightly better quality than NV PureVideo, you can see it on the very top left corner of the iphone case and the text on the screen is less interlaced, most visible on the word reiceve+travelling.

Edit2: Here are a couple of comparisons with more motion in the scene:

Source: Purevideo: Yadif: QTGMC:

[huhn]

here so test from me:
http://forum.doom9.org/showthread.php?t=170420

btw. your screens are not always the same screen.

[8-BaLL]

Oh thanks, the source one is wrong. Ill fix it now and add Yadif example.

eidt: it is the same image, its just because the frames carry 2 subfields (pal 50i), and when it gets deinterlaced, the source images slightly changes.

Yadif example added- Yadif has slightly better quality than NV PureVideo, you can see it on the very top left corner of the iphone case and the text on the screen is less interlaced, most visible on the word reiceve+travelling.

[stax76]

How many fps do you get with 1080i and QTGMC very slow with which CPU?

Wouldn't quality with QTGMC be always better if you would output double frame rate?

[8-BaLL]

I have around 2 fps with i5 Quad Core @ 4Ghz and QTGMC(preset="very slow"). If I add resize to 720p (spline64resize), the fps is around 1.7fps. Without QTGMC, the encoding speed is around 11-12fps with the exactly same x264 preset.

1st native 50fps is usually on live TV shows, reality shows, some documentaries, sports, etc. Film and tv series are usually 23,976 / 25.

The quality as such doesnt change if you encode native 50fps material to 25fps. The only change is the motion fluidity, and its REALLY noticeable to me and I prefer super fluid motion, esp. for sport stuff. From my experience, encoding with 50fps makes the file around 20-30% bigger, the quality stays around the same.

Most ppl dont encode at 50fps though, because its basically 2x the frames you have to encode so it takes twice longer to encode. Its personal preference. Id rather wait longer but have fluid 50fps than going down to 25fps.

[stax76]

I thought it would be slower, are all German HD channels always 50 fps? Are you using KabelBW? Can you record everything are are some movies locked? What is your recording device or software? Which software do you use to demux? Which software do you use to cut? Also I wonder if 50i can be much better then 50p. Thanks for sharing your experience and sorry about all the questions.

[8-BaLL]

Originally Posted by stax76

I thought it would be slower, are all German HD channels always 50 fps? Are you using KabelBW? Can you record everything are are some movies locked? What is your recording device or software? Which software do you use to demux? Which software do you use to cut? Also I wonder if 50i can be much better then 50p. Thanks for sharing your experience and sorry about all the questions.

Slower? I think 1.5 fps is slow enough, esp. when encoding with 50fps, which doubles the frames.

No, tv series and movies are usually 24fps/25fps native. Stuff like Live TV shows, sports, reality tv shows, documentaries are usually pal50i. And its not just on german tv, its on most HDTV channels.

I use a SAT receiver to record stuff.

50i is usually beeing deinterlaced by Sat receivers/TVs to become a 50p image, so in a certain way the quality is around similar with a good deinterlacer. But qtgmc is better than any deinterlacers in most devices up to like 1000 dollars and probably even above that if its an AV receiver. The thing is, that qtgmc removes interlacing artefacts, and applies certain denoising and smoothng filters, which I dont think hardware deinterlacers do.

For the other stuff I think you should go check out ba eginners thread or something like that.

[huhn]

germany is a pal region so everything in free TV is 50p or 50i.

movies in pal regions are usually 25 PSF so you can just ignore the interlaced flag and be done.

BD are usually still the normal 23/24p.

[foxyshadis]

The phone in your first set looks field-blended, making it a bad choice for comparing deinterlacers. The second is good, and definitely shows how soft purevideo is.

[tebasuna51]

Of course qtgmc(preset="very slow") is better, and is recommended to recode when you don't have time problems.
Congratulations to qtgmc developers for the good job. We know than AviSynth have powerfull tools to improve the image.

But take in mind than Purevideo deinterlacer must be designed to work in real time, try to play the two scripts.

[hello_hello]

8-BaLL,
for a bit of a speed increase when encoding have you tried de-interlacing to 50fps and resizing to 720p? I don't know how much it'd speed things up but usually there's no noticeable difference in respect to picture detail. Your second lot of screen-shots don't look like they'd have anywhere near 1080p worth of picture detail.

They're only standard definition, but I'd be interested to see how Purevideo does compared to Yadif and QTGMC with the samples in the zip file I've linked to. The one called "text" has text at the bottom of the screen that appears to test de-interlacing quality, especially at the end where the text scrolls. The one labelled "blinds".... well if you keep your eye on the window blinds in the back corner as the camera pans from about the halfway point you'll probably see some de-interlacing differences. If you feel like doing some comparisons......

Even Yadif at 25fps doesn't look good de-interlacing the "text" sample to me. Not so much because of the wobbly text, but because the picture motion is obviously "jittery" compared to 50fps. For QTGMC it's similar, but the difference between 25fps and 50fps doesn't seem as obvious.

interlaced.zip (30mb)

[8-BaLL]

Originally Posted by foxyshadis

The phone in your first set looks field-blended, making it a bad choice for comparing deinterlacers. The second is good, and definitely shows how soft purevideo is.

Its not field blended, the reason why the text looked blurred is the Iphone animation, when you open the Messenger, it kinda does that zooming in animation, and the 1st image is during that animation

@tebasuna51- yes sure it was designed to work in real time with Pure Video. However the GPUs have insane ammounts of calculational power, Im sure something like qtgmc would be possible if they cared to implement something like that. Also their Deinterlacing solution was introduced with Purevideo HD first generation, when GeForce 6600 came out (http://en.wikipedia.org/wiki/Nvidia_...n_PureVideo_HD). Ever since they didnt change it, at least its not mentioned on the Wiki page. And 6600 came out in 2004. So their solution is stone age old by now. We didnt even have proper HD stuff back then.

@hello- thank you for your suggestion, I will do some tests with your files. I wonder how all 3 deinterlacers will perform in that scenario. If you want me to test a specific frame in both files, let me know.

[hello_hello]

Originally Posted by 8-BaLL

@hello- thank you for your suggestion, I will do some tests with your files. I wonder how all 3 deinterlacers will perform in that scenario. If you want me to test a specific frame in both files, let me know.

I'm just interested to see how they handle the hard to de-interlace bits. The text on the screen in general for the first one. And where you'll often get de-interlacing artefacts where there's straight lines and slow motion. Like the blinds in the back corner halfway through the second sample. But you need to compare movement, not still frames. QTGMC should do reasonably well. They're only small samples.

I've linked to a second small sample below. Watch the top of the desk frame at the edge of the desk closest to the camera (it looks to be metal at the top). Also watch the paper tray just behind it. With Yadif de-interlacing they'll both wobble a bit and even QTGMC doesn't seem to de-interlace the desk frame perfectly. You'll see a bit of "jaggedness" across the top when the camera moves. I don't know the exact term for it, but you should see what I mean. I wonder how PureVideo will do.

Wherever video gets de-interlaced on playback using my PC and MPC-HC these days (I haven't really thought about it given I virtually always de-interlace with QTGMC and encode before watching interlaced video), the edge of the desk is one time it's not de-interlaced very well.

interlaced2.zip (6MB)

[CruNcher]

It would be nice to compare how Nvidias Purevideo (VPX) Deinterlacer compares vs Intels Hardware Deinterlacing since Sandy Bridge and Yadif don't you think

And are you sure Nvidias Deinterlace didn't changed throughout VPx Versions ?

As you actually can see in GPU Reviews it got better and better over time.

And im not sure that even their Deinterlacer (especialy now with Maxwell) is done through the Shaders anymore in the Days of the 6600 that might have been still the case but nowdays ?