De-interlacing Grass Valley HQX files....

[pippas]

Although it's perhaps a bit of an 'old fashioned' approach these days, I still prefer to convert the raw (.MTS) files form my HD camcorder into an intermediate format before editing. I find the intraframe files make the editing process 'feel' more like editing DV, with 'smooth' timeline scrubs, fades and transitions, etc.... even on a less than high spec computer!

My own choice of intermediate codec is Grass Valley HQX (was Canopus), which is now available as a freebie from HERE.

I find the easiest way of using that codec is in conjunction with their free AVCHD2HQ converter utility, a copy of which you can find HERE - or you can register with Grass Valley to get the very latest version.

Although originally intended to be used with the Grass Valley Edius software, both the utility and the codec work well without Edius being installed, although not all the converter options will work (Conversion to HQX works fine)

My HD camcorder records as 1080i , and I use the de-interlacing option in the AVCHD2HQ utility to produce either 1080/50p or 720/50p HQX files to work with. The de-interlacing results seem to be on a par with other methods I have tried.. certainly apparently as good as Yadif, for example.

Is there any way of doing objective comparisons between the quality of de-interlacing options (i.e. other than just 'looking' at the results)?

The Grass Valley de-interlacer seems to be very good. But then as it probably has Procoder in it's 'genes', that's perhaps only to be expected.

Or should I keep the HQX files as interlaced, and de-interlace right at the end of the editing process, using Yadif or QTGMC perhaps?...

Experiments so far seem to look pretty good using progressive HQX files.......it would be nice to find a way of comparing results objectively....

[jagabo]

Is there any way of doing objective comparisons between the quality of de-interlacing options (i.e. other than just 'looking' at the results)?

Not really. You could start with a 1080p60 source, interlace it to 1080i30, then deinterlace back to 1080p60 and run PSNR or SSIM tests with MSU Video Quality Measurement Tool (the free version is limited to SD resolution) comparing the deinterlaced video to the original 1080p60 video. But what those kinds of test measure does not correspond well to human vision.

To use the free version of MSU VQMT you could resize high resolution source (say, 720p60) to SD resolution and work from that as your starting point.

[poisondeathray]

Other free ways to test with no resolution limitations would be avisynth's compare() for PSNR, or the SSIM filter, or ffmpeg's PSNR filter . But as jagabo mentioned, those measures don't correlate that well with subjective perception

[newpball]

In case someone makes tests I am interested in seeing the results.

Seems like an interesting option.

Just downloaded this and tested it with Premiere Pro on some UHD material.
Looks good, low CPU usage!

[pippas]

@jagabo and PDR. Thanks for the suggestions guys. I shall investigate further. As you have said though, comparing results 'mathematically', as it were. doesn't always convey what any differences may look like in practice.
Even looking at the same file on different monitors - or using a different player - can make quite a difference in what you see anyway.

One thing that has surprised me though. I didn't expect that my WD Live player would playback a 1080/50p file 'smoothly', but it does.
Mediainfo reports the file as having a 'High @L4.2' profile and that's outside the spec of the Sigma chip fitted to this WD unit).
Still, don't look a gift horse in the mouth, as they say!

@newqball. Yes, pretty low CPU usage... which is why one reason I like it. I also like the fact that I'm not trying to edit long GOP files with 'tricks' like smart rendering and/or proxy files. I like the intraframe format. As I said in my earlier post, reminds me of editing DV !

There's a white paper on the HQX codec HERE which you might find interesting.

It certainly offers a useful way of editing 4K footage simply - and for free!

[poisondeathray]

In terms of "visually lossless" codecs, cineform is also free, fast , high quality way of editing UHD and 4K..

More importantly, it has been recently certified by SMPTE and is now the industry standard, open format, VC-5 (hopefully it will make it into open source tools like ffmpeg soon) . There was always the question of whether or not something proprietary will stop getting developed or no longer supported. (e.g. will it support 8K or other colorspaces in the future? Bt. 2020? etc..) Now that it's both open, and a professional standard, it's no longer a question

https://www.smpte.org/news-events/news-releases/gopro%C2%AE-cineform-codec-standardize...-vc-5-standard

[pippas]

That's an impressive result for David Newman!..... IIRC, Cineform and Grass Valley both made their intermediate HD codes available for free at around the same time, although some of the more advanced Cineform features are still payware.

I seem to remember that some folk always considered Cineform more 'modern' than Canopus HQ (now Grass Valley HQX) because it employed wavelet technology rather than DCT. I'm guessing that will count in its favour, moving forward?.....

I have certainly found them both very impressive. As an amateur, I was also attracted by the Grass Valley AVCHD2HQ utility which I mentioned earlier. This made HQX easier to use than Cineform - at least from my own amateur perspective.

I'm guessing the Cineform may now come on in leaps and bounds, following the SMPTE announcement. Whether any of the more advanced features will now become freeware?.......

The HQX codec is already usable up to 10 bit and with an alpha channel... and can edit up to 8K if needed. And those features are already freeware.
It'll be interesting to see which way Cineform develops......

[poisondeathray]

Yes, things like RAW support, 12 bit, etc... are all still only available if you bought a paid version as of right now. As of right now, only the 10bit version is available for free. But they don't even sell those other versions anymore - so I suspect everything will become all available soon - someone just has to compile things from the library, and/or make a GUI around it.

But for the old cineform GUI's and accessory tools like HDLink (they had accessory functions like deinterlacing, IVTCing, non destructive metadata transforms like grading, etc...) those are part of cineform's (and thus GoPro's) IP, so I doubt those will become available for free

Wavelet compression has different sorts of problems, different sorts of compression artifacts - so its not all necessarily good.

[racer-x]

My PC (i7) is fast enough and my editor (Aviutl) is robust enough to edit native 4k footage, even HEVC source with ease. I downloaded this HQX codec and agree it's pretty good. It would be good to use if Your PC isn't fast enough.

I can't seem to find where to specify 10-bit or Alpha Channel in the settings though....

[pippas]

According to the description in this comparison thread the alpha channel support for HQX is 'internal'........

The white paper on the HQX codec HERE mentions that the alpha channel support is 10 bit and 'built in'. Page 3 also suggests there is more detail on alpha support later on in that paper, but I haven't found it so far.....

[poisondeathray]

Alpha channel works for HQX - it's enabled by default.



I had HQX installed on one of my computers for an old project, but I updated it to the recent version run some quick observations:

HQX is good (much better than long GOP), but still noticably slower than cineform (just scrub around on a timeline for a UHD source) in 64bit applications - it's about 1/2 to 2/3 as fast for decoding latency (at least on UHD) using it's native decoder .

You can also use various benchmarks to test and quantify VFW and directshow performance directly:

In 32bit VFW applications, it's noticably slower - you can quantify this by using avsmeter on an avs script to test VFW it's about 1/3 as fast. Or another method would be to use 32bit vdub and test the speed by running analysis pass . (you can repeat for 64bit versions) . VFW is what most editing applications would use on windows

You can also test directshow decoding performance (what would be used in most media players) with the graphstudio benchmark with both 32bit and 64bit tests. It's quite slow in directshow, about 1/2 as fast as cineform

Interestingly, ffmpeg's decoder for HQX is more optimized - it gets about 50% more speed that the native canopus/grass valley decoder . ffmpeg doesn't have a cineform decoder yet


Next would be to do some compression tests /qualilty / PSNR at various levels, encoding speed, compatibility testing, quicktime testing (repeating the tests with the MOV variants). I might do this if I have time later but it takes a lot of time to do it properly