FAQ Video Enhance AI

Hello everyone.

I would like to start a FAQ here. Main purpose is to have something to link to in case of.,… well, FAQ… you know…

Some FAQs will be beta or alpha related, some “generall”, so I´ll probably do sections…

I will be filling this up - and if it makes sense - will copy the generall stuff over to the VEAI Generall forum. So if anyone wants to contribute - feel free to answer here, I will add it into this first post (which will get rid of this introduction stuff in the future). Hope it´s possible to edit older posts …

One request for the forum: Is it possible to add a generall FAQ or sticky function? Also, the layout is squeezed in width, my FHD Monitor is only used in half.

Hi Reiner
I use a Mac Pro 5,1, I believe it is still quite popular basis for video/photographic media processing. It is becoming rather tricky to balance conflicting requirements for various softwares. In particular, I am not able to upgrade to Catalina or beyond. Even for Mojave that is required by BMD Resolve 16, I have had to change my NVidia GPU to an AMD card (RX580).
As I understand it, this makes it a non starter for GPU acceleration with VEAI?
I know that this machine is over 10 years old, but there are probably 1000s still in active use due to unattractive/uneconomic Mac alternates.
A FAQ specifically for ‘MacPro 5,1 support’ would be great!

For inclusion in VEAI FAQ:

Will VEAI work on DVDs without first deinterlacing?

What deinterlacing software is recommended:
• for Mac?
• for PC?

The only deinterlacer that doesn’t reduce picture quality (when properly used) is QTGMC. Read up on interlacing and how the method used in QTGMC is unlike any other.


http://avisynth.nl/index.php/QTGMC

Download QTGMC with all the needed plugins from here:
https://forum.doom9.org/showthread.php?p=1928244#post1928244

Examples from Avisynth scipts I’ve used. What setting you use depends entirely on the video. I could literally write a book on what settings are used when and where.

This a baseline reference (does only deinterlacing, no noise reduction):
by -Vit- (the creator of QTGMC, term loosely used, hundreds if not thousands of developers created QTGMC):
“The most “accurate” reasonable settings I’ve measured (measured by best SSIM that doesn’t introduce any noticeable artifacts) are:
QTGMC( Preset=“Very Slow”, EZKeepGrain=0.5, SourceMatch=2, Lossless=2, Sharpness=0.1, Sbb=0 )”

I have use that followed by noise reduction (FluxSmoothST) with good results.

Examples:
In a 1080i video:
DGSource(“q:\wc\cpw-S01E13_track1_und.dgi”, deinterlace=0)
AssumeTFF()
QTGMC( Preset=“Slow” )
SelectEven()

In a 480i film DVD:
AVISource(“lg480i-29.97-huffy.avi”)
AssumeTFF()
ConvertToYV12(interlaced=true)
TFM()
TDecimate()
QTGMC( Preset=“Placebo”, EZKeepGrain=0.1, SourceMatch=2, Lossless=2, Sharpness=0.2, Sbb=0 )
SelectEven()
FluxSmoothST(temporal_threshold=7, spatial_threshold=7, luma=true, chroma=true)

In a 1080i video where every field has unique data (no SelectEven() to delete every other frame):
DGSource(“x:\w6\c-qsf.dgi”, deinterlace=0)
AssumeTFF()
QTGMC( Preset=“Slow” )

Note interlaced video CAN sometimes have unique data in each field. This is especially nice when the double framerate is wanted in the final video (sports or live concerts). If you can get that 50fps without having to interpolate frames (SVPFlow, FrameRateConverter, Dain, etc), it can be a good thing (speedwise, quality typically the same either way).

In a 480i film VHS and with QTGMC to also denoising. This is NOT recommended if using VEAI to upscale as it reduces the maximum amout of possible detail. It’s very important to understand that too much pre-processing can limit VEIA.
AVISource(“tatt480i-huffy.avi”)
AssumeTFF()
ConvertToYV12(interlaced=true)
TFM()
TDecimate()
QTGMC( Preset=“Placebo”, EZDenoise=2, NoiseProcess=1, NoisePreset=“Slower” )
SelectEven()

A better script for VHS film to VEAI:
AVISource(“tatt480i-huffy.avi”)
AssumeTFF()
ConvertToYV12(interlaced=true)
TFM()
TDecimate()
QTGMC( Preset=“Very Slow”, EZKeepGrain=0.1, SourceMatch=2, Lossless=2, Sharpness=0.1, Sbb=0 )
SelectEven()
FluxSmoothST(temporal_threshold=7, spatial_threshold=7, luma=true, chroma=true)

Clean noisefree 1080i video:
DGSource(“thkd.dgi”, deinterlace=0)
AssumeTFF()
QTGMC( Preset=“Fast” )
SelectEven()

The take away from that is you need to study the QTGMC.avsi script and note how TR0, TR1, and TR2 work. Preset Placebo usually oversmooths (for VEAI). Understand Lossless mode undoes temporal denoising done by QTGMC, and is sometimes desirable. Fast is only to be used with clean noisefree video.

Progressive is far simpler to work with.

And now maybe you understand why VEAI does not do deinterlacing. To do it right, you’d have to create an AI with the experience of someone who’s done digital video for 20 years, and I’m not talking about folks who work in TV station. Those folks don’t even know what QTGMC is… You need a video guru who’s spent many years doing video on a PC with doom9 and videohelp tools. I’ve used Vegas on many projects, and it’s great in certain use cases, but useless for most VEAI projects. The tools on doom9 and videohelp are much more practical. Get to know Avisynth, Virtualdub (and all the forks including Virtualdub2), and the thousands of plugins written for each.

2 Likes

Thanks very much for your reply. It should be part of a VEAI FAQ. I’ll proceed with what you’ve laid out.

Hi, I am trying to learn what compression factor is, but I cannot find a good answer anywhere. What does it mean and how should I find a good setting for it?

1 Like

Hi. Are we talking about compression factor with some MPEG compression (AVC;HEVC,h264,. etc…) ?

Yes, when I am in VEAI there’s an option for Compression Factor for MP4 on a scale from 0 to 35 with 17 being the default. What do these numbers mean? Is higher or lower better ?

In a nutshell: The lower the number, the better the quality, the bigger the file size the bigger the bitrate.

17 is around “visually losless” for most cases.

0 would be losless (big files sizes)

the lower the resolution, the bigger the Q-Number should be to look visualy “ok”.

try and do two testclips. one with 10, omne with 30 - you will see the difference in the output file right away. pay attention to edges near uniform areas and small details, it´s most visible there.

for more information, google “constant quality x264” for example:
https://slhck.info/video/2017/02/24/crf-guide.html

1 Like

Thank you for the explanation - that really helped!

1 Like

you can also use staxrip beta which has a GUI interface. it has the QTGMC plugin built in.

We have a version in beta with a new deinterlace model that you may be interested in. You can join our beta group here:
https://community.topazlabs.com/g/Video-Beta-Testers

Hi Guys - my orriginal intend was to start a FAQ for Beta testers in the first post - the post was originally put in the BETA section.
Someone moved it without consulting me.

For this to make sense, I would have to be able to change the first post and move it back to the beta forum. None seems possible at the moment.

Any way some mod could give me the propoer rights to do that?

That’s an amazing and comprehensive explanation but above my pay-grade!

I have over 100 hours of 720x486 SD interlaced footage digitized to ProRes422 at 29.97 frames and 47k bit rate. Since I’m not a video engineer with your experience, I’ve been waiting patiently for the new VEAI with interlacing to tackle this project.

I just tried the new Dione TDv1 and the results have been jaw dropping, spectacular. Kudos to Topaz! My only issue is I’m missing frames 27, 28, 29 and 30 so it is unusable presently. Any idea what is happening and how I can fix that? Cheers! WLS