DGDecomb

These CUDA filters are packaged into DGDecodeNV, which is part of DGDecNV.
Post Reply
DAE avatar
Guest

Re: DGDecomb

Post by Guest »

Ubuntu and its derivatives are the most common / popular, so I would say Linux Mint

Did I mean hamster?
Hamster or gerbil, I couldn't afford either one if I bought those Telsa units
PS
Don't about your neck of the woods but up here in Canada some people keep gerbils as pets, go figure, and why I don't know
DAE avatar
Guest

Re: DGDecomb

Post by Guest »

Is your software compatible with winXP?
And would a Virtual Machine with XP get around the WDDM limitations?
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

Should be, or could be. Maybe at one point I stopped compiling for XP but it would be easy to restore it. The question is what nVidia drivers are still supported on XP.

Yes, it would get around WDDM. Just for fun I'm thinking of putting XP on one of my shed machines and putting one of my 1050Ti's in there, to compare to my win10 system with the 1050Ti.
DAE avatar
Sharc
Posts: 233
Joined: Thu Sep 23, 2010 1:53 pm

Re: DGDecomb

Post by Sharc »

Hi Don
Thanks for the latest DGDecomb with the new deinterlacing modes substituting the former FieldDeinterlace. I did first tests and it works very well here. The show and map options are very helpful.
Just for my understanding: DGTelecide(mode=2) does the same as DGTelecide(mode=3,pthresh=0.0), correct?
DAE avatar
Guest

Re: DGDecomb

Post by Guest »

@Sharc
Thanks for the info on a new DGDecNV

@DG
Win7 XP mode might be usable on Win 8.x and 10 with a slight workaround
So a XP install might not be neccessary
I haven't had a chance to test it yet
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

@Sharc

No, Sir. For mode 2, the whole test of whether a frame is combed is omitted entirely and all frames are assumed to be combed. Then they are adaptively deinterlaced.

mode 2 = FieldDeinterlace(full=true)
mode 3 = FieldDeinterlace(full=false)

BTW, for me DGSource() deinterlacing is so good and so fast that the DGTelecide(mode=2) stuff is only needed when using a different source filter, or if you need the map for inspecting video.

Here's an interesting tidbit. DGSource(deinterlace=2) sustains a frame rate almost twice that of DGSource(deinterlace=1)! Weird, huh? The reason is that the CUVID deinterlacer is invoked twice on the same frame in device memory (using different fields) and it is so fast that two frames are delivered in almost the same time as one.

BTW2, I have been comparing DGSource(deinterlace=2) to QTGMC. For natural video there is no way I can trade 500 fps for 3 fps! I can only make QTGMC appear better with highly artificial test pattern clips. Do you have any natural clips that demonstrate sufficient superiority of QTGMC to justify its pitiful frame rate? The answer will also determine whether I spend any time trying to accelerate it.

@gonca

I don't really want to install a virtual machine. It's very simple for me to swap in a different hard drive and thereby not lose my Win10 install. Why make things more complicated than they need to be just for an experiment?
DAE avatar
Guest

Re: DGDecomb

Post by Guest »

Win 7 had what Microsoft called Windows XP mode
It allowed the user to install and make use of older programs
It was a downloadable exe from Microsoft
This executable would install a XP VM in Windows 7
Sadly, This feature was dropped in Win 8 and above, in fact the exe will refuse to install on anything but Win 7
However
if one downloads XP mode exe from
http://www.microsoft.com/en-us/download ... px?id=8002
Then extracts the files from the executable with 7zip or similar software, a folder called sources will be found
Within this folder there is a file called xpm, which one must use 7zip or other program to extract these files.
The resulting folder has a file called VirtualXPVHD which is the XP virtual image.
Apparently this image can be loaded into Virtual Box to generate a Win XP VM

As I said, I haven't had a chance to test the final result
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

Thanks, but it's too much trouble for a simple experiment, especially when we are unsure of the result in any case. I have two unused PCs and several hard disks in the shed. I also have 5 XP licenses.

I'll perform this experiment tomorrow.

My 20 Tflop Tesla supercomputer arrives tomorrow too. :lol:
DAE avatar
Guest

Re: DGDecomb

Post by Guest »

My 20 Tflop Tesla supercomputer arrives tomorrow too. :lol:
Are you putting that one in the shed
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

Hell no. I put it on my credit card. The ??? hits the fan in 30 days. :roll:
DAE avatar
Guest

Re: DGDecomb

Post by Guest »

Pay off the first card with a second card, and then in 30 days reverse the cards. Just keep moving it from card to card
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

Brilliant, thanks!
DAE avatar
Sharc
Posts: 233
Joined: Thu Sep 23, 2010 1:53 pm

Re: DGDecomb

Post by Sharc »

admin wrote: For mode 2, the whole test of whether a frame is combed is omitted entirely and all frames are assumed to be combed. Then they are adaptively deinterlaced.

mode 2 = FieldDeinterlace(full=true)
mode 3 = FieldDeinterlace(full=false)
Thanks for clarification. Actually I should have asked whether the result of DGTelecide(mode=2) is the same as for DGTelecide(mode=3,pthresh=0.0) rather than asking "does it do the same....".
BTW2, I have been comparing DGSource(deinterlace=2) to QTGMC. For natural video there is no way I can trade 500 fps for 3 fps! I can only make QTGMC appear better with highly artificial test pattern clips. Do you have any natural clips that demonstrate sufficient superiority of QTGMC to justify its pitiful frame rate? The answer will also determine whether I spend any time trying to accelerate it.
QTGMC() vs DGSource(deinterlace=2):
I did not do such comparison yet, so I can't comment. Main reason for me using QTGMC has been that certain footage like my .avi (YUY2) captures can't be decoded with DGSource, and QTGMC() produced very good results (denoising, sharpness, detail retention, suppression of bob shimmer) with "less than perfect" natural video sources. I can live with the few fps of QTGMC for real encoding jobs (running overnight), its low speed is mainly annoying for testing.

Any plans to include a bobbing mode in DGTelecide()?
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

Sharc wrote:Actually I should have asked whether the result of DGTelecide(mode=2) is the same as for DGTelecide(mode=3,pthresh=0.0) rather than asking "does it do the same....".
Yes, the result is the same, but mode 3 runs an extra kernel and associated data transfers so it will be slower. Otherwise I wouldn't have had a reason to distinguish them. ;)
Main reason for me using QTGMC has been that certain footage like my .avi (YUY2) captures can't be decoded with DGSource
Ah, yes. Have you considered an alternative capture utility that can capture to AVC in transport or MP4? AVI is so yesterday.
Any plans to include a bobbing mode in DGTelecide()?
I mentioned it earlier as a possibility. I need to think about design as it interrelates with the whole motion estimation subject, if I want to do it properly. Would you have any use for a simpler version that does not use full motion estimation?
DAE avatar
Sharc
Posts: 233
Joined: Thu Sep 23, 2010 1:53 pm

Re: DGDecomb

Post by Sharc »

Have you considered an alternative capture utility that can capture to AVC in transport or MP4? AVI is so yesterday.
Yes, for example I can capture with x264 vfw in fast mode (--preset verifast) and low CRF (say 10, "transparent") without dropping frames.
The problem is however that I captured these tapes about 10...15 years ago in .avi (Huffyuv), and the quality of the tapes has not improved since, I assume. Therefore I am a bit hesitant to capture the aged tapes again. Anyway, no big deal in using the slow QTGMC on the existing .avi for this case.
I need to think about design as it interrelates with the whole motion estimation subject, if I want to do it properly. Would you have any use for a simpler version that does not use full motion estimation?
I imagine that a simpler version would be good enough in most cases and a nice completion of the DGDecomb function package. There is always the possibility to fallback to QTGMC if need be - as long as it is maintained and remains stable. Whether this justifies to put more efforts into DGDecomb I am not sure though..... Maybe it's something to put under your pillow ;)
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

Well, it's fun to develop this stuff, so why not?
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

@gonca

I can't do the experiment because the last working XP driver does not support 1050 Ti, and I don't have two 7800s in the shed.
DAE avatar
Sharc
Posts: 233
Joined: Thu Sep 23, 2010 1:53 pm

Re: DGDecomb

Post by Sharc »

admin wrote:Well, it's fun to develop this stuff, so why not?
As long as you have some fun with it I would be the last one to stop you :D
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

Ooh, I just had a beautiful Downy Woodpecker feeding at my suet station! I'm going to have to set up my HD video camera to record. I have had Cardinals and lots of Sparrows, of course.
DAE avatar
Guest

Re: DGDecomb

Post by Guest »

admin wrote:@gonca

I can't do the experiment because the last working XP driver does not support 1050 Ti, and I don't have two 7800s in the shed.
Want me to try the virtual box VM to see if it works
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

Thanks for the offer, but I don't think it will matter, unless you have a circa-7800 card that you can test in both environments.
DAE avatar
Guest

Re: DGDecomb

Post by Guest »

Don't have anything that old in my shed
User avatar
admin
Posts: 4551
Joined: Thu Sep 09, 2010 3:08 pm

Re: DGDecomb

Post by admin »

OK, gals and guys, I am going to make an adaptive bobber with EDI or NEDI. That should be excellent quality and remain very fast.
DAE avatar
Sharc
Posts: 233
Joined: Thu Sep 23, 2010 1:53 pm

Re: DGDecomb

Post by Sharc »

This looks promising.
DAE avatar
astrayred
Posts: 11
Joined: Wed Dec 04, 2013 8:55 pm

Re: DGDecomb

Post by astrayred »

admin wrote:OK, gals and guys, I am going to make an adaptive bobber with EDI or NEDI. That should be excellent quality and remain very fast.
Very promising! Finally I don't need to wait a few hours to encode with QTGMC. It's too long even with my Ryzen 1700 :/
Post Reply