DG NV tools

[Guest]

Originally Posted by adiabatic

I tried Beta3 with the problem capture (irregular pulldown) mentioned above. Used Force Film in DGIndexNV then encoded with

LoadPlugin("<path>\DGDecodeNV.dll")
DGSource("SpeedNov03b_de.dgm",deinterlace=1, resize_w=1280,resize_h=720)
colormatrix()

and my 45:07 29.97 fps source video ended up as a 55:42 23.976 fps file... I think I'm misunderstanding what you said in Post #544.

As I said before and as the manual clearly states, if you set deinterlace=1 or 2, then DGSource() forces the field operation to Ignore Pulldown (but the frame rate is left at film rate). It's a wrong setup. To use the GPU deinterlacer you need to have pure interlaced video as source. Then either Ignore or Honor will get you what you need.

For your sample clip, because it has so much pure video and just sporadic soft pulldown, you should use field operation Honor Pulldown and then deinterlace with a filter in your script. There's no point to try to get to film rate with that clip because all the video parts will become jerk city.

I'll try to improve the user manual discussion of all this when the code settles down.

Originally Posted by neuron2

The field operation is applied only when serving through your script. I forgot to mention that.

yup i got it working now. I had an out of date DGDecodeNV.dll in my avisynth plugins directory. Put the latest one in there and everything is working fine now. Cheers.

[turbojet]

Thanks for the force film, worked for me as expected so far. Now that it's implemented do you plan on adding a Video Type FILM/INTERLACED like dgindex has?

Also is there a chance to get information window docking back in DGIndex(NV)?
It seemed to disappear around DGIndex 1.5.0 or so.

[Guest]

Originally Posted by turbojet

Thanks for the force film, worked for me as expected so far. Now that it's implemented do you plan on adding a Video Type FILM/INTERLACED like dgindex has?

I removed that because people always used to to say "my video is interlaced because DGIndex says so". I got so tired of telling people that it just tells how the video is encoded, not how the actual content is, that I just removed it. I'm not totally opposed to restoring it, but let me ask you, what useful information would that give you?

Also is there a chance to get information window docking back in DGIndex(NV)? It seemed to disappear around DGIndex 1.5.0 or so.

Yuck, I hated that. I suppose it could be a configurable option if a few others also would like to have it.

[turbojet]

Originally Posted by neuron2

I removed that because people always used to to say "my video is interlaced because DGIndex says so". I got so tired of telling people that it just tells how the video is encoded, not how the actual content is, that I just removed it. I'm not totally opposed to restoring it, but let me ask you, what useful information would that give you?

I made a typo instead of FILM\INTERLACED video type in dgindex reads FILM\NTSC. It helps determine to force film or not with ntsc dvd content.

Yuck, I hated that. I suppose it could be a configurable option if a few others also would like to have it.

With DGIndex it's not that big of a deal with undocked because the video display was always the same size unless ntsc vs pal. Now with HD and SD video sizes the info window might sit on the video window. Alternative to an option could be 'docking when touched' like ImgBurn and DVDDecrypter does.

Also I noticed DGIndexNV takes 3-4x longer to process then DGIndex with the same video\audio. Is there a chance to speed DGIndexNV up? Maybe by multithreading.

Originally Posted by turbojet

Also I noticed DGIndexNV takes 3-4x longer to process then DGIndex with the same video\audio. Is there a chance to speed DGIndexNV up? Maybe by multithreading.

Id say it is slower cause from what i can see there is more information in the index file. On a 40 minute episode my d2v file is around 300kb where the dgm file is a bit over 1mb. I don't think its that big of a deal though cause you easily make that time extra indexing time back, plus more, in the encoding stage so the overall indexing/encoding is faster with NV.

[Guest]

Originally Posted by turbojet

I made a typo instead of FILM\INTERLACED video type in dgindex reads FILM\NTSC. It helps determine to force film or not with ntsc dvd content.

OK, I'll add that after I get things working for AVC and VC1.

Alternative to an option could be 'docking when touched' like ImgBurn and DVDDecrypter does.

Please explain how that works.

Also I noticed DGIndexNV takes 3-4x longer to process then DGIndex with the same video\audio. Is there a chance to speed DGIndexNV up? Maybe by multithreading.

I tested with the Stuart Little DVD, which I had lying on my hard disk. I got 1:30 for DGIndex and 2:19 for DGIndexNV so I don't know where you get 3-4 times slower. There is a speed penalty for the revised parsing architecture (much easier to understand!) I used in the NV tools. I'll have a look at optimizing it when the basic functionality settles down.

[turbojet]

Originally Posted by olex99

Id say it is slower cause from what i can see there is more information in the index file. On a 40 minute episode my d2v file is around 300kb where the dgm file is a bit over 1mb. I don't think its that big of a deal though cause you easily make that time extra indexing time back, plus more, in the encoding stage so the overall indexing/encoding is faster with NV.

Ya it's not significant just thought that there could be more speed indexing than I was seeing (and I found a way to)

Originally Posted by neuron2

OK, I'll add that after I get things working for AVC and VC1.

Thanks. Also is there or can there be a value in the project file assigned to a frame that signifies film or ntsc?

This is very helpful for TIVTC's d2v parameter when there's some but not all FILM frames.

Please explain how that works.

Tough to explain in words but here's a video example

I tested with the Stuart Little DVD, which I had lying on my hard disk. I got 1:30 for DGIndex and 2:19 for DGIndexNV so I don't know where you get 3-4 times slower. There is a speed penalty for the revised parsing architecture (much easier to understand!) I used in the NV tools. I'll have a look at optimizing it when the basic functionality settles down.

Speeds with DGIndex and DGIndexNV with task manager.

I found out that if I minimize DGIndexNV it finishes the same set of vob's in 2:48 a huge improvement. I tried disabling display while the screen was showing and it made no difference. This is with Windows 7 x64. Any idea why minimizing DGIndexNV has such an impact?

Also a few things I see different between DGIndex and DGIndexNV (by design?):
- Force film still shows 29.97030 fps in DGIndexNV
- Saving a project auto adds .dgm (or .dgv/.dga) even if it's already in the file name in DGIndexNV

After some comparing dgnv resize to lanczos I found out dgnv kind of blurs the image and misses a lot of details. Kind of a bummer since most of the speedup comes from dgnv resizing but is there any chance of changing/improving the resizer that's used?

Here's an example, it's most noticable in the bushes (trees):


DGNV

Lanczos

[nixo]

Hi neuron2

I'm experiencing a little weirdness with dgdecnv200b3. This file indexes to 63 frames but DGDecodeNV.dll only appears to serve 62 frames.

--
Nikolaj

[Guest]

Post the DGA file. Probably you lose one at the end that is nondisplayable due to a bad cut.

[nixo]

It's from the very end of a Blu-ray.

http://www.mediafire.com/file/1ylzzztwi2b/sample.dga

--
Nikolaj

[shortkud]

Possible typo in DGDecodeNVManual.html?

use_pf: true/false (default: false)

Use the progressive_frame indication from NVCUVID. If deinterlace=1 or deinterlace=2, and use_pf=true, then only frames marked as progressive by NVCUVID will be deinterlaced. If deinterlace=1 or deinterlace=2, and use_pf=false, then all frames will be deinterlaced.

Why would it de-interlace progressive frames?

[Guest]

Originally Posted by nixo

It's from the very end of a Blu-ray.

Here is the end of the DGA:

IDR 109472
FRM 7 0
PIC 0
FRM 5 3
PIC 0
FRM 6 1
PIC 0
FRM 6 2
PIC 0
FRM 5 5
SIZ 1920 x 1088
FPS 25000 / 1000
CODED 63
PLAYBACK 63

You can see that the POCs for the last GOP's frames are:

0 1 2 3 5

4 is missing so 5 is discarded.

[Guest]

Originally Posted by shortkud

Why would it de-interlace progressive frames?

Because the flag cannot always be trusted.

[shortkud]

Originally Posted by neuron2

Because the flag cannot always be trusted.

I thought the point of use_pf=true was to use the indication of progressive frames in a way so that only the interlaced frames would be de-interlaced and the progressive frames would "hopefully" be left alone.

use_pf: true/false (default: false)

Use the progressive_frame indication from NVCUVID. If deinterlace=1 or deinterlace=2, and use_pf=true, then only frames marked as progressive by NVCUVID will be deinterlaced. If deinterlace=1 or deinterlace=2, and use_pf=false, then all frames will be deinterlaced.

[Guest]

OK, I wrote progressive when I should have wrote interlaced.

Thanks for pointing it out.

[nixo]

Originally Posted by neuron2

4 is missing so 5 is discarded.

Makes sense. Thanks.

--
Nikolaj

[Turtleggjp]

Originally Posted by neuron2

As I said before and as the manual clearly states, if you set deinterlace=1 or 2, then DGSource() forces the field operation to Ignore Pulldown (but the frame rate is left at film rate). It's a wrong setup. To use the GPU deinterlacer you need to have pure interlaced video as source. Then either Ignore or Honor will get you what you need.

Here's my dilemma:

I would like to use deinterlace=2 for a particular TV show that is pure interlaced, except for a few places during the commercials. When the video is being served, none of these frames with soft pulldown will ever be served (because I edit the commercials out with Trim commands in my script), but because they are in the index file, if the pulldown is not honored it will obviously cause audio sync issues. Is there any way you can make it work in this situation, or would it be just as hard as getting honor pulldown working with deinterlacing in all cases? If it is, I know I can always fall back to another Bob filter in my script, but I really like using the Nvidia one. Thanks!

Matt

[Guest]

Yes, I see the dilemma. I'll see what I can do. It's easy for deinterlace=1 but for 2 there is a gotcha.

[Turtleggjp]

Thanks. They don't seem to be doing that anymore, as all my shows this season do not have any pulldown. Now that my tool to fix audio sync when there are glitches in the stream is almost done, I can finally start working on all the shows I have recorded from last year...