PBS HD - Story of India conversion

[txporter]

I have 6 episodes of The Story of India that I recorded a little over a year ago. They are 1080i MPEG2 files recorded off of the PBS HD network. I want to compress them with x264 to save a bit of space. I want to make sure that I have analyzed them correctly before I get rid of the originals. They appear to be 25fps material that has been fieldblended to 29.97fps. I am using the following script to convert them:

setmtmode(5,0)
MPEG2Source("Story.of.India.E1-clip.d2v",cpu=3)
setmtmode(2)
lanczosresize(1280, height)
tdeint(mode=1,order=1)
srestore(frate=25)
lanczosresize(width, 720)
removegrain(1)

I initially meant to convert these using TGMC as I have been doing with DVD PAL video blended to NTSC. However, I was getting <0.1fps on my Q6600 for conversion rates and I don't know that I really want to spend that much time converting them. The above script is my compromise and I think they look ok, but it seems like I still have some blends left afterward. Also, the source material isn't great. Perhaps I need to add some denoise in here? The deblocking from dgdecode does seem to help a bit (although I am not sure it is OK to use with interlaced material??).

Can someone confirm that srestore'ing to 25fps is the correct thing to do on this clip? Any other ideas to improve overall quality without bringing my Q6600 to its knees? Thanks.

Clip: http://txporter.tonidoid.com:10001/app/websharepro/share/India/

[edDV]

The 1080i/25 fps original was probably electronically interpolated to 29.97 fps. There is no direct way to reconstruct 25 fps.

The local PBS station here rebroadcasts the 1280x720p feed at 8-12 Mb/s VBR. A typical 55 minute PBS show fits nicely on a DVD-5 without conversion.

[jagabo]

I would try using Yadif() instead of TDeint(). Otherwise your script looks ok. SRestore() should do a decent job of restoring the original 25p frames (I'm downloading the clip now).

Later... Yes, SRestore() worked well and it looks like the source was 25 fps.

[edDV]

I would try using Yadif() instead of TDeint(). Otherwise your script looks ok. SRestore() should do a decent job of restoring the original 25p frames (I'm downloading the clip now).

Later... Yes, SRestore() worked well and it looks like the source was 25 fps.

They shot this with Sony HDCAM @ 1080i/25.

The first step of standards conversion would be deinterlace to either 25p or 50p.

[jagabo]

They shot this with Sony HDCAM @ 1080i/25.

The first step of standards conversion would be deinterlace to either 25p or 50p.

They must have deinterlaced to 25p then. The clip he posted had the common field blending to go from 25p to 30i. Unfortunatly, he won't be able to get back the smoother motion of 25i.

[txporter]

Thanks for looking guys. Can I ask, why yadif over tdeint?

Also, is there a way to get rid of the bob shimmer without processing with TGMC? Or maybe which part(s) of TGMC are responsible for eliminating/reducing the shimmer? Maybe I can turn off a bunch of stuff to speed it up but still fix that.

[jagabo]

Speeding up TGMC: https://forum.videohelp.com/threads/315286-Best-Method-of-De-interlacing-Mini-DV-AVI-files?p=1982161&viewfull=1#post1982161
and a few posts later: https://forum.videohelp.com/threads/315286-Best-Method-of-De-interlacing-Mini-DV-AVI-fi...=1#post1992195

[txporter]

Thanks, jagabo. I saw some of those threads earlier but I guess I forgot about them. Will test out those TGMC settings.

[txporter]

Ok, tested what I use for DVD NTSC back to PAL conversions (I added tdeint to TGMC because I am using Avisynth64):

TempGaussMC_beta2(1,1,1,edimode="tdeint",Smode=1,SLmode=1,Sbb=0,SVthin=0.0)

...and I am getting around 0.05fps.

I then tried what you suggested (mostly):

TempGaussMC_beta2(1,1,0,0,0,0,edimode="tdeint",Smode=1,SLmode=1,Sbb=0,SVthin=0.0)

...and I am seeing around 1.43fps. Its slow, but not your-newly-conceived-kid-will-see-the-light-of-day-before-this-conversion-will slow. Thanks!

I tested it with the clip first and even the greatly reduced TGMC seems to get rid of the bob shimmer, so I am happy!

[txporter]

Ok, did a few more tests with TGMC on the India clip. Trying to find any other possible speed improvements. These were done on a dual core laptop rather than my Q6600 at home, so the numbers have changed.

Repeat of settings that I posted for the Q6600:

TempGaussMC_beta2(1,1,0,0,0,0,edimode="tdeint",Smode=1,SLmode=1,Sbb=0,SVthin=0.0)

0.70fps

TempGaussMC_beta2(1,1,0,edimode="tdeint",Smode=1,SLmode=1,Sbb=0,SVthin=0.0)

0.42fps (repair functions added back in)

TempGaussMC_beta2(1,1,0,4,0,0,edimode="tdeint",Smode=1,SLmode=1,Sbb=0,SVthin=0.0)

0.64fps (MDegrain repair seems to be the speed hit)
Changing the 3rd 0 (MDegrain temporal gauss filtering) to a 1 (settings that I use for DVD conversions) brings the fps to <0.1fps

TempGaussMC_beta2(1,1,0,0,0,0,edimode="--",SVthin=0.0,pelsearch=1)

0.70fps - recommended fast settings from other thread

TempGaussMC_beta2(1,1,0,0,0,0,edimode="--",Smode=1,SLmode=1,Sbb=0,SVthin=0.0,pelsearch=1)

0.87fps - incorporation of faster sharpening modes

TempGaussMC_beta2(1,1,0,0,0,0,edimode="tdeint",Smode=1,SLmode=1,Sbb=0,SVthin=0.0,pelsearch=1)

0.77fps - tdeint vs bicubic

All of these settings look much better for bob shimmer than running yadif or tdeint alone. I didn't pull out stills on this source, but from previous work I did, I preferred yadif (or tdeint) to any other mode for sharpness of the scenes (bicubic, EEDI2, nnedi, nnedi2 all seem to blur out the image more). Many folks report much better results for diagonal lines using one of the more advanced edimodes, but I didn't notice any particular problem with just yadif/tdeint.

[jagabo]

Nice bunch of tests. Another thing to keep in mind is that the success of the deinterlacers varies depending on the type of material. Results with anime can be very different than results with live video.