1080/60i to 720p (for AVCHD Disk) and 480i (For DVD)

[multiAVCHD]

HD DVD muxing requires .264 + .ac3 file. I believe multiAVCHD converted the LPCM audio to AC3, right?

A log from such compilation will be useful, too.

[greymalkin]

thanks again I'll grab the log next time (or does it save it somewhere in a .txt file?)

I also unchecked the option to clear the _TEMP folder so if it should happen again I won't have to re-encode them! .

Thanks again for all your help. This tool is going to replace a lot of other peices of software for me!

[multiAVCHD]

Yep, the log is saved under _LOG\ folder and in the output folder as multiAVCHD.log.

[greymalkin]

ok I'm finally getting back to doing this..I used multiAVCHD to make my AVCHD disk and I'm just going to have it show a "welcome screen" instead of a menu and then just go into the first chapter, then 2nd chapter.

For the DVD I'll make an .avs file to just resize to 480i60. then feed it into HCEnc, set the AR flag to 16:9 and pray for the best.

The only reason I was trying 480p is because on my upscaling toshiba A3 HD-DVD player it was showing interlace lines even though it's supposed to be doing an upconversion..it seems like "budget" dvd's still show interlace lines while bigger production movies don't show the lines when the upconversion is taking place?

Also, a lot of the AVISynth scripts I see show convterttoYV12..isn't this a colorspace conversion and isn't that what I don't want to do? Can I keep it in YUY2 the whole time? and which deinterlacers/etc. would allow me to do that? I just want to open the edited CanopusHQ AVI, deinterlace (using something that doesn't force a colorspace conversion), resize, then interlace keeping it @ 60i.

[multiAVCHD]

x264 encoder requires YV12 colorspace for AVC/H.264 encoding.

[manono]

The only reason I was trying 480p is because on my upscaling toshiba A3 HD-DVD player it was showing interlace lines even though it's supposed to be doing an upconversion..it seems like "budget" dvd's still show interlace lines while bigger production movies don't show the lines when the upconversion is taking place?

None of them should show interlacing. Unless you resized the video improperly, which from the sound of it is what you did. You can't just resize interlaced content any old way you might like. You have to resize in an interlace aware manner.

Can I keep it in YUY2 the whole time?

HCEnc requires YV12 input.

[greymalkin]

that very well may be the case..and many of the "budget" dvd's we've gotten from the library/etc. were probably also resized/encoded wrong as well..I did check the "interlace" box when I resized in virtualdub and I'm aware of the deinterlace/resize/interlace process...but I just haven't found the right sequence/tool/etc. that gives me a proper mpg/mpv file. Once I get a process that works I'll never deviate from it for the rest of my life. Now allow me to further embarrass myself by attempting to put together the entire AVISynth Script

AVISource("<path><CanopusHQ1080i60filename>")
ConvertToYV12(interlaced=true)
tempgaussmc_beta1() <-- must download/add plugin
LanczosResize(720,480)
AssumeTFF().SeparateFields().SelectEvery(4, 0, 3).Weave()


stick that baby in notepad and save as an .avs file.
feed to HCEnc, set AR to 16:9
end up with a 480i60 mpv file that isn't slowed down, sped up, hideous, blown out...etc.
dance dance dance

Once I get it to a useable MPG the rest ofthe authoring/etc. process is an old hat.

[greymalkin]

some day my prince will come...

so I tried loading the script in HCEncGui and it basically said I don't know what EEDI2 is..so I did some looking and found there's 8000 dependencies (give or tak 7995) for TempGaussMC_beta1 to work..so I downloaded/installed those...got an stlport.dll error on the next import..spent some more time digging and finding there was something wrong with the version of mvtools I had...then finally loaded again and get "there is no function named "TempGaussMC_beta1". Now all the other plugins are .dll files..but the TempGaussMC_beta1 is an .avs file. Does that sound right? This is about as difficult as installing anything in linux (when the package manager isn't working) .

[manono]

In addition to all the supporting DLLs you need, you'll probably have to import the TempGauss script itself:

Import("C:\Path\To\TempGaussMC_beta1.avs")

[greymalkin]

wow this is absolutely ridiculous! I used import to load the avs for tempgaussmc_beta1..then went through this plugin missing after that plugin missing, then this argument for this plugin doesn't exist...and theres many many versions of each plugin. Is there anywhere where some kind soul packaged them all together? or does everyone just go easter egg hunting until they find the magical combination?

[manono]

Have you seen this already:

http://avisynth.org/mediawiki/TempGaussMC

[greymalkin]

yep..unfortunately it doesn't contain links to the right versions. I've tried several different "flavors" of the tempgaussMC .avs(i) file and many versions of the plugins and it looks like it's just a free for all when it comes to writing these things.

[greymalkin]

wait...wait!!! the stars have aligned..I have made it through the lost woods....and the thing is finally working... I mvtools.dll instead of mvtools2.dll. I'm surprised noone out there has bundled these together to save the hours of frustration.

[greymalkin]

well at 10 hours it says I'm 8% done with pass1 :P...I did say I didn't care how long it takes but it looks like the whole production would take well over a week...so all that work to switch back to yadif :P.

[txporter]

Here are some suggestions to speed up your encoding:

avs file:

#setmtmode(2,0)   # uncomment this line if you have avisynth 2.5.8MT and have a multi-core processor)
LoadCPlugin("<path>\Yadif.dll")
Import("<path>\TempGaussMC_beta2.avs")   # or change it to a .avsi file and drop in your avisynth 2.5\plugins directory
AVISource("<path><CanopusHQ1080i60filename>")  # I am pretty sure that avisource works with setmtmode2. If you are getting crashes, change top line to setmtmode(5,0) and add setmtmode(2,0) below avisource call
ConvertToYV12(interlaced=true)
LanczosResize(720,height)
tempgaussmc_beta2(1,1,0,0,0,0,edimode="Yadif",Smode=1,SLmode=1,Sbb=0,SVthin=0.0,pelsearch=1) #  (see plugin download list below)
LanczosResize(width,480)
AssumeTFF().SeparateFields().SelectEvery(4, 0, 3).Weave()

TempgaussMC_beta2 - download by following link in this thread.
Download latest masktools v2.
Download VerticalCleaner.dll (also linked in thread with link to TempgaussMC_beta2.avs file)

Drop all of the necessary plugins into your Avisynth 2.5/plugins directory so they autoload (Yadif.dll doesn't autoload because it is a C plugin).

If you have avisynth 2.5.7MT or 2.5.8MT (and a multicore processor), you can uncomment the setmtmode(2,0) line to get a bit more speed improvement.

[greymalkin]

hi txporter..thanks for the super optimized script! I did hear about resizing it vertically before the deinterlace to try and speed things up a bit..I'll get the right plugins and try this out. I tried plain old yadif but wasn't happy with the results..I think I won't be until I have the whole package..I'm learning there's a lot more to a good deinterlace then I originally thought :P.

I do have a quad core machine and it's frustrating to see it sitting at 23% the whole time it's encoding so I'll definitely use the lines for Multithreading. In a moment of weakness I tried AVStoDVD which did a pretty good job but my first pass it has some blurring/ghosting and then I went in and used it's deinterlace options which adds the leakdeint(mode=0) filter in the script for me..it seemed to fix the blurring/ghosting during pans but there are some jaggies every once in a while..so still searching for the golden ticket.

[txporter]

Well...you might see some additional CPU utilization, but TGMC is a pretty slow filter and you likely not get near 100% utilization.

Actually you need to NOT resize vertically before you de-interlace. You can resize horizontally safely though.

I agree that there is a lot more to deinterlacing than I thought as well. The most important part of TGMC (to me) is the reduction/elimination of the bob shimmer. If I do a straight yadif(mode=1,order=1) deinterlace, you can see shimmering in stationary images. Whatever magic that Didee performs in TGMC eliminates these. Even with the drastically reduced settings that I posted above. (see the stockholm video to see what I mean by bob shimmer) I did samples with all of the various edimodes in TGMC ("--" is a simple bicubic resize, as well as Yadiff, EEDI2, NNEDI, and nnedi2) and I found that Yadif (in my eyes) leads to less blurring of the image and I found that I like the results better. You may want to cut a small sample from your original source and test out a few settings before letting the whole thing run.

If you open up the tempgaussmc_beta2.avs file in notepad or something similar you can see the documentation.

The defaults for the string that I posted above are the following:

TempGaussMC_beta2(2,2,1,4,0,4,edimode="EEDI2",Smode=2,SLmode=2,Sbb=1,SVthin=1.0,pelsearch=2)

I will try to put together a speed/setting table based on a 1080i conversion of a PBS HD video (Story of India). I will eventually run the 25s through the default TGMC filter just to get the numbers (but it will likely take a LONG LONG time to run it).

Here is the avs file that I used with the PBS video (it is PAL video that was field blended to NTSC, so I am undoing that):

setmtmode(5,0)
MPEG2Source("Path\To\Story.of.India.E1-clip.d2v",cpu=3)
setmtmode(2)
lanczosresize(1280, height)
TempGaussMC_beta2(1,1,0,0,0,0,edimode="tdeint",Smode=1,SLmode=1,Sbb=1,SVthin=0.0,pelsearch=1)
srestore(frate=25)
lanczosresize(width, 720)
removegrain(1)

The following fps numbers are for a dual core laptop, so a quad core desktop should be quite a bit faster.


edit: Forgot to mention that I updated my TGMC script to use tdeint instead of Yadif for the deinterlacer. This is mainly because I use the 64bit avisynth at home and Yadif wasn't recompiled for 64bit environement yet. The results between Yadif and tdeint are very similar. Also note, that case sensitivity is important with edimode in TGMC. Look through the script and use the edimode exactly as written there (EEDI2 rather than eedi2), otherwise it will default to using bicubic for the edimode (same as "--").

Commentary post runs:

You can see that there is a very large difference in conversion rate between the default settings and some of the reduced sets. The biggest knobs in performance seem to be MDegrain temporal filtering and repair (3rd and 6th number in the parameter strings), the temporal radius for the temporal gauss after motion compensation (2nd number) and the edimode.

This is what I can gleam from reading through the TGMC thread on doom9 (called "exist any filter specially done for (deinterlace) flicker") and reading through the script itself
**anyone that knows better please correct me where I go astray**
- the first 3 parameters (defaults @ 2,2,1) set the temporal gauss filtering for the script. These are responsible for the reduction/elimination of the flicker/bob shimmer. Some sources will need more or less filtering to reduce the shimmer.
- the next 3 parameters (defaults @ 4,0,4) are the repair parameters used to combat artifacting
- edimode sets which program/plugin is used for interpolating the missing pixels. I don't know what each one brings to the table as strength/weakness compared to see each. I have read that EEDI2, NNEDI, and nnedi2 are better at connecting up straight lines whereas bicubic and Yadif (and tdeint) are much faster.
- Smode sets the method used for re-sharpening the motion compensated temporal gauss
- SLmode sets the method used to limit the re-sharpening
- Sbb is defined as the back-blending of the gauss-blurred sharpen difference (this sounds like another way to limit the sharpening???)
- SVthin can be set to thin out horizontal edges that are broadened with the interpolation
- pelsearch can be toggled between 1 (fast) and 2 (accurate). This defines the running of the motion vector tools.

P.S. There are main more parameters that can be toggled in the script, these are just the ones that I chose to play with.

[Gavino]

Also note, that case sensitivity is important with edimode in TGMC.

I don't think that's true, since string comparison is case-insensitive in Avisynth scripts/functions.

[txporter]

You could be right. I thought I read in the TGMC thread on doom9 that folks were complaining about the fact that they need to use NNEDI or nnedi2 and were asking for the script to be adjusted to the same case. Maybe I read something into it that wasn't really there...

[Gavino]

I thought I read in the TGMC thread on doom9 that folks were complaining about the fact that they need to use NNEDI or nnedi2 and were asking for the script to be adjusted to the same case.

See Didée's reply that no change was necessary.
Also posts #387-389 lower down the same page.

[txporter]

Thanks, for posting that Gavino. That was the portion of the thread that I had seen. Apparently, I need to work on reading comprehension though!

I updated the post above with the finished table to show TGMC speed in fps (on a dual core laptop) for a 1080i video encoded with x264 (1080i --> 720p run through srestore to restore PAL framerate) for various setting changes.

[greymalkin]

oh boy! I'll start with a small clip and the first script. The deinterlacer AVStoDVD uses is working OK for side to side motion, but whenever something moves up and down quickly I see distortion..like someone holding an open binder with a nice sharp edge and it's sort of wobbling up and down and stairstepping pretty badly..or people walking with a bounce in their step and their face is distorting slightly. Also when people take a picture with a flash in the video it seems like one section of the screen flashes before the other..weird.

I think another part of my problem is the bitrate I'm using when I actually encode the DVD. There's a lot of trees and leaves blowing in the wind. I tried to get by using 7Mbps so I could fit it on a single layer DVD, and on my upscaling dvd player it looks pretty good (except of course for the stairstepping on vertical motion), but when I played it through my regular S-Video connected DVD player on the same TV (47" LCD) it looked absolutely horrible..like a bad youtube video..I didn't think upscaling dvd players could fix the video that much! So I'll be maxing out the bitrate and just use dual layer DVD's. The "bad" thing about the HD camcorders is everything is in razor sharp focus and when everything's moving it makes it very difficult for the poor MPEG encoder...I guess that's why it's good sometimes to have that depth of field. It was so much easier getting from my SD camcorder to DVD! Now it's easy getting to AVCHD Disk with this camcorder..I've even wondered if I could just play the AVCHD disk and output to a capture card, letting the hardware scaler do the work for me..it shouldn't be this difficult!

[txporter]

If you are interested, I decided to keep working on this a bit to make sure that I was happy with the method that I chose for deinterlacing. I am playing around with a SD clip rather than HD since it speeds things up. I have done a few stacked comparisons of a 2min Torchwood clip (again NTSC release, so need to deinterlace and then srestore). In the following link, you can find 3 stack comparison files:

stack-deint_compare.mp4 - comparison between minimum TGMC tdeint settings, tdeint, yadif and tdeint+tmm
stack-tgmc_edimodes.mp4 - comparison of TGMC edimode at default temp gauss & repair settings: tdeint, EEDI2, nnedi2, bicubic
stack-tgmc_tdeint_settings.mp4 - comparison of various TGMC settings for tdeint edimode (basically default --> minimum set posted above)

http://txporter.tonidoid.com:10001/app/websharepro/share/Deinterlace/

If you look at around 14-15s where the aerial pan looks over the water in the bay and again at ~25s where there is a vertical zoom-out from the dead dude, this is where I was seeing the most evidence of flicker. I really don't see any flicker on any of the TGMC runs. And I frankly can't note any major/minor difference between the various edimodes (although I didn't step through them frame by frame or zoom in on them). Any comments on what to look for or things noted are appreciated.

[greymalkin]

I finally ran the first script you provided but couldn't use the MT plugin for some reason. I installed the 64 bit version and put the MT plugin in the 64 bit plugin folder.. I even tried using loadplugin/loadcplugin to point to it but no luck..

anyways...the script kicked off and averages around 3fps even without the MT goodness which is a big step up..it was still running this morning so I'll have to check it out tonight when I get home.

earlier I had tried the hd2sd package someone created where they provided a function and all necessary plugins to make it work for AVISynth..it worked OK but still suffers from the same jagged lines issue during vertical motion..I went ahead and bumped the bitrate up to 9500 and I think it needed it as I noticed the VBR getting up to 9300 at times.

If this doesn't work I'm wondering if theres some fundamental issue with the field order or some other parameter I've never had to worry about in the past.

[txporter]

Did you install 2.5.8MT (doom9 thread) and then install the 64bit version of avisynth2.5.8MT? Setmtmodes should work for you then. Odd.

Can you upload a sample of what you are trying to convert?

[greymalkin]

I definitely installed the 54 bit version of Avisynth 2.5.8MT. It was just some .dll files that I right clicked on and installed...then it told me the 64 bit version was installed and somewhere in the readme it told me I now had a 64 bit plugins folder...but I believe in this package there is an MT.dll that I copied to the 64 bit plugins folder..I figured the plugin bundled with the update itself should have been the correct one but I'll download the other one. I don't understand why someone who goes through all the trouble of making these .avs/.avsi files doesn't bundle the plugins necessary for it to function .
oh and if this render has the same problem I'll post some problem clips in a few days!

[txporter]

I am not 100% certain, but I thought I had seen that you needed to have the 32-bit version of 2.5.8MT avisynth installed and working before installing the 64bit with the installer from the 64bit thread.

As far as the functions being bundled, it is really too much of a hassle for scripters to keep up with. Many times the plugins used by the scripts are not built by the same person. It is tedious to track down new releases, bug fixes, etc for the various plugins. All they really ought to do is include the dependencies. Then the user can get the needed plugins (whether you used 32bit or 64bit or avisynth 2.58 or 2.60, etc).

[greymalkin]

...As far as the functions being bundled, it is really too much of a hassle for scripters to keep up with. Many times the plugins used by the scripts are not built by the same person. It is tedious to track down new releases, bug fixes, etc for the various plugins.

I agree 100% it's a hassle to keep up..especially for those of us trying to download everything required. At least the person who wrote the scripts already has the necessary plugins/versions unless they wrote it without ever testing it.

As far as finding the new releases, etc. again I agree that is a tedious chore...even more so for joe average , and I wouldn't expect anyone to continually re-package all the new plugins, but I think it would make sense to at least package the plugins you used when you wrote the script so it functions.

...All they really ought to do is include the dependencies. Then the user can get the needed plugins (whether you used 32bit or 64bit or avisynth 2.58 or 2.60, etc).

yeah, I'm not trying to say they ought to give them to me...I know they don't really owe me anything and I am in their debt for the work they've already done...I just think it would be a relatively small effort for them to bundle the right combination of plugins/etc. that they already have in their posession vs. days of trial and error and crying on a forum for me .

oh and finally I realized reading this you might take this as me saying YOU should have given me the plugins..that's not what I was trying to say at all...you gave me the links for the correct ones and I'm thankful for that..I just didn't do something right and I'm sure if I practice my reading comprehension I'll get it working :P

[greymalkin]

ok I uploaded 2 small clips. The first one is using the hd2sd pre-packaged avisynth script/filters with no additional parameters. This file seems to do fine on side to side pans, but you will notic crawling lines on up/down motion

unfortunately the next one which is the tempgaussmc script from here suffers from both the horizontal and vertical interlace issues...this is what makes me believe there is something I'm not telling avisynth to do to the video or something about the way I'm encoding it...as far as I know everything detects it as 29.97 TFF but after that I'm not really sure what to do to make it not look like this. The avchd file I made with multiAVCHD looks perfect with no jagged lines or blurring pans at all..the mpeg file was made with the same source canopusHQ file and also set to interlaced/tff...

hmm..OK I uploaded the 2 clips but I don't see how to attach them to this post...when I go to attach a file it asks me for a web address? now I went back to upload files/manage attachmens and the files are gone! I don't see any option to add them to this post.

[poisondeathray]

If it's >30MB it will be rejected. You can use a free hosting site e.g. mediafire.com, sendspace.com