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

[greymalkin]

thanks...both of them add up to 26mb..I'll try the free hosting places.

[greymalkin]

OK here's the files:

hd2ds file: has crawlies/jaggies on high contrast sharp lines when camera moves up/down, but side to side motion seems correct.
TempGaussMCscript: has problem from hd2sd file with addition of blurriness on side to side motion and also seems to be jerky on top of the side/side blur

http://www.mediafire.com/file/zw2htewfkyq/hd2sdNoExtraFilters.m2v
http://www.mediafire.com/file/emgwmzlldho/TempGaussMCscript.m2v

Thanks again for everyone who's helped me with this so far..It will definitely be a good learning experience for the future!

[poisondeathray]

how are you watching/examining these?

There was a field swap somewhere, the tgmc was BFF, the hd2sd was TFF

I'm not sure about quality of the canopus intermediate or the original avchd file, but both the sd MPEG2 encodes seem a bit low in quality to me from an HD resize - I wonder if the encoder has something to do with it? The metadata says womble, which is know to be a poor mpeg2 encoder (at least it has been in the past, not sure if they have improved it)

you mentioned vertical panning, but I didn't see very much in the sample clips (maybe a tiny bit at the end), but it was mostly horizontal ?

Yes, the TGMC version is more blurry, which is to be expected if you used default settings. You can adjust the denoising strength.

But there is something else wrong with the motion in the TGMC version, it's more apparent at the last section where everyone is seated, or the pan by the tree trunks. If you apply a bob deinterlacer and step through, you will see there are a few frames that are almost repeats; something very screwy going on there. The horizontal pan on the TMGC version is almost like a big jump, then little jump, and looks something like a blend deinterlace in between. I suspect something went wrong, as it should have been TFF instead of BFF. The motion is better with the HD2SD version (each bobbed frame looks discrete, and the distance between each frame is even).

There is also something wrong with the levels, the black appears to be a bit high (so a bit washed out) , and there are blowouts on the TGMC version which are attenuated on the HD2SD version. I'm not sure if it was fed to the encoder like this, or if the encoder screwed it up?

Can you upload a corresponding section of the canopus intermediate you are working with? Then we can gauge what the source quality you are starting with, and if there is anything else going on

[greymalkin]

thanks poison..

i'll post the original avchd clip and the canopus intermediate later today. as far as I could tell it looked good...but as you say every MPEG encode so far looks like a bad low bitrate file although I'm using almost the max I can..

as far as the mpeg encoder used..I fed my scripts to HCEnc with a 9500kbps VBR. i set the GOP length to 12 (although I know the max should be 18) and set it to interlaced/TFF. I set the quality to best...AQ to 2 and LumGain to 2..

Womble was only used to cut out the peices i uploaded using stream copy..it did not encode either of the videos. OK...back later with the original and canopusHQ files!

oh and I'm pretty sure it's the script/mpeg encoder where i am failing..as the avchd disk generated from the same canopusHQ file using multiAVCHD looks great! It's a litlte softer but I did have to re-encode to 13mbps to fit on a 8gb dl dvd and of course it had gone through 2 generations of encodes. anyways..none of the jerkyness or artifacts are present in the avchd disk

[greymalkin]

ok..here's the original file followed by the avchd disk I made with multiAVCHD.

http://www.mediafire.com/file/xv2tgrdfdez/New Sniper Project.ts <- needs to be pasted in address bar?
http://www.mediafire.com/file/m4menetnloy/AVCHDDiskFinal.ts

I used the TS Sniper tool to cut out a peice. Oh..and to answer the question how was I viewing them..I was viewing them on both my computer, my toshiba A3 (upscaling) and a pioneer s-video connected dvd player connected to my 47" samsung. the MPG files looked bad on all of them.

The canopus HQ file is 316mb for the same scene which is above mediafire's limit. I tried winzip but it did a terrible job. Is there a special video data compression utility? or I could cut it in half.

[poisondeathray]

you can use 7-zip or winrar to do split archives (so other people can re-assemble them when downloaded)

or use a free site like megaupload.com that allows larger filesizes

[manono]

It looks to me that AVCHDDiskFinal.ts was encoded as BFF (Info() gives that information), and it plays jerky with the fields separated. But it's really supposed to be TFF because putting AssumeTFF in the script before separating the fields straightens it out again. And the whites are badly blown out, both in the source and in the encoded result. Fixing that will help somewhat with the compression.

[greymalkin]

manono...thanks for weighing in . yeah I was debating on lowering the exposure setting on the camera but made the wrong decision and left it alone!

as far as it becoming BFF I'm not sure how that happened..when I exported it out of neo 2.5 I had it set as interlaced/TFF..but I don't see any field order settings in multiAVCHD so I assumed it just grabbed the info from the source file and set it to the same thing...

What do you mean by it plays jerky with the fields seperated? So I'm taking it even the AVCHDDisk version I thought was OK (except for the whites which I will conquer later) is bad.

So far it sounds like I need to go back to the neo2.5 project, try to fix the whites somehow (I'll figure out that white balance filter!). Then re-create the AVCHD Disk and of course the DVD from there AFTER I figure out how I got switched the BFF. I'm hopin the Canopus HQ file will answer the question...It's going to take my DSL connection several hours to upload I'm sure so it will probably be tomorrow before I can post links.

[manono]

It might really the MBAFF encoding that's causing the problem, and maybe AviSynth is just using the default BFF in the absence of anything definitive. However, adding AssumeTFF to the script before separating the fields seems to fix it. I don't know if you're using scripts throughout this process and I know nothing about Neoscene.

What do you mean by it plays jerky with the fields seperated?

Separating the fields is a good way to check if the field order is correct.

[poisondeathray]

It might really the MBAFF encoding that's causing the problem

I don't think MBAFF is the issue. x264 uses MBAFF syntax, but the macroblock adaptive part hasn't been implemented yet (the end result is like PAFF), and recent decoders have no problems with it. Old libavcodec >1 year ago did decode with errors (DGAVCIndex can decode with errors). If you have a current ffdshow build, for example it should decode fine. Recently I've been testing x264 interlaced encoding quite a bit, and am quite familiar with it - so I suspect the errors are from something else. It might be from where/how the file was cut

As I said above, for the MPEG2 version, the TGMC was BFF, but the HD2SD version was TFF (determined by separating fields and stepping through) , and the TGMC showed the similar weird jerkiness. So something was different in their processing; the TGMC version should have resulted in a TFF file not a BFF file. Both showed blowouts, but the HD2SD version was not as bad.

The AVCHDDiscFinal.ts also has some weird alternating chroma bleeding when you separate the fields (or even when you bob it for playback) . For bobbed frames # about 290-320, have a look at the green rain drain pipe it alternates colors, as well as the lady's pink shirt, or the bald guy's (sitting down) head. They alternate colors (pink=>white=>pink=>white)

For levels you can use histogram() or videoscope() to see what you are actually feeding into the h.264 or mpeg2 encoder (ie. open up the export from edius). It might be canopus HQ's fault, because there is a RGB=>YUY2 conversion, and you can specify IIRC either full range or Rec.601 16-235

[greymalkin]

hmmm...the NewSniperProject one was directly from the camera.

So just to recap my workflow..I imported the m2ts files directly from my camera into edius neo2.5..edited and outputted to this canopusHQFile which SHOULD be TFF.
HD Version: Imported canopus file into multiAVCHD..set it for Bluray/AVCHD Disk and set bitrate to 13Mbps. I didn't see or change any field order settings.

SD Version: loaded canopus file via avisynth script into HCEnc using the hd2sd pre-packaged function/filters and also tried the tempgaussMC script provided in this thread. Set bitrate to 9.5Mbps and a few other tweaks and both mpg files as you saw had some interlace related issues, color issues (whites blown out) and overall just looked like a bad encode (artifacts, etc.) even at 9.5Mbps

So to recap..the field order on the hd2sd version is correct, but I still need to go back and somehow fix the blown out whites to try and get a better MPEG encode. I will try to do that in neo2.5.

Everything after that seems to be an interlace/field order issue...so maybe I should do what you originally proposed and just make a 1080p master...then do a straight resize to 480p with some blurring and encode it at 9600kbps all the way through?

I'm messing with the YUV curve filter right now...it's way over my head :P. One other puzzling thing is I import the file as 59.94i, but everything after that sees it as 29.97

[poisondeathray]

the NewSniperProject one was directly from the camera.

Now that is weird. Maybe something is wrong with my decoder. I'll do some tests tomorrow, swap a few decoders and report back, but I definitely see the alternating color issues. It might be from how it was cut. EDIT: SORRY, it's the final version, not the NewSniperProject, I mixed them up.

For the SD: If you used the TGMC script earlier, it should have resulted in a TFF file. So something went amiss somewhere.

Check the canopus export levels with histogram() => because that is what multiavchd or hcenc or any encoder is "seeing" and will pass on to the encode. Often the problem comes when exporting out of the NLE because of RGB=>YUV conversions, so what you see in the NLE is often different than would you get. Also what you "see" on the PC monitor is usually different than what you get on TV, unless you have it calibrated specifically for TV levels. This is where waveform monitors, histogram come in handy to see if your video is "legal"

I'm not sure what's going on TBH. These examples your have provided are acting very weird compared to other AVCHD projects I've done.

One other puzzling thing is I import the file as 59.94i, but everything after that sees it as 29.97

59.94i is the same thing as 29.97i. The first is fields per sec, the second is frames per second, but mean the same thing just different syntax

For the histogram() filter in avisynth, it's just layed on the side vertically compared to ones in most editors. Areas in brown are illegal . The left side is 0-15, the right side is 236-255, the middle black area is video legal range 16-235

[greymalkin]

here's the final peice of the puzzle! I'll do the histogram and feed it into virtualdub to see if I can make sense of it. thanks again!

http://www.mediafire.com/file/izmzt5nhdmz/CanHQFine.part1.rar
http://www.mediafire.com/file/nnjowzxfjye/CanHQFine.part2.rar
http://www.mediafire.com/file/znmnlyl2lzy/CanHQFine.part3.rar
http://www.mediafire.com/file/zvdjmkeiztw/CanHQFine.part4.rar


EDIT: I checked the histogram and theres a lot of "smoke" in the right as was suspected . Can a basic color correction or white balance fix..err..let me rephrase..help minimize this?

[poisondeathray]

EDIT: I checked the histogram and theres a lot of "smoke" in the right as was suspected . Can a basic color correction or white balance fix..err..let me rephrase..help minimize this?

you can't "fix" it per se, in that those areas are blown out (i.e there is little detail to recover). You can bring the levels back to "legal range" with a levels filter, either in your NLE or avisynth ; but this is largely a shooting problem. Notice your black level is high too. If you bring it back down to ~16 you will notice the picture "less washed out." A balanced color correction has most of the data in the range 16-235 (with small excursions). Some higher pro level camcorders can record detail in the 235-255 range, and more detail is recoverable, but this is rare for consumer level camcorders

[greymalkin]

OK..I found this little gem that brings the levels out of the brown: ColorYUV(levels="PC->TV")
looking at the histogram it kind of looks like a "hard limiter" for white levels .
Then I can lower the brightness to bring the black levels back down.

I'm doing something like this
loadplugin(".....yadif.dl")
AVISource("MyJackedUpVideo.avi")

AssumeTFF()
ConverttoYV12() <-- I need to look at my script again..If it's here I need to add interlaced=true!!!
ColorYUV(levels="PC->TV")
Tweak(brightness=-20) <-- don't remember the exact syntax, going from memory here...
yadif(mode=1)
SelectEven()
lanczosresize(720,480)


I should end up with a 480p 29.97 file with levels well in range and black level brought down. I gave it a visual inspection in virtualdub and may go back and tweak the black levels..-20 seemed excessive to me but it seemed to look OK.

I guess the AssumeTFF() doesn't go away after the deinterlace as HCEnc kept trying to set it to interlaced/TFF but I set it to Progressive.

In HCEnc should I use the CQ parameter? The default value is 5.

And finally..if after all that my video still looks bitrate starved....should I IVTC down to 24fps? I really don't like 24fps and I'm afraid it will introduce some serious jerkiness with all the panning/moving around...but just though I'd ask so hopefully my next script is my last .

[greymalkin]

OK everyone..please bear with me as I'm getting very close! I created the following script and fed it into HCEnc making sure it was now set to progressive, 9600kbps VBR...CQ=5, LumGain=2,AQ=2..etc.

LoadCPlugin("C:\Program Files (x86)\AVISynth 2.5\plugins\yadif.dll")
AVISource("G:\JackedUp1080iCanopusHQFine.avi")
ConvertToYV12(interlaced=true)
Tweak(bright=-10)
ColorYUV(levels="PC->TV")
AssumeTFF()
yadif(mode=1)
SelectEven()
LanczosResize(720,480)

The resulting file looks beautiful! and several hundre MB smaller than previous attempts, Narrowing the color range made a huge difference in the compression and perhaps setting it to CQ 5.00 also helped...it also rendered at about 1/2 realtime too which is much better than the other options! I might just use the same script for a 720p encode and just remove the SelectEven and of course change the resize parameters..

[poisondeathray]

OK everyone..please bear with me as I'm getting very close! I created the following script and fed it into HCEnc making sure it was now set to progressive, 9600kbps VBR...CQ=5, LumGain=2,AQ=2..etc.

LoadCPlugin("C:\Program Files (x86)\AVISynth 2.5\plugins\yadif.dll")
AVISource("G:\JackedUp1080iCanopusHQFine.avi")
ConvertToYV12(interlaced=true)
Tweak(bright=-10)
ColorYUV(levels="PC->TV")
AssumeTFF()
yadif(mode=1)
SelectEven()
LanczosResize(720,480)

The resulting file looks beautiful! and several hundre MB smaller than previous attempts, Narrowing the color range made a huge difference in the compression and perhaps setting it to CQ 5.00 also helped...it also rendered at about 1/2 realtime too which is much better than the other options! I might just use the same script for a 720p encode and just remove the SelectEven and of course change the resize parameters..

You have AssumeTFF() but no order argument in yadif, which will assume BFF. If this is a TFF clip, I think it will be jerky. You probably want yadif(mode=1,order=1)

Are you re-interlacing it? because if you just use SelectEven() it will be progressive 30p , not 60i like the other clips (i.e you throw out 1/2 the temporal information)

EDIT: oops, I missed the post above, are you planning 30p for DVD now? If so , all you need is Yadif(order=1,mode=0), and that will single rate deinterlace (ie. you dont need selecteven() then)

Also, what is your new "jacked up clip " ? does it refer to something different done in edius ? (i.e did you adjust the levels in edius)

You have more control if you use the smoothlevels filter (or even normal levels filter) vs. using a preset TV->PC and vice versa. Those presets are good only if you have a clip scaled 0-255 or 16-235 and you want to swap back and forth. Your clip was neither. Essentially you are overadjusting, then adjusting it back. As a general rule you want to minimize levels adjustments, because it hurts quality - especially if you do it in 8-bit avisynth. If you notice the histogram, you will see banding or discrete quantization. If you look over in gradients like sky or shadow, you will notice banding introduced by the levels adjustments. Smoothlevels() adds dithering so it's slightly better than original Levels() - you'll notice less banding

[greymalkin]

poison....DOH! I'll look up the field order argument and add it to my Yadif statement (found it! yadif(mode=1,order=1))..but I didn't really notice anything being off at all when I viewed it on my computer or standalone dvd player.

I am not re-interlacing, I de-interlaced with yadif, but when I tried to feed it to HCEnc it told me I had a 60p file which of course is not DVD compliant..so threw SelectEven() in there to bring it down to 30p. I gave up on interlaced because I kept getting bad interlace jaggies in the video. Maybe if I set the right yadif command it might fix it then I'll get rid of SelectEven()...oh and find the command to re-interlace again.

The jacked up file name..I just put that in for humor :P..it's the original canopusHQ clip. I used the ConvertYUY("PC->TV") command to bring the levels into the right range, then used tweak(bright=-10) to bring the black level down.

I'm still pretty new to AVISynth, so something that's dead wrong isn't immediately obvious to me..but you've been a great help a I fumble through! I'll go back and try one more time to get a good 60i version.

[poisondeathray]

ok give me a few minutes , I'll put together a post to try to explain the levels a bit more clearly

EDIT:
make sure you view these at full screen, and use an image viewer that allows you to see it at 1:1 pixels if your monitor isn't high enough resolution so you can scroll around

1 - yadif alone . This is just to show how poor a deinterlacer yadif really is. Compare this to #2. and flip back & forth. Notice all the dot aliasing everywhere. Look at the red leaves. The guy in the red shirt's face. This aliasing and noise eats up bitrate and is harder for the encoder. Didee over at doom9 did some tests and TGMC even in the weak/fast modes uses about 50-75% the bitrate for better quality by PSNR measures. Now, if you resize (downsize) the dot artifacts are a lot less noticeable, especially if you go down to SD. But if you keep full rez, you want to avoid yadif

2 - nnedi2 alone. just to show the contrast in deinterlacing quality. nnedi2 is intra-field deinterlacer, so you should use in combination with yadifmod or TGMC which work temporally instead of looking at 1 frame. Notice both 1&2 are blown out, and have lots >235

3 - nnedi2 + levels(17,1,255,0,235,coring=false) - notice the histogram is brought back within 16-235 range. There are small overshoots but that's ok. Notice it's less "washed out". This is because the black level is set for ~16. The arguments are input low, gamma, input high, output low, output high. Coring=false means it doesn't scale to 16-235 automatically. I suggest you play with the values , and see how they affect the image and the histogram. You can do this in avsp for example. Notice you were able to recover a tiny bit of the blowouts - notice the houses and tree in the background are more visible. The arguments are the same for the levels filters in your NLE's or vdub too, so it's worth your time to learn how to use them.

4 - nnedi2+smoothlevels - looks almost identical, except the histogram doesn't have as banding. You can't see corresponding banding in the actual image in this particular example, because this wasn't a huge levels change

5 - nnedi2+tweak+coloryuv - this is what you were proposing. Notice the dynamic range is depressed (the highest & lowest values aren't optimized) , so the image is very flat. Also notice the banding in the histogram, it's much more extensive. Now you can't see banding in this particular image, but you can usually with fine gradients like sky and shadows.

I'm NOT suggesting to use these values for levels or smoothlevels because your luminance values change over time when you move outdoor to indoor. IF you did these change in NLE, you could keyframe or adjust them dynamically over time. I purposely chose this frame for demonstration purposes ONLY, those values would not be good for the rest of the video.

[poisondeathray]

Now the blinking chroma issue for the AVCHD version.

I took the canopus file and fed into x264 (with minor levels adjustments to make it more legal) , encoding interlaced and never got the same chroma issues. It's almost as if it's getting associated with the wrong field, or doesn't know it's interlaced.

I included screenshots below, but it's better if you step through it you'll see the "blinking" better.

Look at the pink dress in (a) bleeding into the woman's sweater, and (b) pink shirt on lady, guy's bald head (which alternates white & pink), pink glass bleeding into the back of the guy

Can you re-iterate the exact steps you did for that version? did you input that canopus file directly into multiavchd? or was it the original or cut clip that you used as input? do you still have that logfile?

EDIT: I forgot to mention you should convert to rec.601 for your SD version. You can add it after your source filter
ColorMatrix(mode="rec.709->rec.601")

[greymalkin]

phew! OK...so I just thought I was done . OK...for the AVCHD version I fed the canopusHQ file straight into multiAVCHD and set it to BluRay/AVCHD compatible and set the bitrate to 13Mbps...that's about it.

so it's back to the interwebz for more schoolin on yadifmod+nnedi2.and the levels() function will be super helpful too! I knew my way was probably quick and dirty...but at least wanted to show you I wasn't sitting here waiting for a handout :P.

I updated my script to use yadifmod + nnedi2 + levels after stepping through and finding parameters that kept me in an acceptable range throughout (baby steps...it's too late for dynamic levels right now!). I remembered theres a text file with all the syntax included with each plugin (duuuh.)

I'm excited now to see how a properly processed 720/60p file looks...but I'll master the DVD creation first!

[greymalkin]

so my 6 mo. old son woke me up crying and I had to come in and check on my file..I'd feed him but he's still attached to his mother if you get my drift...but i digress...

it is the most beautiful MPG file I have ever created...watching it on my computer connected to my 47" HDTV (scaled to 1360x768 so I can see it from the couch) I can't believe I'm watching a SD MPG file! it is so sharp with so much detail..and the colors are definitely much better..I cannot thank you enough for saving me from mediocrity! and it only needed a little over 6.9Mbps to do it!

tomorrow I tackle the HD version!

[Gavino]

You have AssumeTFF() but no order argument in yadif, which will assume BFF. If this is a TFF clip, I think it will be jerky. You probably want yadif(mode=1,order=1)

By default, yadif will take the order from the current Avisynth setting, so yadif(mode=1,order=1) is equivalent to
AssumeTFF()
yadif(mode=1)

(there is a slight difference in that the TFF property will be retained in the result of the 2nd case, but as far as the deinterlacing operation is concerned, they are the same)

[poisondeathray]

I think I figured out what the AVCHD chroma/blinking problem was - I think it's multiavchd's transcoding error. It's not using ConvertToYV12(interlaced=true) in it's YUY2=>YV12 conversion so the chroma gets associated with wrong field. I tested with most recent multiavchd build

When I used .avs script in x264.exe (what I usually do) => no problem

When I fed canopus AVI directly into multiavchd => PROBLEM

When I fed .avs script into multiavchd => no problem

AVISource("CanHQFine.avi")
ConvertToYV12(interlaced=true)

I'll PM Dean and see what he makes of it. I'm not sure if he can test it, because CanopusHQ is proprietary codec - unless you have one of their programs installed that is bundled with it.

So as a workaround for now, just use proper .avs script as input

[multiAVCHD]

I think the better way is to use an avisynth script. It is not easy at all to add detection and proper colorspace conversion.

If "ConvertToYV12" works for all interlaced sources with interlaced=true, then I can add it if the source is detected/marked as interlaced.

[greymalkin]

that's fine guys...I have no problem feeding an AVISynth script to multiAVCHD.. it will handle all the rest beautifully!

[greymalkin]

ok...1 question for multiAVCHD...when I choose x.264 encoder should I use 2-pass VHQ or one of the 1-pass EXTREME presets for the super max quality?

[multiAVCHD]

Insane/Extreme are fast profiles not for quality but for speed. [2 pass] and [2 pass HQ] are okay, but quite slower.