Very poor quality export from Premiere Pro using mpeg2-DVD.. HELP!!

[hello_hello]

Nice samples, hello_hello. If newpball had the decency to look at them he would be wondering why it doesn't quite look like you're watching only 8.33% of a video...

He prefers to impersonate an ostrich.

So who'd have guessed, The encoder makes a difference.
What I'd put in the "that's not what I expected" category though, is Nero 7 seemed to do a better job than AVSToDVD (default settings for each program). Lots of keyframe pumping with AVSToDVD's default settings (ffmpeg encoder). I think the Nero encoder is possibly better than the HCenc encoder too, based on my monumentally limited testing, but converting the colours and some subtle (quality) sharpening helps.
(Edit: After a second look I might withhold judgement on Nero vs HCenc encoder until I've played around a bit more. I think maybe at the moment the differences I'm seeing are more resizing and not so much encoding differences)

It'd still be nice to have a sample of the video in question to play with. The consensus seems to be, expected detail reduction due to downscaling aside, the problem is the encoding quality.

[poisondeathray]

One of the things that will help for sure for his specific problem, is higher bitrate - even within the same encoder . Although he said DVD9 wasn't an option, perhaps using AC3 audio is an option instead of PCM wave. You can redistribute ~1Mb/s into the video instead

[manono]

One of the things that will help for sure for his specific problem, is higher bitrate - even within the same encoder. Although he said DVD9 wasn't an option, perhaps using AC3 audio is an option instead of PCM wave. You can redistribute ~1Mb/s into the video instead

And once he's done that he can gain much more quality at the top end by raising the max bitrate to at least 9000 and help improve the overall average quality quite a bit by lowering the minimum bitrate substantially (I'd recommend 500). I don't know how possible that is with HCEnc, but his first picture using the Premiere Pro encoder showed a minimum of 4000. As it is now the relatively static scenes look very good but once there's movement it all goes to hell in a handbasket.

When I use VBR 2 pass, the quality is perfect, but there is a pixelated distortion around everything that moves quickly.

[hank315]

... As it is now the relatively static scenes look very good but once there's movement it all goes to hell in a handbasket.

The reason for this is because the used quantisers for I, P and B-frames are distributed wrong, quants for I-frames are
1 - 3, for P-frames approx. 6 and for B-frames in the range 25 - 112 (112 being the highest value possible in MPEG2).
So the B-frames are bit starving which is what you see around moving objects, for the static surrounding the B-frame MacroBlocks
are skipped so are in fact just copies from the I/P frame MacroBlocks which look OK.

More reasons why this video looks so bad:
- Quant matrix used is standard MPEG (all 16 for non-intra)
- Gop size too small (12), raise to 15 for PAL
- The stream is padded with zero's, bits are simply thrown away, this should never be done using VBR
- stream is progressive (progressive_sequence is flagged) but alternate scan is used

AFAIK Premiere uses the Mainconcept encoder and that can certainly do better than this.

[poisondeathray]

AFAIK Premiere uses the Mainconcept encoder and that can certainly do better than this.

Rovi/Mainconcept certainly can, but not many options are not exposed in the "watered down" version that Adobe licenses

[manono]

The reason for this is because...

Outstanding! Thanks for your very informed comment, hank315.

[kieranvyas]

... As it is now the relatively static scenes look very good but once there's movement it all goes to hell in a handbasket.

More reasons why this video looks so bad:
- Quant matrix used is standard MPEG (all 16 for non-intra)
- Gop size too small (12), raise to 15 for PAL
- The stream is padded with zero's, bits are simply thrown away, this should never be done using VBR
- stream is progressive (progressive_sequence is flagged) but alternate scan is used

Sorry I have only just seen everyone's posts here! First and foremost thank you to everyone who is trying to help me I really appreciate.

Now PLEASE LET ME CONFIRM I fully understand that when going from HD to SD I will lose an enormous amount of quality. I have no problem with this - in fact I am pretty pleased with the general quality of the DVD. It is simply the distortion that surrounds a lot of the motion in various scenes that I have a problem with. Interestingly enough, I exported a quick HD tester using the H.264 codec, dropped it into Apple Compressor and encoded to mpeg2 DVD - guess what... no distortion whatsoever!! So I ended up with far better quality encoding in Apple Compressor from a H.264 file than I did exporting directly to mpeg2 DVD from Premiere! This has led me to believe that the problem possibly lies with my sequences using a mix of footage with frame rates of 25fps and 50fps.

Anyway! At the bottom I have included 4 examples for comparison:

1) Clip1 Mpeg2-DVD.m2v - The mpeg2 export directly out of Premiere. Awful distortion.
2) Clip1 Lossless>Apple Compressor.m2v - The mpeg2 export using Apple Compressor from the 'Clip1 Lossless.mov' file. No distortion!
3) Clip1 Lossless.mov - The lossless version. Exported using the animation codec in Premiere.
4) Clip1 H264.mov - Just a H.264 export in case you don't want download the extremely large lossless version!

Apple Compressor wins hands down! Now everybody tells me the encoder within Premiere is far superior to Apple Compressor, so this must mean I am doing something majorly wrong.

Here are my sequence settings:
1920x1080
50fps
Square Pixels
Progressive Scan

And here are my settings I am using to export mpeg2-DVD directly out of Premiere:
MPEG2-DVD
720x576
Quality:100
Frame Rate: 25
Progressive
Widescreen 16:9
PAL
VBR, 2 Pass
Minimum 2.25 Mbps
Target 5.7 Mbps
Maximum 8Mbps
GOP Settings
M Frames 3
N Frames 12

Rendering at maximum depth
Using maximum render quality

I will be eternally grateful and owe my life to anyone who can explain to me how to achieve the same quality as the Apple Compressor clip but through Premiere Pro! MANY THANKS

[manono]

Since pdr said you can't adjust a lot of the settings, then why not switch to using a different and better encoder, one such as hank315's HCEnc?

I can't explain most of the differences but can point out that the information bears out that the MainConcept result is definitely inferior to that produced by Apple.

The first two pictures are from your MainConcept encode, the lower two from Apple's encoder. The most striking thing to me is the huge difference in the quant values (the Q-level). The MainConcept one at 10.14 is guaranteed to produce mosquito noise and macroblocking while the Apple one is quite low.

One explanation is that the different encoders use different quantization matrices and hank315 earlier criticized the one MainConcept uses. I don't much like either matrix but the one Apple used is certainly preferred when the source is hard to compress. I don't use either encoder but am fairly sure you can't choose to use a different quant matrix. But that won't explain the entire difference. The higher max bitrate employed by the Apple encoder (9000 .vs 8000) allows the more complex movements to get more bits. That's an important factor. In addition, the MainConcept encoder encoded a progressive stream using Alternate Scanning and the Apple one used the preferred ZigZag Scanning. Apple used a lower DCT Precision (8 .vs 9). But those are fairly minor. There are other reasons, probably some of the ones hank315 pointed out earlier, ones over which you have no control.

Again, I'd suggest using a different encoder.

[kieranvyas]

Again, I'd suggest using a different encoder.

I already tried using HCencoder but the trouble is I'm on a Mac. I exported losslessly from Premiere as a quicktime format (as avi is not an option on macs) and used the script:

QTInput("dvdtest.mov")
hd2sd(interlaced=false,OutputColorSpace="YUY2")

with avisynth and then encoded using HCencoder. Unfortunately however I was getting darker colors, as if the contrast had been turned up slightly. I troubleshooted this issue a while ago and the only conclusion anyone could reach was to somehow use avi rather than the quicktime format. But as I'm on a mac that isn't possible.

However I have to admit I find HCencoder quite confusing. If you could help me with the settings I would be very grateful! Thank you

[Skiller]

Thanks for that lossless sample, I had some fun with it.


Here's my result at downscaling and encoding it.

I used a custom AviSynth script with linear light resizing in 16 Bits, low-pass and a slight FineSharp to carefully bring out a little more grass details.

Encoded with CCE SP3 with Manono3 Matrix and and an average Bitrate of 5700 Kbit/s (max 9200, min 2500, although the sample is too short to have much of a fluctuation). Personally I would try to use a higher average bitrate because the video is so detailed but I used 5700 here because it seems to be your required average bitrate.



I noticed the levels seem to be wrong in your Apple Compressor sample.

Apple Compressor


Me


The levels of your Premiere export are the same as mine, so it's quite obvious Apple messes it up somehow.

I also noticed there seems to be no low-pass at all with both of your encodings which results in heavy interline twitter (jaggies) whenever it's played back through interlaced playback chains. Here's an approximate simulation of what that would look like:



The Premiere MPEG2 encoding is horrible (despite >7000 Kbit/s in that sample!) but the actual downscaling is better than Apple Compressor which is rather blurry. Also, Apple scales 1920x1080 to 720x576 (rather than 704x576) which both I and Premiere consider wrong (hence the small black bars in the Premiere encode).

[kieranvyas]

I used a custom AviSynth script with linear light resizing in 16 Bits, low-pass and a slight FineSharp to carefully bring out a little more grass details.

What was the avisynth script you used for this? And you are right about the low pass, i noticed that! Can I apply this in HCencoder?

[Skiller]

And you are right about the low pass, i noticed that! Can I apply this in HCencoder?

No, only via AviSynth.


Complete script is attached (because it's rather long).

You need Dither.

[_Al_]

kieranvias - that lossless, full HD, you posted Clip1 Lossless.mov is 25fps, your original is 50p. To mke a DVD m2v file one would need 50p original. Skiller made 25p m2v sample out of it. Is that intended because you do not want to deal with interlace ? That DVD should be interlaced, imho, because original was shot 50p. With proper resize and low pass filter 50i would give better fluent feel. But sure it will not be that sharp, maybe that is what you want, not sure now.

[kieranvyas]

kieranvias - that lossless, full HD, you posted Clip1 Lossless.mov is 25fps, your original is 50p. To mke a DVD m2v file one would need 50p original. Skiller made 25p m2v sample out of it. Is that intended because you do not want to deal with interlace ? That DVD should be interlaced, imho, because original was shot 50p. With proper resize and low pass filter 50i would give better fluent feel. But sure it will not be that sharp, maybe that is what you want, not sure now.

Some of my original footage is interlaced, which I deinterlaced within Premiere. Most of my footage is progressive though. So are you saying I should be exporting my DVD as interlaced?

[_Al_]

I am not sure what was said in the whole thread above, ...,

If you make 25p m2v out of 1080 50p you loose temporal resolution, you get jerky movement on DVD. You focus on sharpness and detail but you also loose temporal resolution.

If you make 50i m2v you keep temporal resolution, but you have to introduce low pass filter in script. But footage will not be that sharp as above. Your decision. I think 50i would be more appropriate.

You know now that you cannot do it in Premiere , bad results, or at least you try to allocate more bitrate to to your content by using AC3 as audio, in PAL land you might even use mpeg1 layerII audio, bitrate 224kbp/s or so (because Premiere cannot encode AC3 without extra plugin I think). That was mentioned by manono, not sure if you tried that and set minimum bitrate lower (not sure how much it would help, it depends on your video content).

Get YUV lossless out of Premiere. And possible avs could be:

#load clip here
#posible YUV conversion to YV12, not sure what video you load into avisynth, not sure why RGB, YUV would be more appropriate I guess if you do not work with colors
Spline16Resize(720,576) # or Spline36Resize(720,576)
Blur(0,0.5) # something between 0.5 - 1.5, whatever works
ColorMatrix(mode="Rec.709->Rec.601", clamp=0) #to have proper SD resolution color space for DVD
AssumeTFF().SeparateFields().SelectEvery(4, 0, 3).Weave()

that low pass filter could be different , there is more ways to do it. To just blur lightly vertical lines is most simple. This way flickering is eliminated, how much it is up to you. You do not want to do it too much to have it blurred too much. While judging overall result, do not look at snaps at microscope. Use bigger screen and look at it from regular viewing distance, possibly on TV. Because TV deinterlacer is important here, you are making DVD. Do not overanalyze some snap samples, made by player, deinterlaced etc.

[Skiller]

Some of my original footage is interlaced, which I deinterlaced within Premiere. Most of my footage is progressive though. So are you saying I should be exporting my DVD as interlaced?

It's hard to judge. I would base the decision on:

• how much (percentage) of the final exported video was originally filmed at 25i rather than 25p?

• how much movement/panning is in those shots that were filmed at 25i?


If there are lot's of 25i shots I would go for an interlaced DVD.
If there are few 25i shots but with lots of panning and movement I would go for an interlaced DVD.
If there are few 25i shots with few movement and ideally no panning I would go for a progressive DVD.

If you make 50i m2v you keep temporal resolution, but you have to introduce low pass filter in script.

Actually, for DVD one should always apply a low-pass – even when you're dealing with pure progressive 25p video. If the video, progressive or not, is played back via an interlaced playback chain there will be jaggies otherwise, because there's no way to differentiate the two in that scenario.
That's why I applied a low-pass even though the lossless sample is 25p and was encoded as progressive by me.

[manono]

There's no reason you can't have both interlaced and progressive footage in the same DVD. Just encode the whole thing as interlaced. Most PAL DVDs are done that way anyway, even for movies. I don't know how good of an idea that is, but it's the norm rather than the exception.

Some encoders (CCE, for example) can low-pass the footage while encoding.

[Skiller]

Some encoders (CCE, for example) can low-pass the footage while encoding.

Yes, but CCE optionally does it to improve encoding efficacy (to reduce mosquitos) whereas we do it to reduce interline twitter. Not the same thing (horizontal vs vertical).

[kieranvyas]

[QUOTE=Skiller;2387354]

If you make 50i m2v you keep temporal resolution, but you have to introduce low pass filter in script.

Actually, for DVD one should always apply a low-pass – even when you're dealing with pure progressive 25p video. If the video, progressive or not, is played back via an interlaced playback chain there will be jaggies otherwise, because there's no way to differentiate the two in that scenario.
That's why I applied a low-pass even though the lossless sample is 25p and was encoded as progressive by me.

Thank you guys for your continued support. I'm having such a nightmare I think I would have exploded if it wasn't for this forum! Basically because I'm on a Mac I'm running Windows through Parrallel (which is like Bootcamp) and using HCenc. It was working fine but now I can't implement your script Skiller as HCenc returns an error saying it requires Avisynth version 2.5.4 or higher, despite it also telling me I have v.2.5.8 installed successfully!

I am just about to build my own windows workstation and I'm thinking maybe I should just wait and transfer my project over to that machine and try HCenc then...as I'm currently at a loss and tearing my hair out!

[manono]

Yes, but CCE optionally does it to improve encoding efficacy (to reduce mosquitos) whereas we do it to reduce interline twitter. Not the same thing (horizontal vs vertical).

I've never used it myself but figured it was the same thing. Live and learn. Thanks.

[kieranvyas]

If you make 50i m2v you keep temporal resolution, but you have to introduce low pass filter in script.

Actually, for DVD one should always apply a low-pass – even when you're dealing with pure progressive 25p video. If the video, progressive or not, is played back via an interlaced playback chain there will be jaggies otherwise, because there's no way to differentiate the two in that scenario.
That's why I applied a low-pass even though the lossless sample is 25p and was encoded as progressive by me.

Is there any way to reproduce a low pass in Premiere Pro?

[Skiller]

It was working fine but now I can't implement your script Skiller as HCenc returns an error saying it requires Avisynth version 2.5.4 or higher, despite it also telling me I have v.2.5.8 installed successfully!

It may fix it or may not but it's worth a try to install AviSynth 2.6.0 RC 3 [150419].

Also try if you can look at the script's output by dropping it into VirtualDub or if it returns an error message instead.

I am just about to build my own windows workstation and I'm thinking maybe I should just wait and transfer my project over to that machine and try HCenc then...as I'm currently at a loss and tearing my hair out!

Yeah, that might be easier on your nerves.^^

Is there any way to reproduce a low pass in Premiere Pro?

Some vertical blur + sharpen would basically do it but... that's the kind of stuff I would never do in an NLE because that's not what they are exactly good at.

[LMotlow]

Having learned a whole lot of stuff from poisondeathray even before I signed up last year, I must disagree. Whatever resizing was used for that video, it softened everything, and whatever other processing was done didn't maintain the original gamma or IRE levels and generated a mushy image, artifacts or not. We're not talkiing about a $50 NLE here. Just out of curiosity I got better MPEG results right off the bat with a simple TMPGenc Mastering Works re-encode directly from the .mov file (but that won't play as well as the Avisynth version discussed earlier, so why bother with it?). Not that the original is that great to begin with, being slightly overexposed and fuzzy as the dog passes in front. I played with it myself and found straight Spline36resizing cleaner than the 16-bit mod version and way sharper than the Adobe m2v. But at the low bitrate the O.P. wants, he'll need some mighty upper-tier software to get better results whether he uses Adobe or not. I'd go for double-layer DVD. But the O.P. requires lower quality, so that ends another part of the debate.

What 50p source? The .mov sample is 25p.

[smrpix]

Let me suggest another approach for a moment. Go back to Premiere and rather than using the default DVD values, raise your target bitrate to 7 and your maximum bitrate to 9. You can also set your source settings to "stretch to fill." See if that solves things.

[Cornucopia]

Why is the OP trying to swim upstream?!

Go with the flow and do what the veterans here have been telling you:
Use the right/best tool for the job at hand.
Premiere excels at editing, but it doesn't (and wasn't intended to) excel at processing.
AVISynth excels at processing.
Output from Premiere a lossless clip that isn't resized. Resize+process in AVISynth and frameserve to HCEnc (2pass VBR at highest avg. bitrate for your intended DVD disc size). Done! Best output possible.

Why should there even be more to the conversation?

Scott

[kieranvyas]

Why is the OP trying to swim upstream?!

Go with the flow and do what the veterans here have been telling you:
Use the right/best tool for the job at hand.
Premiere excels at editing, but it doesn't (and wasn't intended to) excel at processing.
AVISynth excels at processing.
Output from Premiere a lossless clip that isn't resized. Resize+process in AVISynth and frameserve to HCEnc (2pass VBR at highest avg. bitrate for your intended DVD disc size). Done! Best output possible.

Why should there even be more to the conversation?

Scott

You are absolutely right, HCenc is the only way forward. However, I've actually tried the HCenc method now using two different scripts featured above, but the resulting footage is darker than the original, as if the contrast has been turned up. I'm very nearly there except for this final problem!

**In case anybody else reads this forum trying to use HCenc on a Mac. I figured out you cannot create AVS scripts through the mac app Texedit. You must use Notepad on Windows.**

[manono]

but the resulting footage is darker than the original, as if the contrast has been turned up.

Did you change the colorimetry as suggested by _Al_ earlier. It should always be done when going from a hi-def source to standard-def output:

ColorMatrix(mode="Rec.709->Rec.601", clamp=0) #to have proper SD resolution color space for DVD

http://www.avisynth.nl/users/warpenterprises/

[poisondeathray]

The difference is in levels, it's the "quicktime gamma bug" difference. Older macs use ~1.8 gamma, but rest of the world use 2.2 gamma for displays.

The compressor version is flatter and black level is elevated. It might look right on some Mac displays but it's technically wrong if we assume the MOV is correct. The MOV sample you uploaded is RGB in quicktime animation, so if that was representative of the original, that "darker, increased contrast" version is the correct one. That is what it will look like using the industry standard ITU Rec matrices

[kieranvyas]

The difference is in levels, it's the "quicktime gamma bug" difference. Older macs use ~1.8 gamma, but rest of the world use 2.2 gamma for displays.

The compressor version is flatter and black level is elevated. It might look right on some Mac displays but it's technically wrong if we assume the MOV is correct. The MOV sample you uploaded is RGB in quicktime animation, so if that was representative of the original, that "darker, increased contrast" version is the correct one. That is what it will look like using the industry standard ITU Rec matrices

That explains everything!!!! Thank you for that information.. and thank you to everyone who has assisted me on this thread. I am quite new to this and you have all taught me a LOT! The community here is awesome!