I know the process of achieiving a high quality SD DVD from HDV footage is an ongoing point of debate with many different ideas and suggestions. I have used Jeff Belunes suggestions in his HD to SD workflow and am encountering less than satisfactory results, perhaps because of the 2hr length of material.

I have a 2hr Hdv Pal CS3 timeline I want on a SD Pal DVD for replication (a documentary that has been screened in HD as Blu-ray) where 500 will initially be distributed.

I exported the 2hr timeline as 5 de-interlaced smaller files & sent them via VirtualDub (VirtualDub couldn't handle a single 2hr m2v file) to create avi files for Encore.

Encore transcodes the avi files.

For a singde layer SD DVD the quality is very poor on occasions where there is motion/ camera movement and I am assuming the variable compression is high when there is motion and thus the quality deteriorates.

To get the best quality image for content this long (ie 2hrs)
- would TMPGenc be an improvement over AviSynth/VirtualDub? No transcoding within Encore?
- would a dual layer master be an aim rather than a single layer master?

I have the resources (and wish) to upgrade to Master CS5.5 and wonder if I should re-export from CS3 to create files for importing into Encore CS5. Is there any improvement to the DVD creation process in Encore CS5 to CS3? Is the transcoding quality improved?
Or should I try to open the CS3 Premiere file in CS5.5 and export in 5.5 to Encore 5 to obtain high quality?
Or are third party software (VirtualDub/TMPGenc etc) still recommended to be used for a high quality HDV to SD DVD with CS5.5?

Any thoughts/ suggestions would be appreciated.

I exported the 2hr timeline as 5 de-interlaced smaller files & sent them via VirtualDub (VirtualDub couldn't handle a single 2hr m2v file) to create avi files for Encore.
Don't deinterlace since you lose quality by doing that. If the source footage is interlaced, keep it interlaced when creating DVD's.

Roger, I compared a SD DVD that had used the exported de-interlaced files with a SD DVD using interlaced exported files and the de-interlaced was notably improved.

Are you using CS5.5? If so have you experience with creating SD DVD's from HD footage?

Thanks

Matthew, there is a bug in Premiere that makes the direct encoding of mpeg SD files from any HD timeline a mess. My workaround is to export first to SD PAL uncompressed, either quicktime or avi, and then convert them to mpeg-2. Then you won't have to deinterlace, and the image quality will be much better.

If you look around this forum, it's an issue that has been tackled numerous times recently.

Note to moderators - perhaps we should make a sticky of one of these threads?

Like Roger said do not de-interlace interlaced footage for DVD especially in CS3.
CS5.5 does a far better job but why should you de-interlace.
DVD players can handle interlaced footage just fine.
I have been working with HDV for years now and never de-interlaced.
Frameserving from CS3 to Tmpgenc is a very good option. Use to do that all the time and still do.

I've de-interlaced as the image on the DVD looks much better than if left interlaced.

Perhaps there is a bug in the interlaced workflow? (fields being flipped?)

From Premiere (AME) I create an mpeg blu-ray m2v - then create an avi from VirtualDub - then the avi's are used in the transcoding within Encore.

Where could fields be flipped there?

Matthew...
Is the source HDV interlaced or progressive?

I've de-interlaced as the image on the DVD looks much better than if left interlaced.

If there is not much motion and you shoot with a 1/60 or slower shutter, then interlaced HDV source will look better on a DVD after it has been deinterlaced. If there is lots of motion and you shoot with a 1/120 or faster shutter, than an interlaced DVD will look better. In either case, quality can be significantly improved by using a high-quality external scaler and encoder to master the DVD instead of relying on Premier and Encore to do everything.

Peter - the source is HDV interlaced

Eric - what high-quality external scaler and encoder options are there?

Thanks

Eric - what high-quality external scaler and encoder options are there?

An inexpensive option is to use Dan Isaacs hd2sd AviSynth script and HC Encoder as described in

HD to SD DVD – Best Methods | Creating Motion Graphics Blog | Blu-Ray DVD Authoring Menu | Precomposed (http://www.precomposed.com/blog/2009/07/hd-to-sd-dvd-best-methods/)

The main difficulty when down sampling HD to SD is to avoid aliasing artifacts which lower the quality of the resulting DVD. Even when there are no objectionable artifacts, a low-quality down sampling algorithm will reduce the apparent resolution. Aliasing can also reduce the efficiency of subsequent video encoding.

To show how much difference there is between scaling algorithms, a sinusoidal test pattern is used to compare the the aliasing properties of different scaling algorithms in

Down sampling methods (http://bvdwolf.home.xs4all.nl/main/foto/down_sample/down_sample.htm)

see also

http://www.mir.com/DMG/Software/freq-frame-table.html

It would be interesting to use these test this patterns to compare various HD to SD work flows.

Eric hit it on the head....

The only way to do proper downconverts is with third party softwares (AVISynth)
Adobe needs to step up to the plate and take this issue more seriously..
Until there's a plugin to properly address field manipulation, deinterlacing, resizing and reweaving, nothing will come of Adobe's claim to "Better quality" downconverts.

Heck, i can live with some artifacting that Eric mentions, but Adobe hasn't addressed the issue of properly dealing with interlaced sources.
How does your deinterlaced Vdub transcodes look?

Here's an old drawing i made, just to verify one of the scripts i used years ago, to better understand how resizing interlaced materials work.

I might be missing something, but in my experience the problem is limited to simultaneous downscaling and field switching in interlaced material (like from HDV to DV or MPEG-2). Progressive material does not seem to be affected. It is a major timewaster for many people (myself included), but let's not overstate the issue.

Don't know if this would work.
Save to hard disk the use DV Shrink to shrink it onto DVD

Download DVD Shrink v3.2.0.15 (freeware) - AfterDawn: Software downloads (http://www.afterdawn.com/software/cd_dvd/copy_dvd/dvd_shrink.cfm)

I might be missing something, but in my experience the problem is limited to simultaneous downscaling and field switching in interlaced material (like from HDV to DV or MPEG-2). Progressive material does not seem to be affected. It is a major timewaster for many people (myself included), but let's not overstate the issue.

Sorry Bart, perhaps i'm not understanding better....
But to properly do a downconvert of interlaced material, you need to separate the fields, resize, and properly weave them back.
I used to run these programs, and it took forever..Just think of the processing power involved. Also, think of all the different deinterlacers, options, and parameters in each deinterlacer, and we can come to the conclusion that Adobe hasn't taken the issue seriously...

Resizing interlaced frames any other way is wrong.
Of course, progressive is a different story. It's a non-issue. Whether somebody chooses bilinear, bicubic, or bicuspid, is all personal preference.

But nonetheless, interlaced downrezzing and progressive downrezzing are different animals..
That's why i always film in 30P.
AME correctly scales and crops progressive material..

I know the process of achieiving a high quality SD DVD from HDV footage is an ongoing point of debate with many different ideas and suggestions.

That was three years ago, I'm afraid.


I have a 2hr Hdv Pal CS3 timeline I want on a SD Pal DVD for replication (a documentary that has been screened in HD as Blu-ray) where 500 will initially be distributed.

I exported the 2hr timeline as 5 de-interlaced smaller files & sent them via VirtualDub (VirtualDub couldn't handle a single 2hr m2v file) to create avi files for Encore.

Encore transcodes the avi files.
...
To get the best quality image for content this long (ie 2hrs)
- would TMPGenc be an improvement over AviSynth/VirtualDub? No transcoding within Encore?
- would a dual layer master be an aim rather than a single layer master?

... and wonder if I should re-export from CS3 to create files for importing into Encore CS5.

First things first, you don't need to upgrade to CS5. CS3 handles HDV to SD brilliantly on its own. I've had the exact same workflow for my feature with spectacular results.

Once you've locked your edit:
Export your movie as an uncompressed image sequence (I used TIFF 16-bit from AE, since I had finished in AE). I strongly recommend using AE to render your project as opposed to Premiere Pro. Deinterlace footage unless your material is true progressive (unlikely). If you need to, sharpen your footage before exporting.

Open a new project in Premiere and import your image sequence and export to Encore. Use CBR with a value of 6.5Mbps (should be just about right in terms of size on a single layer DVD). One uses VBR to save space, especially when you have lots of additional video tracks, soundtracks, etc. For best quality, use CBR.

Premiere will create an m2v file on your system and this is loaded into Encore. It is critical that you do your bit calculations prior to exporting from Premiere. To learn how, read the Encore manual. It's pretty straightforward. If you do this right, Encore will not re-encode the footage, but if you have made an error, Encore will re-encode to 'correct' the error. You need to avoid that at all costs.

This workflow provides excellent results. You DON'T need any other software. Hope this helps.

Deinterlace footage . . . Use CBR with a value of 6.5Mbps

Deinterlacing before down converting to SD yields about 30% greater apparent spatial resolution while losing 50% of the temporal resolution. If your footage doesn't have very much motion this is a reasonable tradeoff. As a bonus the resulting video may compress more efficiently, especially if you encode as 30p progressive flagged for 2:2 pulldown using an external encoder. Note, however, that 6.5 mbps will only fit 1.5 hours on a single-layer DVD. For 2 hours try a rate less than 5 mbps.

I used to run these programs, and it took forever..Just think of the processing power involved. Also, think of all the different deinterlacers, options, and parameters in each deinterlacer, and we can come to the conclusion that Adobe hasn't taken the issue seriously...


OK, I never thought I would actually be defending Adobe in this regard. However I would recommend reading their article on scaling here:

some details about scaling in Premiere Pro CS5 and CS5.5 Premiere Pro work area (http://blogs.adobe.com/premiereprotraining/2010/10/scaling-in-premiere-pro-cs5.html)

You get different scaling algorithms depending on whether you have or don't have nVidia CUDA card in your system, and whether you choose Maximum Render Quality, or not.

Let's also separate three issues that are involved in this discussion:

1. The bug of improper deinterlacing in case of different fields on input and output (HDV to DV/MPEG-2)
2. Scaling of interlaced footage in general
3. Encoding of MPEG-2 files.

Ad 1.

The bug is there, hands down, and it has been there for past two iterations, so it's the highest time Adobe should step up and really fix this issue. You know, like half a year ago? I guess most people who complain about the poor quality of downscaling are the victims of this bug. The difference is best seen in generated graphics like lower thirds or credit rolls. If you have rounded corners where you should have sharp ones, this is the result of this bug.

The workaround is to separate the steps - first scale down to the footage with the same field order (like uncompressed 4:2:2), and then reverse the field dominance during encoding to MPEG-2 or DV.

So - before you complain, first make sure that you avoid this bug!

Ad 2.

The algorithm that is used for scaling depends on the hardware that you have or don't have. Lanczos filter, touted as the recommended method in the link that you posted, is used only when you have CUDA card on board. Otherwise it is very computationaly expensive - as you yourself have mentioned. But with GPU acceleration, it is much quicker and feasible. And it works quite well in my experience.

There is a noticeable difference between Gaussian (MRQ and no CUDA) and Lanczos (MRQ with CUDA) resizing. Frankly, I was surprised, comparing the results of one render with another. But to make sure that you export with the highest quality, you should have a CUDA card.

I agree that allowing the power-user to choose the scaling algorithm would give you more control, and allow non-CUDA users more choice between encoding time and quality, however the defaults that Adobe uses seem quite sensible to me, and to say that they "don't treat this issue seriously" is unjust. Perhaps it should be re-phrased as "they don't treat this issue seriously enough for me".

Ad 3.

Various encoders give various results. AME is not the best encoder out there, although it is very good, and will be sufficient for most cases. We routinely publish the DVDs and don't feel the need to resort to third parties software. However, I usually export CBR 8 Mbit/s files for the best quality that the DVD player can handle.

TMPGEnc and Cinema Craft come to mind as a little better software encoders. If you really want the highest quality, you would go for the third party encoder anyway, perhaps even a hardware solution. Some time ago the best MPEG-2 encoders were big boxes which would give you total control over everything that you needed, and many things that you didn't. But it was a solution for big authoring studios, costing hundreds of thousands of dollars.

Thanks all for your replies.

CS3 handles HDV to SD brilliantly on its own.

Sareesh - thanks for your suggestion and wonder if anyone else has successfully tried that method. I will give that a try.

Perhaps I could ask for some suggestions on the current worklow:

Source HDV 1080i, mostly interviews with still interviewees, some motion (eg interview with car driver with scenery passing behind outside the car window, also some moving camera - but only about 5% of the film is with motion)

Based on HD2SD (for SD PAL DVD's mainly on SD DVD players)

1.Export file(s) from Premiere -

Via AME - Mpeg2 Blu Ray (m2v), Upper Field First, Interlaced-or-De-interlace???
VBR 2 pass, Level:High, min Bitrate 25, Target 30, Max 35

2.AviSynth Script -

To keep it interlaced ("xxx.m2v", OutputFieldRate=50)
Should I convert from UFF to LFF for Encore??
If so ("xxx.m2v", OutputFieldRate=50, OutputBFF=True)
I have read "the HD2SD function uses TFF by default. Output BFF=True can be set to reverse the field order, which seems better when encoding with Adobe Encore"

3. VirtualDub -

Fast re-compress with Lagarith codec (which is YV12 colourspace) & save as avi

4. Import avi's to Encore

Transcode final as VBR 2 pass with bitrate min 0, average 4000, Max 7000
Audio PCM (I dont have Dolby Dig as an option although I could use mpeg).
Use Dual layer if 2hr 12min content goes over limit.

Cheers

Matthew, in my opinion you will waste quite a lot of time, especially encoding to Full HD H.264.

I suggest you first try encoding dkirectly to uncompressed SD 4:2:2 (upper field first), either mov or avi, then import it directly to Encore, and see if you like the results.

If you're going for double-layer, make it CBR 7 or 8 Mbit. 2h12 should fit about that. Plus Encore automatically encodes audio to Dolby Digital, but only if you transcode it in Encore. For some reason AME can't do it, probably something to do with licensing.

But this is what I myself would do.

Export your movie as an uncompressed image sequence (I used TIFF 16-bit from AE, since I had finished in AE). I strongly recommend using AE to render your project as opposed to Premiere Pro. Deinterlace footage unless your material is true progressive (unlikely). If you need to, sharpen your footage before exporting.

Open a new project in Premiere and import your image sequence and export to Encore. Use CBR with a value of 6.5Mbps (should be just about right in terms of size on a single layer DVD).

Sareesh, I am trying your method and realise I have a few questions- particularly where the actual HD to SD takes place.

1. AE Export - I have imported the HDV Premiere project into AE. When I am exporting the first of the files (I'm doing the 2hr 12 as 5 separate files) am I maintaining the HD resolution or down sizing to 720 x 576 (would have to use the stretch command)?

There isn't an option for bit size for the TIFF's but I notice they are 24bit.

Where can you de-interlace in the AE Export? "Field Render OFF"?

2. Premiere New Project - HD or SD here?

3. Regarding the bit calculations, is this one ok? DVD-HQ : Bitrate & GOP calculator (http://dvd-hq.info/bitrate_calculator.php)

There isn't an option for bit size for the TIFF's but I notice they are 24bit.

I think exporting 2 hours of video as an uncompressed TIFF image sequence may be a problem. At HD resolution you will get 1 TB of data; at SD resolution you will get 208 GB. If the DVD looks better deinterlaced and you've already made such a DVD, then you are done unless you want to experiment with different encoders and down sampling methods.

Regarding the bit calculations, is this one ok? DVD-HQ : Bitrate & GOP calculator (http://dvd-hq.info/bitrate_calculator.php)

The calculator recommends an average video bitrate of 4700 kbps for fitting 2 hours onto one DVD. This sounds right.

1. In AE, the bit depth is determined by each project, and is a global setting. It is under File-->Project Settings --> Color Settings -->
a. If you have heavy grading and finishing work, choose 32-bit float. Or, if you are using a modern system, select 32-bit float.
b. If you just want to render without making major changes choose 16-bit

For color space, Rec 709/HDTV is good enough for simple work. Choosing and working in a higher color space is beyond the scope of this post, and let's not complicate matters at this point.

Your comp resolution should match your source resolution, and fps should match, pixel aspect should be 1, with no deviations from your source material. Most HD systems are upper field, so under File-->Interpret Footage-->
a. Select upper --> then double click the footage to open it in the footage window --> use the time control to step forward five or more frames:
i.If the motion is consistent in one direction, you have set it right
ii. if the motion steps back every other frame, choose the lower field order setting.

Render settings: The point is to only render once.
a. Under Render Queue-->Output Module-->Lossless-->Format: TIFF sequence--> Choose any one
i. Millions of colors - 16-bit
ii. Trillions of colors - 32-bit
iii.Trillions+ (32-bit float)

An ideal master should be 16-bit, and that's good enough for broadcast work. Use 32-bit only for VFX work. Note: There's a difference between this setting and the project setting, and depending on what you've chosen in the project setting, your choices will vary.

b. Render Settings-->Best Settings-->
i. Color Depth: Match what you have chosen, or leave it at Project Settings
ii. Regarding Field Render, the simplest way is to perform a test on your footage. The CS3 manual explains it perfectly. In short, the idea is to test a 3-second clip with both settings and choose the one that works for you. The test should take 10 minutes. Do it.

2. The new Premiere project is HD and matches your master, with no changes, at 25p.

3. The link you have provided seems okay, and you could use that to recheck your calculations. However, I strongly suggest you read the Encore CS3 manual --> Planning your project --> Everything, including Bit Budgeting. Once you have learnt how things work, you can confidently use a calculator (or build your own with MS Excel!) for future projects.

Regarding deinterlacing:

There are two ways:
1. User the deinterlace flicker tool or another tool. If you have really weird interlacing artifacts, this will help to a certain extent, with the downside that it reduces resolution + you need different settings for each 'problem' and it's a lengthy process.

2. Read 'Separate video fields' in the AE CS3 manual. It explains things perfectly. I recommend option 2, because the goal is to preserve resolution. Use only when required. This might seem daunting at first but once you get the hang of it, it's pretty easy.

Here's one of the tricks it suggests (use sparingly!):
a. Import the same footage twice.
b. Interpret one with field upper, the other with field lower.
c. Blend both to get film-like blur.

Hope this helps.

I use TMPGenc TMPGEnc - Products: TMPGEnc Video Mastering Works 5 (http://tmpgenc.pegasys-inc.com/en/product/tvmw5.html). Has a very good encode from HD. It allows you to see how much the disc is being used for the data rate chosen.

Ron Evans

Are you using CS5.5? If so have you experience with creating SD DVD's from HD footage?
Yes, i am using CS5.5 with a certified graphic card for hardware acceleration. When the graphic card handles the downscaling it is really good.

Note that the export settings do default to Lower Field First in CS3, so manually changing it to Upper Field First must be done. If you accept the default, then the de-interlaced export do look better. If you change to Upper Field First, the interlaced export looks best. Export directly to MPEG2-DVD and avoid going to DV PAL/DV NTSC.

What "best" is is of course in the eye of the viewer. I don't like the look of de-interlaced material...

I use TMPGenc TMPGEnc - Products: TMPGEnc Video Mastering Works 5 (http://tmpgenc.pegasys-inc.com/en/product/tvmw5.html). Has a very good encode from HD. It allows you to see how much the disc is being used for the data rate chosen.

Ron Evans

Hi Ron,

What's your workflow for HD to SD with Premiere & TMPGEnc (& Encore if you use it).

Thanks

My normal editing is with Edius and with that I would edit HD export a Canopus HQ file and then encode with TMPGenc. Since this codec is on my PC I can use this with Premiere or Vegas as well as Edius. Not sure what options you may have on your PC for higher quality file export but you would need to use the best format short of uncompressed that may create a very large file !!!

I author in one of Encore, DVD Architect or DVDLab Pro. All will take the files from TMPGenc without re encoding.

Ron Evans

Hi Ron,

What's your workflow for HD to SD with Premiere & TMPGEnc (& Encore if you use it).

Thanks

If you do not use an intermediaire codec frameserving from Premiere to TMPGEnc is the easiest way.
Download Advanced FrameServer 0.3.118 Free - Frame server for your Premiere application - Softpedia (http://www.softpedia.com/get/Multimedia/Video/Video-Editors/Advanced-FrameServer.shtml)

If you do not use an intermediaire codec frameserving from Premiere to TMPGEnc is the easiest way.
Download Advanced FrameServer 0.3.118 Free - Frame server for your Premiere application - Softpedia (http://www.softpedia.com/get/Multimedia/Video/Video-Editors/Advanced-FrameServer.shtml)

If you can't get frame serving to work or want to master the DVD on a different computer, you could also use DNxHD which is available for free download from Avid.

Avid | DNxHD Codec - Beauty Without Bandwidth (http://www.avid.com/US/industries/workflow/DNxHD-Codec)

DNxHD works well for me when moving video between different programs and computers.