Maximizing HD to SD Quality
 

There has been a lot of discussion about the poor quality SD video that results from a HD down conversion. The linked article describes a rather tedious procedure that is Premiere-centric. Precomposed Blog - HD to SD DVD - Best Methods Can anyone suggest a Vegas work flow for the same result? I also use Cineform NeoScene.

When you are required to shoot in HD for whatever reason, my suggestion is of no help: shoot in SD.

I maintain that when delivering in SD it is best to shoot in SD. Since there are times that you may be delivering on Bluray and also need to deliver additional copies of a project in SD, I believe simply rendering out m2t files to mpeg 2 is sufficient. No, the quality may not be the highest possible, but the difference, IMO is so slight that it truly means little to me or my clients. I think it is more important to acquire the highest quality to begin with, and these little nuances in post mean absolutely nothing. Just my two cents.

Jeff,

Thanks, but that's not what we wanted to hear.

I too am on a quest for better quality delivery. So much of my work ends up on the web, that I hadn't been delivering much DVD from HD source footage till this wedding season, and I must say the DVDA trancoding sucks. I haven't done many HD weddings til this year. The few I delivered over the last couple of years looked bad on dvd and I just assumed I hadn't tweaked my settings as well as I could have. I read thru the article Jim referenced and it confirms that the HD to SD workflow doesn't work.

I recently rendered a 28-minute piece, that is about 28gb, in very HQ(.avi). By the time I encode to mpeg2 for DVDA it's about 2.x gb and looks like absolute crap compared to the original.

I'm looking at the Procoder 3 route, but at $500, is it worth it.

I'm even thinking about offering an upgrade to the Western Digital HD Media Player and delivering on USB. YouTube - Western Digital WD TV HD Media Player

There has to be a better way, without a 15-step workflow (exaggerating).

You mention DVDA transcoding....I encode my mpeg so that is doesn't have to be re-encoded for DVDA, isn't that what you are doing? Or am I missing something?

Jeff,

Yes, I'm transcoding in Vegas 8 to mpeg2 for DVDA. I think it's the same process, but Vegas is quicker. In reading the article that Jim refers to, this HD-SD problem is inherent in most NLE.

I suppose if you're not looking at the very nice .avi files (I store a HQ .avi version for archive), you wouldn't be so sadly disappointed in the mpeg2 version.

As the article points out, the issue is really the downconversion, not the MPEG-2 encoding itself.

In my own experience, I must concur that the best SD is from SD. But when that isn't possible, I've done something in Vegas that seemed to help my quality.

I shoot in XDCAM HD. To downconvert my project, I render it to IMX 50. That's a Sony I-frame only MPEG-2 mastering format at 50Mb/s. The downconversion in Vegas to that format seems cleaner than any other method I've tried, and I think the reason is that IMX is one of the few SD formats that is top field first, just like my HD (or HDV) is. And this seems to help in the quality of the downconversion in Vegas.

Then from the IMX file I can go to MPEG-2 any way I want, myself typically through the Frameserver to TMPGEnc. Note that NTSC IMX is a 720X486 format, so you need to match output aspect to 720X480 for your MPEG-2 render.

Also, be sure for all downconversions that you have the deinterlacer in project properties set to BLEND or INTERPOLATE and not NONE, since Vegas must deinterlace when scaling, though it will restore the interlace later, if it is needed for your format.

I do HDV 1080 50i (I always have my in-camera colour desaturated a bit during filming, adding colour/saturation in post and have noticed it does help in the final product whether DV or HD)
> DI w/Cineform
- apply a almost unnoticeable sharpen
- render progressive w/interpolate enabled
> AVI uncompressed
> TMPGEnc mpg2

It’s fine but not perfect

But lately I’ve been a little interested in the HDV > IMX instead of cineform approach!

If the quality is important, I recommend a bicubic spline downcoverter. If you can't do that, then a Lanczos rescale is also very clean, and what I do for any of my projects that go from HD to SD.

Unlike some others, I find that my HD to SD downconversion yields superior results in every way possible. Frankly, they look stunning as I have demonstrated on this forum in the past. The workflow is a bit inconvenient, but gets the results I want.

Doing this, I find I need to add no sharpening when going to Mpeg2. When I am going from SD to Mpeg2, I add just a bit of an edge with sharpening knowing that the compression will knock nearly all of it back off.

I'd be happy to post screenshots of downconverts for you guys. It CAN be done very cleanly.

Ok, so here's an example.

I took 2 frames of a RED 4k file and laid it onto a timeline. I conformed that timeline to 1024x512 to preserve the 2:1 layout of the original. I took a frameshot of that 1024x512 file. I then went back to a 4k timeline, rendered that 4k RED file to a 4k uncompressed AVI file. Pulled that into virtualdub, did a lanczos resize to 720x360, and rendered back out uncompressed. Put that onto a 720x360 timeline, and took another frameshot.

So here, we have an SD sized frame derived from a 4k source. This site won't let you upload a real HD sized image, so I had to pass on that.

Look at the clean edges on the signs in the background and the engine. The clean edges of the lines on the car. All those problematic things people say you can't get on a downscale from HD. If coming down from HD was a problem, then coming down from 4k should be terrible. Judge for yourself.

There is a color space shift that happened that I didn't bother to correct for. Sorry. Other than that, this photo got zero processing. Just laid on the timeline and a frame shot taken.

Simple question - - Since much higher quality HD to SD (MPEG-2) conversion is possible, why don't the so-called "Pro" editing software companies such as Adobe, Sony, Avid, Grass Valley et al offer it? Has anyone asked these companies? If so, what did they say? Personally I think it is ridiculous to ignore something like this in products that cost many hundreds of dollars each.

I believe Apple's "Compressor" program includes the Lanczos rescaler, but you have to know the setting to use it. In typical Apple fashion, the underlying choice is hidden from the user, so it's not obvious how to turn it on. There was a long thread here about it.

As to why this option is not available in NLE's, that I couldn't tell you. Perhaps they see down conversion as something not many people would want. Certainly Lanczos and Bicubic Spline algorithms are well understood and free to implement. They do take quite a lot of time to do what they do and as is proven time and time again, 99% of people are more interested in speed than quality of results. So they seem to be hitting their targets quite well.

I just can't see that rationale. If they are that disconnected from their users, maybe they need to find a new line of work. I believe the NLE market is ripe for a new company with a keen desire to capitalize on the state of the NLE industry. The current NLE suppliers are bloated with everything from the internal politics of massive companies to kissing up to stockholders.

I suspect one aspect of the problem is that they don't have the skill set in-house to fix the problems. Adobe and Sony license the Main Concepts encoder. If they were that "skilled" with encoding, they wouldn't need to go out-house would they. I believe they are all more focused on the superficial glitz and glitter of their products and aren't nearly as concerned about the "hidden" functionality of the core elements of their products.

I've been fairly happy using TMPEG enc. for downscaling and converting my Cineform 1080i masters to MPEG-2. Inside of TMPEG in the filters section it lists Lanczos-3 as the default picture resize methed. This is probably why the downscale looks good.

Marc

Thanks Marc. You may have just given me a practical work flow. I use Cineform. If all I need to do is output my final edit as Cineform and then encode it to MPEG-2 with TMPEG, that's a reasonable work flow.

From my experience in the print and web production world, it is almost always is better to start with higher-resolution source material then "down-res" to a lower resolution.

This does not seem to be the case starting with HD footage then going to SD...or is it? As recently as last night I watch a television program, half on the HD1080P television in my living room, then switch to my old SD television in my bedroom. The program looked great in HD, and it looked great in SD.

I would assume that the show was shot only once in HD, then "down-resed" for SD viewing. Is this technology not available or easily in reach of desktop editors?

I just started mixing HD with SD footage. I thought the big issue was going to be differing aspect ratios, but it seems to be more complex than that.

Ken

Kenneth, as outlined earlier in this thread, the tools to do high quality conversion are available, and in the case of the PC, they are free. They require more effort on the part of the editor, and in many cases, through either inexperience, ignorance, or just unwillingness, these tools are simply not used.

The amount of short-cuts taken and desired by most editors is simply shocking to me for those being paid for their work. I don't mean to be harsh to anyone, but if you are a professional, you are being paid for the quality of your work. If you don't avail yourselves of the (free) tools, then there is no one to blame. Yes, it would be nice if all the NLEs did great down conversions for you at the click of a button. But they do not. So it is incumbent upon us to do what is necessary to create the best product we can.

I'll put the quality of my down conversions against anything on SD broadcast right now. And I didn't pay a dime for the tool to do it. Look at the downconversion quality every time we se a Hollywood film on BluRay, DVD, or TV. That came from FILM scans of 2k or 4k, and through maybe 3-8 different compressions. It CAN be done, and done well.

I understand what you are saying Perrone, and I am actually very familiar with using Virtualdub and its various mods for web video conversion workflows, but I am not yet familiar with the workflows that will give me sterling SD footage from all the various HD sources. I am taking it all in and trying to get it straight.

Do you think the "big boys" are using Virtualdub etc. or are they using a different technology? Obviously a 1-button solution is not realistic even for a big editing house, but I wonder what they generally use for a conversion from HD to SD?

P.S. You can call me Ken - or Kenneth, I don't mind either way. ;-)

Ken

No, I don't think the big boys are using VirtualDub. They can afford to hire a programmer to write a GUI around a bicubic spline algorithm, or they can pay to buy something that does it for them like this: Multimedia solutions - Custom Technology Corporation

As for the VDub workflow it's simple.

1. Import interlaced high quality .avi file
2. Select "Resize"
3. Input desired size (720x480)
4. Select Lanczos as the rescaler
5. Choose compression type for output
6. Save new .avi file.

Done.

Perrone, what is your opinion of TMPGenc? It uses the Lanczos-3 resizer.

Never used it. Couldn't say. I do know that it costs money. And I get Lanczos resizing for free in Virtualdub. Along with a ton of stuff that TMpegenc can't do.

Perrone, with creating the initial .avi, what sort of hard drive space are we talking about here - what .avi settings are you using?

I use the Lagarith codec (works in 32 and 64 bit modes and is lossless). For audio I am using PCM 16 or 24bit 48KHz.

It's about 460 Mbps so works out to roughly 3GB/min. Which is about 1/3 what uncompressed 10bit AVI would be or just less than half what uncompressed 8-bit uncompressed would be.

Just a quick aside. I was invited to Adobe HQ today to give my impressions on improvements that can be made to CS5 (specifically, Encore). The product marketing manager, Sr. product manager and lead engineer from India were eager to get end-user impressions on improvements. I invited two of my colleagues and we chatted for almost 2 hours. I wish I had time to cull from this forum some of your user experience to present...but I did a good job of 'venting my spleen'. Number one on my hit list was Media Concepts atrocious encoder. I hammered that point for all it was worth. Of course, issues such as ease-of-use, glitches, menu transitions, poor quality previews, and much more was covered. If nothing else, I hope that Adobe seriously considers the acquisition or incorporation of Virtual Dub, Procoder, or another high-quality downconversion tool. I tossed out the idea that Premiere is almost useless as an HD cutting tool without aftermarket cards or codecs such as Cineform. I also held out high hopes that they would embrace 64bit Windows 7, GPU processing/encoding, and multithreaded support across their entire line. Don't expect miracles, we're talking Adobe here....but lets hope they take some of our 'mini-panels' suggestions and run with it.

Good on you for venting end user experiences and frustrations. I am a former Adobe user, and probably still would be had they not fumbled the ball in the end zone years ago and forced me to Vegas.

I wish more Companies would sit up and take note of how their products are ACTUALLY being used (and worked around) rather than just thinking they have a handle on things. I give the Vegas folks credit for reading their forum and being as responsive as possible. They might not get everything right, but they sure do listen.

If Adobe (Or anyone) acquired VirtualDub that would be awful. VirtualDub is beautiful in its simplicity and effectiveness, and it is free. I could see Adobe screwing it up. I would hate to see VirtualDub yanked off the freeware scene. It might just be the best free piece of software I own, I'd have to think hard for something better.

Ken

Actually, you wouldn't need to output your final edit as Cineform to send it to TMPGEnc, you could just use the Frameserver and send it without going back out to Cineform, that would save time and disk space, and the quality should come out the same. Actually, the quality should be better not needing to render out to a lossy codec again.

Good point. But I'm still looking forward to the day when I can do all my HD to SD workflow within 1 software piece that will give me the best quality (and hopefully GPU support). I doubt that the good folks making VirtualDub have the time/money/resources/engineering to improve much upon their product for the next version of Windows, 64bit systems, multi-threading improvements and GPU encoding. Adobe has deeper pockets...but let's all hope they don't screw the pooch on this one. Worse case scenario, keep your free copy of VDub handy when Adobe CS5 comes out.

Thanks for the excellent walk-through, Perrone. I guess I missed the earlier threads on this. I thought going straight from HDV to SD MPEG2 in Vegas was the cleanest route. But a quick test using your method looks very promising.

Actually, the Procoder downconversion is not considered very high quality. If you go to the Canopus/GV Procoder forum, you will see a similar discussion about using virtualdub to do the resizing, although Procoder is still used for the final compression stage.

Richard

I have spent all day today as well as last evening doing a comparison of various ways to resize and encode. My starting format was Cineform 1440 x 1080i HD clips. As a reference point, I did a MPEG-2 render using CBR with a bit rate of 9,000 kbps using the Main Concepts MPEG encoder in Vegas Pro 8. The result was the loss of resolution that is typical of HD to SD down conversion.

I then loaded CineForm files into TMPGEnc 4.0 and rendered them as MPEG-2 files. TMPGEnc uses the Lanczos resizing filter. The results of this method were clearly much better. The video was noticeably sharper.

I then loaded the CineForm files into VirtualDub and resized them to 720 x 480 and saved them as RAW RGB files. I then encoded the files to MPEG-2 using both the Main Concepts encoder in Vegas as well as TMPGEnc.

My conclusion is that most of the “damage” to resolution is when the files are resized when encoding with Main Concepts. The files that I encoded to MPEG-2 after resizing them in VirtualDub were of apparently equal quality when encoded with either the Main Concepts encoder or TMPGEnc. It appears that the culprit that degrades the files is the resizing operation in Main Concepts.

My question is: If I can do this in a couple of days, why can’t the “Pro” NLE companies do the same thing? I think it isn’t unreasonable for them to incorporate the Lanczos algorithm in their resizing “engine”.

Good question Jim, and I wish I had the answer. But I don't. What I do have is a workflow that I am happy with and works well. TMpegEnc seems like a nice solution for those wanting to go to DVD, but I often am not delivering that way, so I prefer doing things in VDub, and then making my SD master.

As I said before. The tools are clearly available, and relatively inexpensive (or free). So maybe the Big Boys will realize that this is a real need for some people and offer a solid solution inside the NLE.

This seems like a great tip and the Virtual dub software is in my price range.
One note, looks like it doesn't support M2t files so if you use a Firestore or some native HDV flavor, you're SOL unless you transcode. I've been trying to work natively in m2t, looks like it's back to the future with Cineform again.

I'm not quite understanding you there Jim. I'm not familiar with TMPGEnc - are you saying that it encodes MPEG2 with the Main Concept encoder as well?

I'm sorry I wasn't clear. TMPGEnc does not use the Main Concept encoder. To use the Main Concept encoder, I opened Vegas and loaded the resized RAW RGB file and rendered it to MPEG-2. The point I was making is that when Main Concepts encoded to MPEG-2 without having to resize, the video quality wasn't affected. The video quality loss occurred when it also had to resize.

Cheers for that - I'm with you now.

Jim when you encoded in Vegas did you have the deinterlaced method in the Vegas properties set to either blend or interpolate? If it is set to none it will look bad when Vegas does a downconversion. That said I agree that TMPEG downsized and encoded video looks better.

I also read in another DVD thread that HDV video that is uprezzed in Cineform (during capture) to full HD is supposed to look better. It's supposed to have something to do with the square pixels. Not sure if it's a big difference but I'm trying it on a project right now.

Regards, Marc

I didn't deinterlace, I left it is interlaced. I use CineForm NeoScene. It doesn't support upconverting from 1440 x 1080 to 1920 x 1080.

It's not a matter of deinterlacing. For some reason Vegas downconversions (when converting an HDV Timeline to SD MPEG for example) are affected by the deinterlace setting in the project properties of Vegas. If it is set to none it will look bad.

Following Perrone's basic workflow, I tried a bit of footage and am quite impressed with the results.
I rendered the same clean (no effects etc) 1440*1080i clip from Vegas using:
a) Main Concept DVD Architect PAL Widescreen Video Stream template
b) Cineform Neo Scene Codec at 1440*1080 (.avi)

Imported the Cineform file into Virtual Dub. Applied the resize filter (720*576 PAL) using the Lanczos3 filter mode - rendered back out of Virtual Dub using the same Cineform codec.

Imported both the Main Concept and Cineform/Virtual Dub files into a Vegas Pal Widescreen project and took these two frame grabs. I'm very impressed with the overall sharper image - and there is more detail.

Sweet!

Nothing like trying it for yourself and seeing the results in front of your own eyes. It's a pretty big difference. Considering that the Lanczos isn't necessarily the best at this, its remarkable how poor the downscaling is in the NLEs. And you used Cineform (twice), which is NOT lossless. I use lossless codecs which improve on the workflow you've used here. Not saying Cineform is bad, mind you.