Now that CS5 has been out for a while, its about time we revisit our old, yet still very active article, HD to SD DVD – Best Methods, and see if Adobe’s downscaling algorithms have improved since CS4.
The Comparison
This time around we decided to give Adobe Media Encoder the upper hand by feeding it 1920×1080 30p footage, which it can process much better than interlaced footage. Knowing that it was going to be processing progressive footage instead of interlaced, I knew going in that the results were going to be a little closer to our studio quality downscaling method using Dan Isaacs’ HD2SD script along with AVI Synth, Virtual Dub, and HC Encoder.
To get a basis for the comparisons, I first started by taking 1920×1080 30p footage into Premiere and adding a small title of sample text using a script font. Small script fonts are known for having problems with poor quality downscalers, as the thin parts of the character can show up very differently depending on the algorithm used. The downscaled text that most closely matches the original HD source is the winner.
So lets go ahead and show the original HD text that we will be analyzing (image is at 100% scale, cropped):
Original HD Source
As an added source of comparison, we then downscaled this HD source in Photoshop using the Bicubic method, and here are the results (no mpeg compression was involved):
Downscaled in Photoshop, Bicubic Algorithm
For our first official downscaled video sample, we will be showing you the studio quality version which uses the HD2SD script described in our previous article:
HD2SD Script Method - 7mbit CBR
The HD2SD script method looks almost identical to the Photoshop Bicubic method, with the Photoshop version having very slightly more sharpness to it. We also have to remember that the HD2SD method also went through mpeg compression at 7Mbit CBR, where as the Photoshop method did not go through any compression.
Now lets take a look at the best that the Adobe Media Encoder can offer, using the “Maximum Render Quality” setting:
Adobe Media Encoder - 7Mbit CBR - MRQ Enabled
As you might be able to notice, the text has become a little less visible using the Adobe Media Encoder and Maximum Render Quality enabled. Don’t forget, enabling this feature will cause it to take FOREVER to render. Completely unnecessary as our other third party (free) applications render much faster and look much better. You can also notice an increased amount of compression artifacts compared to our studio quality version, even though the bitrate is identical. The results aren’t terrible, but they do show how the original appearance and thickness of text was altered. In this case the text became a little thinner and slightly more difficult to read. Had the font been a lot smaller, its possible that it could have just disappeared entirely.
What happens when we don’t enable Maximum Render Quality in the Adobe Media Encoder? Lets find out:
Adobe Media Encoder - 7Mbit CBR - No MRQ
As you can see it had the opposite effect. The text became much thicker than the original, much of the contrast in the detailed areas has been lost, and of course there are lots of compression artifacts.
Scaled Results
In case your eyes aren’t as trained as a professional, we are going to focus in on some of the tall grass and blow it up to 400%. Lets go ahead and compare all of those again:
Downscaled in Photoshop, Bicubic, 400% (no mpeg compression)
HD2SD Method - 7Mbit CBR - 400%
Adobe Media Encoder, 7Mbit CBR, MRQ Enabled, 400%
Adobe Media Encoder, 7Mbit CBR, No MRQ, 400%
With this scaled up comparison, it is very easy to see what we were talking about before. The video encoded with the Adobe Media Encoder CS5 and using Maximum Render Quality has strange dark pixels introduced and very noticeable compression artifacts. When not using MRQ, the Adobe Media Encoder produced a slightly softer image, also with noticeable compression artifacts. Though I have to say, the artifacts visible when using MRQ appear to be worse, however the image is sharper… so I guess it is a trade-off.
Conclusion
It appears Adobe has not improved the downscaling algorithms in CS5. This is disappointing news, especially since freeware downscalers have been around for many years which can achieve studio quality results, and encode faster than the Adobe Media Encoder (even while running in native 64bit with 12GB of RAM). Are you ready to jump ship to Apple Final Cut Studio? Not so fast. Apple’s Compressor application actually performs worse than the Adobe Media Encoder. The best method is to of course use the free HD2SD script created by Dan Isaac, and discussed in our popular HD to SD DVD – Best Methods article.
Had we been using interlaced footage instead of progressive, the differences would have been much more noticeable, as the Adobe Media Encoder produces disgraceful and unusable results when downscaling interlaced HD footage without the MRQ setting enabled. We decided to use progressive as this is the format that most video production companies are starting to predominately use, especially with the DSLR boom.
Bottom Line: Avoid using the Adobe Media Encoder for any kind of scaling, where your output resolution is different than your source resolution.
We have heard a rumor that you can achieve great downscaling results in Premiere CS5 using the Mercury Playback Engine, and placing your finished HD sequence into an SD sequence, then scaling it down to fit. This, in theory, uses the CUDA Hardware acceleration to scale the image instead of Adobe’s default algorithms. Again, this is just a rumor we heard and have not tested this method, but we can save that for a future article.
P.S. – If all those comparison images looked the same to you… maybe you shouldn’t be doing anything related to video production. 🙂
Good information. Thanks for sharing.
By the way, love the P.S., it’s true!
Thanks for checking this – it confirms my own testings. But it is always good to have a professional validating such things. Thanks a lot!
Not wanting to start a anti-Adobe flaming but one can really not understand, why such obvious quality problems are not fixed over years. But at least in PPro 5.0.2 and AME 5.0.1 it seems that they changed something with the conversion of colorspaces, it looks more as expected now.
But for scaling and deinterlacing I stay with hd2sd from Dan, probably till CS8 or 9, at least.
I was at a presentation by Adobe this last week in Coventry UK and they had a few interesting things to say about this.
They blamed the poor downscaling in CS4 on frame blending was was on by default.
In CS5 they have brought the Frame blending tick box onto the main screen of Media Encode and this should be OFF unless you are transcoding across frame rates.
In addition they claimed to have changed the downscaling algorithm and media player is now using a photoshop algorithm and 32bit sampling when MRQ is enabled. They said that this had solved any issues with the quality of downscaling… hmmm we shall see – my 64bit machine arrives next week!
I also use your recommended HD2SD encoding method with splendid results. But I have to say it is a pain to have to come out out of Adobe and Dynamic Link to encode.
However the results are so spectacularly better I can’t see how I can justify doing anything else!
The best resizing algorithm Photoshop has to offer is Bicubic, which is still inferior to the spline resizing of the HD2SD script.
See results here:
AVISynth resizing filter comparasion
As you can see from the sample images in my article, Adobe still has a long way to go before the scaling issues are “solved”.
So now the question is, is it possible to use hd2sd with CS5? Now that Media Encoder is in charge…?
Or is the best solution to go back to CS4 when you have this problem?
We still use hd2sd even with CS5 for the absolute best quality downconversions… however, with the new algorithms offered by the CS5 Mercury Playback Engine, the results are so close that it might not even be noticeable by the client. In a recent blog post we link to an article written by Todd Kopriva explaining how to achieve the best downscaling using the Mercury Playback Engine. You will of course also need a video card that supports the technology… where as the HD2SD works on almost any PC configuration, as it is all software based processing.
Thanks for the reply Jon! But how do you use hd2sd with CS5 now that the old Export box is gone and it takes you straight to Media Encoder? The old Export box displayed the DebugMode FrameServer in the pull down…Media Encoder in CS5 does not…
Our original HD to SD article explains how to export using the AME, in the mpeg2 i-frame format at full resolution and 100 mbps. This essentially creates a visually lossless file which can then be integrated into the HD2SD workflow.
Hi Guys
Wow, what a thread! Everything I wanted to know about HD2SD but were afraid to ask! I am loving it – this has made my life a lot easier during the nightmare I have had with my current project in getting 2 hours of 1080i footage down to a fluid and sharp SD Progressive MPEG2 for DVD9. A few lessons learnt, and what a learning curve!
I have a newbie question for all you learned ones regarding scripting. I have 26 short clips, I have got them all to the Lagarith SD Full D1 stage using the HD2SD process mentioned here and am ready to put them through HC for the final encode to MPEG2 for authoring to my DVD.
I would dearly like to make a script that contains all 26 clips and HC would then take these and encode according to my settings and then join them at the same time and leave me with one single 2 hour long .m2v ready for multiplexing using Imago Multiplexer and then finally over to Encore.
Can anyone guide me as to how to do this. When recently searching for batch tips for HC Enc I stumbled upon a script someone had written to do exactly this but I can’t find it now I need it!
I am using the following simple .avs for single files in HC Enc, can I just add the other clips as a list? I rcall the sript I saw before looking something like a list of files..?
Here is my single file script:
AviSource(“JACK30.avi”)
ConvertToYUY2()
Any help truly appeciated 🙂
Thanks
Winston
Nash TV
http://www.nashtv.co.uk
ps – I assume I am doing the right thing in converting the colourspace to YUY12 as without this command the script won’t run. I am using Encore to author. CHEERS! 🙂
If you used Virtual Dub to downscale the footage with the hd2sd script into the lagarith codec, then in the compression settings you should have had the option to save it to YV12 (which is what you were suppose to do). Do not use any other colorspace. The default is RGB I believe. HC Encoder only accepts YV12, and if you properly output to a YV12 Lagarith avi, then you should not have to convert the colorspace.
Now that we have that under control, you can use the following script to combine clips:
AviSource(“clip1.avi”)+AviSource(“clip2.avi”)+AviSource(“clip3.avi”)
You might want to redo the hd2sd process into the correct colorspace so you are not having to do any conversions. If you don’t care about a possible loss of color data and want to just use the clips you already have, then you can just add the ConvertToYV12() after combining all the clips.
Hi Jon
Thanks for your super quick response!
With regards to the colorspace, I took my initial hd2sd script directly from the Jeff Bellhune tutorial, which reads like this:
SetMTMode(2,0)
MPEG2Source(“JACK30.d2v”)
hd2sd(interlaced=false, OutputColorSpace=”YUY2″)
I assume this is why my output is all YUY2, Jeff Bellhune states as he is making this script in the tutorial to do this. I just took this as good advice from a learned one – are you saying the opposite, ie, not to include the “OutputColorSpace=”YUY2″” command in my hd2sd script?
Also, from what I remember, I performed the original MPEG-2 100M I-Frame encodes from PPro CS5 using a YUY2 colorspace, again this is from a set of presets which I sure I got from your website or if I didn’t it was from Jeff Bellhune. So does this mean right from the PPro CS5 timeline into the first encode I was on YUY2?
If so do I continue in YUY2?
Thanks for your time here! 😉 I am so nearly there!
Thanks again!
Winston
Hi again
Quick correction to above post! I went and checked and there is no mention of colorspace anywhere in the original MPEG-2 I-Frame settings….sorry! 🙂
Jeff’s tutorial does not prepare your clip for HC Encoder, which is why he puts it in YUY2 (different Encoders use different colorspaces). For HC Encoder, you must convert it to YV12, and in the compression settings of VirtualDub, set the lagarith codec to YV12 mode.
So, instead of OutputColorSpace=”YUY2″ in the hd2sd script, you must replace that with OutputColorSpace=”YV12″
Doing so will properly prepare your footage into the YV12 colorspace, which is the only colorspace HC Encoder accepts.
Thanks Jon thats great!
My hd2sd script looked like this:
SetMTMode(2,0)
MPEG2Source(“JACK30.d2v”)
hd2sd(interlaced=false)
Do I need to add the OutputColorSpace command? I didn’t and re-encoded the whole thing one clip at a time but did select YV12 in the Lagarith setting within VirtualDub.
Is this ok?
Everything seemed to run ok and looks great but if I have not done something to the detriment of the quality of my final output I would like to know 🙂
Thanks again
Winston
Hey Jon,
I recently purchased your Regal Cinema Package, and I absolutely love it.
I have only had to create BluRays up to this point, and now I’m having to create a DVD from Sony HDV Footage. 1440x1080i 1.333 anamorphic Upper Field, 29.97fps. I believe its called 60i. Anyways I’ve done everything mentioned in your blog here, and have everything loaded in properly, but I’m having two issues.
1. Wrong aspect ratio. (Getting black letterboxing on left and right sides with footage being “squished”.
2. and more importantly, I’m seeing no picture quality improvements in my tests. This could be becuase of PAR though.
Here’s my workflow:
– Sony Vegas 8.0c – Project Settings at HDV 1080 60i (Basically outputs in orginal format)
– Debugmode Frameserver (YUY2 – with audio in signpost)
– VirtualDub with AviSynth Script with “Fast Recompress” and “Lagarith” Codec. Set with “YUY2” and Multithreading. Open with the .avs file.
hd2sd(“PANNATONI CEREMONY.avi)
– Imported into Encore CS5 DVD Project (Aspect ratio problem becomes apparent at this point) Transcode setting are 720×480 16:9 29.97 Interlaced 7.0CBR. When setting the transcode settings the source file appears correct in input window, but output window shows letterboxed. It did this also with default DVD transcode settings before anything was touched.
Thanks for your help, and cannot wait for the Corporate Menu to come out…
Lonnie
Lonnie,
The avi saved out of Virtual DUb will not have the PAR information in the file, which is not a problem, but you just have to make sure that your footage is being interpreted correctly in Encore. Right click the imported file, select interpret footage, and make sure the aspect ratio is set correctly.
Jon,
I did not see a specific setting for PAR in Encore transcode settings. Only the 16:9 or 4:3. No 1.333 which is what the original footage is. Am I missing check box or tab that brings this up? I appreciate you answering these questions on an old blog. It really helps out. I can’t have your crisp menus making my footage look bad. Lol.
Lonnie
Lonnie,
It’s not in the transcode settings. Simply right click on the clip in your project panel, then select ‘Interpret Footage’. You will then have the option of choosing how Encore interprets the file (as either 4:3 or 16:9). This is separate from how it gets encoded.
If Encore thinks the file is originally 4:3, then tries to convert that to 16:9, you might see the problems you are having. If it thinks the original file is 16:9, and is encoding to 16:9, you should be good to go.
The interesting thing I noticed is that Adobe Media Encoder yields completely different results than Encore with the same settings.
I’ve been trying to get acceptable results from 1080p WMV files I wanted to downscale and master into a DVD. I used CBR 9mbps encoding with MRQ on in Encore, and the results were crappy, with blurry and aliased text. Then I found your articles and wanted to try it the long way, with AviSynth and stuff… but I tried one more thing.
I put the same files into Adobe Media Encoder and created a preset with the same settings – MRQ enabled, CBR 9mbps, and PAL output. And guess what? The result is more than acceptable! Maybe not that beautiful as with your recommended way of scaling, but all the blurriness, aliasing and other problem is gone.
I have no clue why is that, granted that both Encore and Premiere use AME internally… I guess I won’t be trying to understand what’s going on. Adobe sometimes really can’t do things right.
Another reason why we recommend encoding your main videos in the Adobe Media Encoder BEFORE bringing them into Encore.
“Another reason why we recommend encoding your main videos in the Adobe Media Encoder BEFORE bringing them into Encore.”
what about if you dont use Pr ? (i use Vegas) … and I need to be able to place frame accurate chapter markers in Encore by hand (not doable with a per-rendered / compliant file)
Gentlemen – here we are a year+ since the former post and I am seriously wondering and hoping some solutions have been found. For me, working with Cs5, I am still plagued by the jaggies in Encore made SD-DVD playbacks on a Sony Bravia 40″ HDTV, using a Sony up scaling standard DVD player from AVCHD interlaced timeline originals. Yet none of those artifacts seem present in commercial DVD playbacks. I am further bummed out by Bellune’s hyper techie and oomplex hoops to jump through just to be able to produce a decent DVD. Re use of the Mercury Engine, I find it’s primary advantage is in helping with navigational speed and rendering, not in making any difference in image quality.
For whatever it’s worth, similar discussion with another buddy who is a Mac user, he states he never has these problems, given he says he always shoots, edits and delivers in progressive. In short he concluded by saying, the quickest fix is to “Get a Mac”.
So – back to my original question, have we really made any progress since the last posting?
Hi Bob,
The Mercury Playback Engine actually does a decent job of scaling. The trick is knowing how to render with it. Your video must be rendered from Premiere, or from a Premiere sequence queued in Adobe Media Encoder. If you load a video directly into the Adobe Media Encoder or directly into Encore, it will not get scaled with the CUDA hardware scaler. If you edited your video in HD in Premiere, you can simply export to Mpeg2 DVD and it should use the CUDA scaler if you have the hardware to support it. You can also import your HD video into Premiere, create an SD sequence, scale down the video, then export to Mpeg2 DVD.
Regardless if you then render directly from Premiere or Queue it in Adobe Media Encoder, the CUDA scaler will be used.
The results are not quite as good as the HD2SD method, but the client may not even notice the difference. I recommend upgrading to the Adobe Creative Cloud so you always have access to the latest version of the software.