HD to SD Downconversion
 

Ok,

So in the past week or so, there has been a lot of talk about the footage coming out of the EXx cameras not converting to SD very well. I tried the experiment for myself last night and mine worked ok. So tonight I thought I'd give something a go.

I took a reasonably clean 1080p interview, and cut a 5 second snippet out of it. I then resized it to 1280x720 and made an .mp4 file as a control. After doing that, I created a 16:9 720x405 file. From that, I created an mp4, and an mpeg2 file ready to drop into my authoring software.

I uploaded all three files on Youtube, and onto Vimeo as well. You can download all three files from Vimeo from here:

Perrone Ford on Vimeo

The Youtube site is here:
YouTube - perroneford's Channel


The workflow:

1. Load 1080p .MXF file into Sony Vegas adjust exposure.
2. Render 1080p uncompressed .AVI file
3. Open 1080p .avi in Virtualdub. Resize to both 1280x720 and 720x405 and save uncompressed versions
4. Open 720p file in Vegas and render .mp4 at 6 mbps.
5. Open 405p file and render .mp4 file
6. Open 405p file and render widescreen mpeg2 file
7. Upload files to youtube and vimeo

From my testing, I am able to VERY clean SD files from 1080p sources with a minimum of fuss. If virtualdub understood the MXF files, it would save a step, but it does not.

This also show me that the creation of SD footage is NOT tied to the camera or the XDCamEX codec. It is entirely possible to get lovely footage at SD sizes from the 1080p original.

I will do a file replacement on the Vimeo site for the 720p file so it can be seen there in HD. The Youtube one shows properly but one of the 405p files is not yet showing.

I hope this helps out.

So I'm back at my assumption that you don't know what you're doing.
Resizing 16:9 to 720x405 for DVD-use just doesn't make much sense.
What's exactly happening in this step?
Any resizing/resampling of the pixels?
If no: It will look distorted on your TV, because DVDs don't use square pixels, but your 405p-file does.
If yes: down to 360 or up to 480 lines?
If down to 360 lines: you're using letterboxing instead of true widescreen.
If up to 480 lines: You're interpolating 75 extra-lines from the 405 initial ones instead of going directly from 1080 to 480. That's sub-optimal.

Thanks for your detailed workflow, I am sure it must work otherwise you wouldn't have posted the formula.

On trawling through this site and other I have found many workarrounds to the problem of converting HD to SD - most seem to be long winded and rather defeat the object of working with Solid State memory. Over the last two days I have burnt over 20 test DVDs trying to find the answer, each DVD had its own set of problems, the main one being artifacts on fine detail. My main reason for choosing the EX3 was that I could shoot in NTSC or PAL format, for my DVD sales the North America market is very important, if this were not the case then I may have purchased the Z7.

The sollution came late last night. The EX3 has a built in downconverter, which will output files in the SD format. You do need to have a video capture card which will accept the Component video cable (RGB cable). I have a Matrox Rx2 card. From the Video Set menu select YPbPr/SDi OUT and select SD. From your NLE select Capture and set the EX3 to Media, select thumbnails and choose a video clip, press the play button and start recording the video to your computer. The footage will be added to the clip bin in much the same way as capturing from tape.

OK, so it isn't as convenient as being able to drag a file straight from the card, but at least you get to review the footage as it is being imported.

I will continue to find a better solution, I am sure you must be able to to downconvert a file from one memory card to another, if this is the case then you should be able to drag the files directly into the timeline.

Finally, the footage captured using my technique above is perfect, certainly as good and better than anything I previously captured using my Canon XH A1.

I used footage that was shot at 1920 x 1080, 1440 x 1080, 1280 x 720 interlaced and progressive, they all worked perfectly. My tests were all done using NTSC, but I will repeat them using PAL footage at a later stage.

ok, let's memorize for a minute, the problem are artifacts on fine details. Perhaps you're using a bad mpeg2-encoder with bad settings, but who cares...
So this back-to-analog workaround doesn't show artifacts on fine details? Did you ask yourself the question: Why? Because there's analog bluring and sub-optimal downscaling of the ex3 happening which finally entirely destroys the fine details? Or noise and other analog artifacts cover the initial problem of artifacts on fine details?

Converting HD to SD isn't a problem.
Problems are:
-the halting problem
-a lot of people don't know the basics of digital media
-people who don't know those basics are answering questions in forums like these, so a lot of confusion gets into them.

Domininik. I know you are trying to help (me and others) here. So I'll be brief. These files are going to youtube, or being tested on my PC. Which uses square pixels. When going to DVD for TV, I am not using square pixels. My tests are not designed to be pixel aspect ratio correct, only to be able to determine if we are losing resolution along the path.

I invite you to view this file which I uploaded some hours ago and I assume you have not seen:

Interview 405p Mpeg2 on Vimeo

That is a REAL render for DVD output from a 720x480 timeline with a 16:9 source.

Here is another I was working on as I was typing:
http://www.vimeo.com/2581948

Ok I believe you, that you know how to do it.
But then what's your posted workflow for? It's not for people who want to do DVDs, though the title "HD to SD Downconversion" let people suppose that, and it's not for people who want to upload youtube-videos, because youtube supports 720p.

This is why I said yesterday I didn't want to get into the whole pixel aspect ratio thing. This workflow is really just so people can compare what they see on the screen between their 1080p or 720p out of the camera, to a rendered mpeg2 file which would be the end of the workflow. Because these comparisons are done on the computer (and not burning a new DVD every time they want to compare) I thought it might be helpful.

So I'll say it here. If you are ACTUALLY BURNING A REAL DVD, keep your project at 720x480 for an NTSC SD DVD. PAL folks can adjust as necessary.

What I am hoping to glean, is whether people are losing quality in the step where the video gets scaled from 1080/720 to 480, or if they are losing the quality from their mpeg2 codec. In following this workflow, if it's the scaler, then step 5 is going to look bad. If step 5 looks solid, but step 6 does not, then it's the codec. If we look bad at step 4, then we have other issues like a possible interlace problem, or something else going on.

Make sense?

If people would start to post stills from their (problematic) mpeg2-results instead of writing a thousand dubious words, we could immediately tell them where the problem in their workflow is. But posting stills from mpeg2-files seems to be even harder for them...

Some have indicated that they don't see the problem on stills on the computer. So my next best thought was to take the thing all the way through mpeg2 render. If it looks good all the way through mpeg2 on the PC, then the problem is hardware / firmware related down the chain.

What I found mystifying was people saying that the CAMERA was not giving them good SD results. I just couldn't fathom that at all. If the camera is giving you stunning 1080p footage as everyone claims, then the fault is not with the camera or the codec. It's elsewhere in the chain.

Here is the 720p downres of that second interview clip: YouTube - iv2 720p mpeg4

Perhaps we see. ;)
That's why I said "stills from the mpeg2-results", so the very end of the chain.
Absolutely. Btw, did you know that your choice of the lanczos-rescaler is a very good one, because the lanczos-rescaler is the best rescaler actually programmable? :)

Surely that is why we come to forums such as this one, to find out what all the settings mean. I will be the first to admit that I don't know all the basics, but I am learning.


I have no problems with the camera, it is a fine instrument. All I want is to be able to produce decent quality as an end result - this may very well be down to the codecs, software etc. I am still finding out.

You are right my DVD has been downgraded in quality when compared to the original footage, but then I would expect this from SD. I am not 100% there yet, but nearer today than yesterday.

Which software do you use? Windows or Mac?
The first hint I can give you is to try to use some blur to make the edges softer. That will reduce mosquito-noise around edges and flickering on CRT-TVs. Also, if done before downscaling, it will reduce aliasing introduced by low-quality downscaling.
Please forget your analog workaround soon. It's far from being reasonable in any way.

I use Premierre Pro CS3 (Windows) and Encore CS3. I also have a Matrox RTX2 card with its own set of Codecs.

Up to now I have been bringing the video files directly into the timeline and then using the Adobe/Matrox Media encoder to output the file ready for burning in Encore. The Media encoder Transcodes the file on the fly. Premierre also has a XDCAM EX preset which does work, although it still relies on the Adobe/Matrox media encoder to produce the DVD ready file. I suspect the encoder may be at the root of the problem.

Perrone, maybe I missed something along the way in your explanation. When making DVDs, why not render straight from the HD timeline (mfx) using MainConcept MP2/DVD Architect NTSC Widescreen Video Stream?

Perrone...

I really hesitate to post much of my own experiences here, but, I'll give it a try. As of late, it seems people on this forum have become quite rude and intolerant of opposing viewpoints.
Having said this, I've played with both Vdub workflow scenarios, as well as other techniques. As far as vegas is concerned, I can achieve results with Vegas that are comparable with Vdub downrezzing by generating the HD .veg file, then opening an SD project and bringing the HD.veg file into the SD project, then rendering in SD.

Also, many people, here, seem to like to use sharpening when shooting with the EX1. As you know, a DETAIL setting of '0' adds sharpening to the captured images. Sharpening done in-camera, is to be avoided as it significantly adds to the twitter problem people experience. For some reason, they refuse to accept that better results can be obtained by turning DETAIL off and sharpening in post. It doesn't help, also, that in camera sharpening adds to the work done by the camera compressor algorithm, sucking up bandwidth that would be better used on image detail, not compression/DETAIL artifacts. There are supposed "experts" here that really believe there are no compression /motion artifacts with the EX1 codec.

Because it means that the encoder has to do two jobs. It has to downres the video, and it has to encode it. I don't know what method the Mainconcept encoder uses to downres, so I like to control that step myself and use the best rescaler available.

Well, after much research and testing, I settled on it. Though I would like to find a bicubic spline rescaler to test againt but no one seems to have one. It would not be nearly as fast though. The lanczos also adds some subtle sharpening, so I wouldn't use it for everything.

Dominik-
Instead of taking potshots at Perrone's method, please give us your version of a step-by-step process of converting HD to SD. I'd love to see how you do it.
Pete

This doesn't surprise me. But I tend to like to know what's happening behind the scenes so I use VDub. There are likely a number of ways to solve these issues and get good SD results. I merely posted mine as I have not had any negative results from deriving SD from the XDCamEX cameras.

This is a shame. The EX1 is the sharpest camera in it's class. It needs no sharpening in the camera at all. My results are with sharpening off, and I don't do any in post either.

Well that's silly. That's why it's called compression. And artifacting is clear and obvious under motion. Just grab a still and zoom in 400%. You'll see all you want. My interview footage is a great example. As I adjust in my chair, you can see the codec get overwhelmed and go soft.

It's ok Clark. I know why he was on my back. But I would love for him to share his method here.

Lanczos is the best because it approximates closest to the Nyquist-Limit without aliasing. That sharper look isn't actually sharpening but just real unboosted high-frequency-information which can't be seen by more high-frequency-muffling methods.
I'm sorry, I initally misunderstood his intention.
My method is actually very simple, because I'm in the lucky position to use Apples Final Cut Studio Pro. Its Compressor Application does in one step excellent rescaling and good mpeg2-compression. What I've actually to do is to pick a dvd-preset, set up an appropriate bitrate, set the rescaling quality to best and press start.

One would think that after all those various generations that one would be moving farther away from the images original quality. Something has to be lost in the process.

It's hard to judge on the Web, but the video samples on both YouTube and Vimeo look soft on my monitor.

Then you clearly don't understand what "uncompressed" means. If I copy a data file from computer to computer 6 times, does it mean it changes from what I started with? This is no different.

Yes, Perrone, I know what "uncompressed" means. But thanks anyway. As I said, above, your Web samples do not support your methods.

Adobes MediaEncoders downscaling produces some visible aliasing. Apply gaussian blur with a strength of about 3.0 on the timeline (not in the mediaencoder-settings) and try again. That should result in better results.

hmmm..I would distinguish between "soft" images and images that demonstrate twitter or flicker. Perhaps I misunderstand the nature of people's discontent. "Softness" is acceptable, to me, provided it isn't as soft as DV material, but on an SD display, some softness is understandable. Twitter or flicker, OTOH, is extremely distracting, annoying and unacceptable for a quality presentation. Even native progressive footage will show twitter/flicker in SD, if one isn't careful.

Well, ANY codec is going to soften footage when going from 1080 to 480. Especially given that your monitor can resolve more than SD can offer. But output that footage to an SD TV and it looks marvelous. Laying my footage back onto the timeline and scrubbing it while looking at my SD broadcast monitor is very nice. I also wonder if you have closely looked at the 720p version. The original footage was "soft" because the focus was out. But the SD looks remarkably true to the original footage. Download all the files and view for yourself.


And if you do understand uncompressed and generational loss, why are we discussing it?

Thanks, will give this a go

Dominik, would you say that MpegStreamClip also does a good downscaling HD > SD? Do you know which algorithm is applied (the famous Lanczos)?

I am asking coz I intend the use Cinema Craft encoder right after New Year, they offer a Compressor plug in now.

I don't know the exact method, but it produces a bit aliasing. In my opinion Compressors rescaling is better. Here's an example, which I previously posted in this forum. Upper one is streamclip, lower one is compressor. The black bars are produced by streamclip because of a reason I don't know.

Danke sehr!

I couldn't stand it I had to post...LOL I know nothing about this but this is what I did.

I did a MXF full res 1980x1080 edit, rendered as full res MXF.

Reset project to NTSC DV widescreen 720x480, EX detail off and on, rendered as a DVD architect 720x480 widescreen mpg.

Pulled the mpg back into the timeline and snapped these.

Thank you Eric, for posting the two sample pictures.

After much experimentation and trying out most, if not all, of the suggestions made by contributors to this thread, I have found the best solution for producing DVDs.

From the Sony Clipbrowser 2.01 select a clip and Export it as a MXF for NLE and drop the resulting MXF file directly into your NLE applications timeline. With your editing completed send the file out to your DVD authoring application, result = superb detailed SD DVD production.

Over the last four days I have created about 25 test DVDs using a combination of suggested workflows and my own. The Export as AVI DV from clipbrowser 2.01 produced a soft mushy picture. The downconvert to SD from the EX3 produced a similar result, although slightly better than the AVI DV, but still didn’t do justice to the EX3.

It strikes me as odd that many people are adopting a complicated, or time consuming work around, when this camera offers a fast time saving workflow, i.e. direct recording to memory card. The MXF to NLE adds an extra step, but the end result is superb.

I have had the camera for four days now and despite my initial disappointment , it is now producing the result I was hoping for. I still have a few other problems to sort out, i.e. blurry panning, but this could be down to me and my panning speed etc. I will start a new thread on this subject.

Thank you all for taking the time to offer your advice and suggestions, it is much appreciated.

Seasonal best wishes to you all.

You aren't sending MXF to the authoring application so what are your project settings and render settings?

John

No, I just select export to MXF NLE and save the file to a new folder, in Premierre I start a new EXCAM EX project and import the MXF file. I export the finished project file using the Adobe Media Encoder. Open up Encore and import the file - simple as that. The quality is stunning, I have produced many DVDs in the past using stadard DV tapes etc. I am comparing the quality next to these. I still think there is room for some improvement, especially with the panning shots, but I think this is another issue.

Vincent, are you using premier pro cs3 or cs4 ?
If so then you don't even need to rewrap the files to mxf, just copy them to your pc.
Use clip browser to copy, drag and drop from the top explorer widow to the bottom window/destination folder.

But apart from that you're doing the same as me, i use vegas and dvd architect, nice and simple, good results.
I also cannot understand why people are going through multiple programs to get good dvd's.

Paul.

Paul,

Yes, I know that you can just drop the files directly into Premierre (I am using CS3), but |I was having problems with my Matrox RT X2 card and MPEG files, I was getting a strange frame every now then have a look here.

http://www.dvinfo.net/conf/attend-wo...s-problem.html


This is why I decided to use the EXCAM EX project rather than the Matrox HD project. Having said that the fault still occurs in Matrox HD but not in EXCAM EX projects. I am trying to find a sollution to this through the Matrox help line, but in past experience I will be waiting for a long time for a reply.

I agree with you that there seems to be so many unecesary work arounds about, when it should be an easy process.

Thanks for your reply

I agree with Vincent, with a caveat. Softness has never, ever been a problem for me. Downconverts with a quality downconverter yields results considerably better than native DV source files. However, I experience line twitter/jitter....that flashing that happens when horizontal lines in the image move vertically inside the frame. This is a vexing problem. I just don't understand what the big bruhaha is over softness since I don't see it. I always work with Cineform intermediates, so, perhaps this softness issue is related to Long Form GOP, donno, doncare.

Does anyone else experience line twitter? Speak up or forever hold your flicker.

The only time I've ever seen it was working on someone from this site's problematic footage. But it was interlaced and clearly sharpened in-cam.