Downconverting XDCAM HD to SD - problems?
 

I've heard that there can be problems downconverting HD material to SD, especially if there's a lot of fine detail in a scene. Does anyone do this on a regular basis - have you, too, found problems?

I guess if I know for certain that the material will be used in the SD realm, I'll just shoot in DVCAM mode - but it would be nice to have the option and not to worry about conversion issues.

I guess it's the same issue that faces other conversions i.e. how do you represent single pixel detail in less than a pixel! ;)

Thanks,

Paul.

Well, simply put, you cannot retain all of the information and detail of a pristine HD source in an SD downconversion.

If you use a very good resizer / deinterlacer, you can get excellent results, but SD will never be HD.

An ideal workflow IMHO would involve AviSynth - Load up the HD source, if it's interlaced, deinterlace it with TDeint() or MVBob.SelectEven(), and then use a sharp resize like LanczosResize(720,480) or Spline36Resize(720,480).

If you need to deliver 480i, then you can use TDeint(mode=1) to bob, or omit the SelectEven() from MVBob to get SD 59.94p output, which can be re-interlaced to 29.97. Doom9 has lots of good info regarding up/down conversion

Thanks Derek.

Please always remember, that there are users from all over the world in this forum. The original poster comes from Great Britain, and so he need to transfer his material to a PAL target: 25 Frames, 576 Lines and so on.

There is a world outside NTSC... :-)

Matthias

>how do you represent single pixel detail in less than a pixel?

Blur (apply 1 pixel, or even only 0.5 pixel, of blur to your image in the NLE of your choice) then downscale.

mmm - ok.

How does that work for you in comparisson to a straight HD to SD convert? Do you find the result much softer?

I'd have to say not. Any percieved softening is counteracted by the sharpening implicit in the downscale process.

That said, I offer it only as a suggested workflow which you may want to try.


You don't mention your target SD format, or your platform, NLE or other details, so I went for a very basic one word answer for you!

I work in FCP, and use a 0.5 pixel channel blur then export a QT Movie w/ Current Settings. (ie I keep the QT export as HD)
Following that I use MPEG Streamclip to downscale the QT to a DV PAL AVI file, I enable the "Better Downscaling" option in Streamclip's AVI exporter options.

The results work well for me. Without the blur, fine detail can cause the image to "sparkle".

OK thanks Andy - some good things to try there :)

I must say - I don't currently have an XDCAM camera but am seriously thinking of going that route shortly, hence the questions.

I'll also wait to see what's announced at NAB before jumping in too! I've heard various rumours...

Rumors are rumors, Paul.

I am in the NTSC world, so I cannot speak for the issues in PAL land but, I find that the downconverts from HD to SD on my F350 look better than material generated as SD (DVCAM). That may not be the case in all situations.

Paul, people in the film production world have been shooting HD for SD since 2001 or so, when the F900 and Varicam became available en masse.

There's a whole business of national and regional commercials, most of which used to be shot on 35 and 16 only, but because of shrinking budgets and the availability of HD 24p, started looking to HD. The F900 was the first video camera to give these folks the 24p they needed, so making good, high quality downconverts became the norm. I jumped from the film business to the video business as this was going on.

HD to SD downconverts (good ones) are better than shooting SD in the first place because of the concept of oversampling. If you take a huge, high-resolution image, and shrink it, you minimize a lot of the undesirable compression/colorspace artifacts that came with the original HD format. By the way, this is also why telecined 35mm looks so good as well...it virtually becomes grainless when all that information is shrunk to NTSC or PAL resolution.

So in the case of the 350, you're much better off shooting HD and downconverting to SD. The resolution might be the same in SD mode (because the image off the chips is still oversampled down to 720x480), but then that oversampled image is compressed to DV, with it's 4:1:1 color resolution.

If you do that same oversampling trick by shooting HD, and downconverting later (in your computer) to a format better than DV (like uncompressed SD), then the 4:2:0 color resolution of XDHD becomes an effective color resolution of 4:2:2 or better, a side effect of the image resizing. You also skip the DV compression. If you see this often enough, you get to the point where you can spot it quite easily; DV has a "coarseness" that you can spot after you get a good look at something better.

One other thing, mentioned by Andy above that often doesn't get mentioned (except when I'm rattling on about it). A lot of cameras don't re-introduce edge enhancement to the downscaled image on their downconverts...I have yet to see an HDV camera that does this. The XDHDs do, and you have 3 steps of adjustment. HDCAM and D5 decks have this as well, as well as standalone boxes like a Terenex. The in-camera edge enhancement gets averaged out in the image re-size, so often it's necessary to re-inroduce it on downconverts. Edge enhancement is something that's added to even a lot of film material coming out of a telecine these days, so it's important to keep it in mind in your downconverts.

I use the camera to downconvert to SD. There are menu settings for SD detail (high/mid/low) and Cross color (on/off) which are applied to the downconverted signal.



-gb-

I'd certainly agree that a hardware downconvert, including those available in your camera or deck, is almost always better than anything acheived in software.

Thanks Nate/Greg :)

How about chromakey? - If I wanted to end up with web based material, say at 400x300 for instance, would I be better keying in HD, SD or 400x300? (I'm thinking especially of what you said about the colour sampling).

Thanks,

Paul.

Yes, but Nate makes a good point about doing NLE downconversion to a higher color space SD, like uncompressed or DVCPRO50.

Part of it depends on how big of a rush you are in to get the material out.

-gb-

I don't understand much about colour space issues and how this is improved by downconverting - is there a primer or other good info out there that I can read up on?

no argument with that. hence why I use a software downconvert ... the story needs feeding asap !

don't forget though that the F70 for example does a high quality downconvert to uncompressed SD, so I'm not suggesting using the downconvert via DV per se.

Absolutely. I keep forgetting that the deck can output SD SDI. Until recently, I sincerely believed the camera was capable of the same.

-gb-

First you need to understand the basics of chroma subsampling. Color space is a different issue (RGB vs. YUV) and I used that term mistakenly in my post above.

Chroma subsampling refers to how much color information is thrown out in the interest of reducing bandwidth of the signal so that it can be reliably recorded on the medium. It's a cost decision on the part of the mfg as well as market segmentation.

4:4:4 means as you go across one horizontal line a distance of 4 pixels, you have 4 luma samples, 4 red samples, and 4 blue samples.

4:2:2 means in the same 4 horizontal pixels, you would still get 4 luma samples, but the red would repeat for two pixels, as would the blue. This reduces the color resolution, but saves on data volume.

4:1:1 means in the same 4 horizontal pixels, you still get 4 luma samples, but the red repeats for 4 pixels, as does the blue. This further reduces the amount of color resolution, but saves even more space.

4:2:0 is a different animal. It employs an alternating line strategy where as you have 4 luma samples then 2 red samples, and no blue samples ON ONE LINE. On the next line, it would give 4 luma samples, then 2 blue samples, and no red samples. The pattern alternates the red and blue going down each line through the frame.

Both HDV and XDCAM HD use 4:2:0. When you down convert, you squeeze all that information into a smaller space so the color resolution has an 'apparent increase' as Nate says.

Hope this helps,

-gb-

Does the 350 output a downsampled SD signal via the HD-SDI output (or component output on the 330)? If so, that might be another workflow. Of course you'd loose you timecode reference and clip organization, but in select circumstances could be a way to avoid conversion within the computer.

No to downconvert out on the HD-SDI of the 350. Dunno about the 330. Something makes me think that WOULD work.

That's the problem with hardware downconvert on the camera, on the 350 you only get composite which is no good for post.

I kind of also disagree about hardware being better than software for downconverts. Hardware would be better in the case of a Terenex box with HD-SDI, but not as good if we're talking about the composite out on the 350.

With the software tools I have, I have a lot of control over the process, including fine control over putting edge enhancement back in.

Not to mention that software is just the most convenient place for it to be for a lot of my work. If I had to run HD-SDI out in real time to tape, and back in again just to author the DVD project I'm working on, I'd go insane.

If I need to run off a Digibeta or SP, all I have to do is export my timeline and go to Starbucks (er, and have a deck rental delivered :-)

I'm interested in the SDI out of the 350 as well - what EXACTLY does it give you - compressed or uncompressed?

Also I'm led to believe it's AFTER the compression sectionof the camera i.e. you don't get anything better than the original 4:2:0 - or is that not true?

Wrong! SDI is ALWAYS uncompressed 4:2:2. Dual HD SDI can give you 4:4:4. However, to capture HDSDI, you need to have a capture board and a RAID array to be able to handle all the data coming in at that speed.

Think about it, Paul. What good would an SDI connection be if it were compressed and chroma subsampled like the other options.

-gb-

OK cool - so provided I had a PC with suitable capture card and RAID array, I could capture to disk in the studio and get a better key with chromakey than with the XDCAM recorded material?

I was confused as I asked this elsewhere and was told I wouldn't get a better result.

Oh, and what kind of data rate do you need for HD SDI - in MB per sec?

Compressor 2 from Apple's Final Cut Studio does a fantastic
job downconverting HD XDCAM footage to SD. If you are editing
in HD XDCAM and your final goal is to put it out on DVD, then
do a straight downconversion to SD MPEG-2. Progressive HD video
downconvered to progressive SD looks great. Adjusting the different filters will yield different results. This advice was given to me by Nate Weaver and it has worked well.

I have had XD HDCAM footage tested/converted with high-end Terenex eqipment. Although impressive, not needed unless doing any live stuff.

The confusion may be coming from whether or not you take a live feed from the camera, or play back the disc via HDSDI. In the latter, the 4:2:0 subsampling and Long GOP MPEG2 compression has taken place already. In order to fit the HDSDI spec, the disc playback would be upsampled back to 4:2:2 but won't be as good as recording it natively via a live camera head feed.

-gb-

Yes that makes sense thanks.

You need OVER 150MB/s for 1080p uncompressed :)

It's quite impractical, but that's where lossless codecs like Sheer and Cineform come into play.

Also, the comment about blurring before downscaling seems awfully misguided to me.

If you're getting aliasing or other resizer related artifacts when doing a large downconversion, then the answer is to switch the resize algorithm. I don't know what algorithm your method uses, but it's probably Bicubic (sharp?), as that's the standard.

A bilinear resize would probably be a little bit softer without all the aliasing, but much better than blurring out all the details before you downconvert! Especially a dumb blur like the one FCP has.

I hate to chirp in and add some Windows workflow, but if you have a Windows system available, and want to do the best possible down conversion possible (in software, without something like a Snell & Willcox Alchemist), then AviSynth is your best friend. It has a vast selection of resizers all the way from a primitive point resize, to the more common bilinear and bicubic, to the more advanced lanczos3, lanczos4, spline16, spline36, gauss, etc... The consensus seems to be that for large scale resizing, spline resizers maintain detail without introducing sharpening halos.

If you want a sharper still result, you can sharpen the HD input first with something like LimitedSharpenFaster, and then downconvert.

It's also got some wonderful deinterlacers, and denoisers. forum.doom9.org is the unofficial 'home' for AviSynth usage and development. Everyone should really give it a read. It's a forum mostly targeted towards DVD backup, but they really know their stuff about processing, and the pros could really learn a thing or two from the open-source guys :)

~Derek

@Greg:

Thanks for the Color Sampling information - this was the best (and easiest) explanation I´ve ever read.

Uli

Thanks for the kind words, Uli. We're here to help each other and that's the best compliment one can receive.

-gb-

Nate, can you point me towards the edge enhancement setting for down-converting to SD in the HDcam(F350)? I was looking through the manual and couldn't find it.

Thanks Peter

Downconverting with PDW-F70
 

In my experience downconverting XDCAM HD with the F70 gives a somewhat poor result compared to DIGIBETA or even DVCAM. To confirm my doubts I captured the F70s SD-SDI (sqeeze mode, high detail gain) to an AVID MediaComposer (uncompressed), exported a frame as TIFF and opened it in Photoshop. Compared to the same frame captured via HD-SDI to a SMOKE-System and scaled down in Photoshop to the exact size of the SD capture I observed on the SD frame:

- an obvious lack of detail in fine textures (skin, hair, trees, fabric, ...)
- a number of white stripes in the first half of the upper line of the frame
- blacked out pixels on both sides of the frame (4 on each side)

When playing around with the menu settings, I found the best result could be achieved with DETAIL GAIN set to HIGH. The CROSS COLOR adjustment didn't make any visible difference.

After mastering to HDCAM there was a conversion done by that recorder to DIGIBETA. The result is very pleasing.

The camera we shot with was the PDW-F350 in progressive scan mode. In the original HD frame (scaled to 1920 x 1080) there are two pixels on the left side and one pixel on the right side that are black. I can't say if it is the camera or the SMOKE-System used for finishing.

Tom

I shoot NHL hockey news in HD with my 350, and down convert using the camera. As far as I can tell it looks better(quite a bit) than my old BVP90/BVV5 SD combo. I generally use the mid level sharpening. I just wish that I had the SDHDI to output the video in component form.
I have also seen some of my stuff cut together with IMX camera footage, and I thought mine came out looking better. They dubbed my composite out using an IMX machine, and later edited using an Avid system.

What about AJA equipment?
 

Right now, we're planning on shooting EVERYTHING in HD, and if it needs to be down converted to SD, our AJA Io-HD does that quite easily. We're still waiting on delivery of our first XDCAM (F355), so my knowledge is ALL theoretical right now. Are there going to be any problems shooting and editing in native XDCAM (on Final Cut) and then simply dumping a letterbox version to Betacam via the Io-HD to cover any SD distribution needs we may have?