Why would you write a final master to a highly compressed format with little color depth?

Thanks Perrone - but if you read this post (Apple Forum) you'll see why I'm confused about this!
Apple - Support - Discussions - Output XDCAM EX to SD via What Card and ...

Please can you describe a workflow for saving files to the HD via Card - I find it very attractive but I've never used one, so it's all a mystery to me, I'm afraid.

I completely misunderstood. You want to play OUT to something, I thought you were trying to play IN. Shane has it absolutely correct. This is an absolutely backward way to get to an SD downconversion. And totally unnecesary. A hardware downconversion is not going to be any better than a GOOD software downconversion. In fact, I can just about guarantee that it won't be as good, since it has to be realtime or near realtime.

All you need is a rescaler that uses Lanczos scaling. I've posted workflows here before that outline how to do it. And the software is free on PCs. Not sure what's out there for the Mac, but there has to be something. Works as good or better than any hardware solution out there.

If you REALLY want to get a hardware conversion, then connect the component outs from the camera to the input jacks on the card, and record to hard drive. It will be a real time transfer, but you'll get your conversion in camera. The EX1 can output HD or SD over that component connection.

Perrone, you're right, had a knot in my thoughts. Too early in the morning;)

No!
(I should have linked to the whole thread - sorry).
I want to try a Hardware Downconversion and I was asking Shane whether I had to have a Deck as well as a Capture Card (it seemed strange that a Capture Card couldn't do the Downconversion and somehow shove it back onto the HD without having to output to Tape and then Capture it again).
Shane misunderstood me.

OK - maybe, but I'd like to try.

That's excellent news! You've explained a lot and I'm beginning to see the light now but I wanted to Downconvert my Timeline using Hardware Conversion - just to try for myself. I can't see how I could do that unless I could export the Timeline as MXF - any ideas?

I'll look for this - maybe another Software method to try!
_______________________

Once again Perrone, thanks for your help and your patience!

Okay, I'm really nervous about making this post. It's shows how bizarre my logic is sometimes. Please go easy on flaming me. :)

Full-screen is full-screen, right? In other words, full-screen equals 100% in scale. If you're playing a video and it's full-screen, but then you shrink it by 20% it gets smaller and is no longer full-screen. Same thing if you increase the scale by 20%, it will get bigger and no longer fit on the screen.

If 1080 30P video is 1920 x 1080, any video with that pixel dimension will be full-screen on an HD television. If 480 30i video is 720 x 480, and video with that pixel dimension will be full-screen on an SD television. Further more, if you shrink a 1080 30P video (1920 x 1080) down to 480 30i (720 x 480) it will still be full-screen when viewed on an SD television.

So.....(here's where my mind get's really weird)

Text that was outlined with a 2 pixel wide outline in 1080 30P will still have a 2 pixel wide outline when shrunk to 480 30i, because it's still the same width when viewed full-screen on an SD television. KEY POINT: But there will be less pixels to describe the outline of the on screen text. It may not look as "crisp" as before.

There's probably a much better way to communicate this, but is there any truth to my "logic"? What am I missing here?

Sorry, maybe I partied a little too much last night. :)

No Mitchell, you're doing fine. The problem is that the rescalers don't know that your 2 pixel wide outline is an outline and not noise. So you need a rescaler that is smart enough to examine the patterns in the image and do their best to preserve the patterns and not just shrink everything down. When a rescaler scales down indiscriminately it's fast and it usually uses a method called "nearest neighbor". And this is what turns everything to mush.

Smarter rescale algorithms are able to look at patterns in the video and try to figure out what's going on before they rebuild a new image. It takes a LOT longer to run these routines and calculate for each frame, so generally you don't see these in realtime hardware or software. But they produce clean results.

Below is a frame grab comparison from a video I am currently working on. You one is native 1080p, the other is from a lanczos rescale down to SD. Look at the detail in both, the shadow quality behind the text, the edges of the letters. This is what I am talking about.

That looks great! I'm going to do some testing today on my Mac (I'm at work now)

Now you understand why I am always so mystified when people talk about how their rescales and how they can't get good results. That rescale was done with a FREE tool on the PC. HD -> SD can be done well, if you have decent tools.

If you downscale your video but let the vector-graphics render at the final resolution, then the edges of the downscaled video may have a different look than the edges of the rendered graphics. But if you first render the graphics at HD, overlay them to the video and then scale the flattened result down, the edges of the graphics and video-content will look more similar, which leads to a more consitent look.
Doing the downconversion in FCP is faster but has worse quality than doing it in Compressor.
ClipBrowsers downconversion-quality isn't very good.
I guess you overrate the intelligence of those rescalers. They don't work with any kind of pattern-recognition. They basically consists just of a lowpass-filter and and an interpolation-algorithm like a polynomial one like bicubic or sinc.
That example has perfect quality. :)

I attached a downscaled version of Perrones 1080p-image to 720x405 with Compressor for comparison.

I've been doing some testing with actual footage.

I created 2 1080 30P movies. One rendered in XDCAM (20mb) and one rendered in ProRes HQ (120mb).

The problem I'm running into is that when I go from HD to SD it wants to make my progressive footage interlaced. That looks like crap when viewing as a QT movie. (too ugly to show you all) But for broadcast my video needs to be interlaced. I need to spend more time with this...

I did learn that it's important when viewing a QT movie for comparison, to make sure that High Quality is turned on. Everything looks bad if you forget to do this step.

I'm going to head to lunch and then do some more testing when I get back.

Dominik, very impressive. Does the same apply to motion?

Yeah, I think compressing a still image isn't as big a deal as actual video (motion). Especially when you're dealing with interlacing issues.

Here's the HD movie clip I'm starting with.
http://www.ssscc.org/ftp/hd-sd/Test-...-1080-30P).mov

Now I just need to get it to look good in SD (720x480 letterbox)

Here's what I came up with (crap). I couldn't seem to get Compressor to transcode to SD without interlacing the footage. I went to the Encoder>Video>Settings and chose Scan Mode>Progressive, but it doesn't seem to make much difference.

http://www.ssscc.org/ftp/hd-sd/Test-...Anamorphic.mov

This is exactly what I predicted. The video in the background looks okay, but the graphics (red circle logo) look like crap.

This is harder than I thought!

Your example has some strong reds. Are you sure that you don't confuse interlacing-artifacts with 4:2:0-artifacts?

That confirms my suspicion: there's no interlacing going on, but 4:1:1-chroma-subsampling-artifacts of DV.

Send me a file to try, and we can see. I need a QT file that is compressed with something I can read on a PC. Like PNG Lossless, or QT uncompressed. Just send 3-5 seconds with high motion. It can be interlaced, and I'll try my de-interlacer as well.

I've done this for footage off the EX1 for myself, but would be interesting to try someone else's footage.

-P

Hey! What's a PC guy doing on a Mac/FCP thread? hehehehehehe (just kidding)

LOL!

Trying to learn!

I have to work with Mac folks in this video editing world, so it behooves me (and other PC users) to learn as much as possible about BOTH systems.

I've created a PNG file and I'm uploading it now. It's going to take a while as it's 150 mb. (about 15 minutes)

http://www.ssscc.org/ftp/hd-sd/Test-(PNG).mov

EDIT: Okay it's uploaded.

Are you sure of the name? And permissions?

Sorry, there was a typo in the name (MOV not MPV) I fixed it.

Processing ...

Done. Images coming!

DVInfo is acting dumb. Check here.
http://vimeo.com/groups/8264/files

Dang! Very nice Perrone! Too bad you're not on a Mac, I'd love to duplicate your work flow. I haven't given up on Compressor yet though.

Anything that I could benefit from on a Mac? (since this IS a Mac/FCP thread)

There has GOT to be some software application out there that uses the underlying rescale algorithm that I am using. The Lanczos one. Its FEELY available for goodness sakes.

Are you running bootcamp? It might be worth it for you to run some of these tools that I have. Especially, the stuff in Virtualdub since it's all free. The denoising programs available rival that $1k stuff I see people raving about. And you've seen the rescaler. There are also sophisticated tools for framerate conversion, color space conversion, etc. And they are all free.

Mitchell, I will say it one more time, but if you don't listen to me I can't help you:
if you compress with a YUV-format like DV, then there's chroma-subsampling going on, so features with strong saturation-jumps + hue-jumps + luma-jumps will look pixelated or blurred. There's nothing you can do against it but using a non-YUV-format or decreasing saturation.
Try to use QTs uncompressed setting and you'll see the difference.

Yeah, it must be somewhere. Does anybody know which algorithm Cinema Craft is based upon, given its exceptionally oustanding results? With the new CC plugin for Compressor, that might be the road to go for high end encodings.

Would ProRes be a viable solution? And/or which other codecs?

Compressor is doing exactly the same than lanczos. I posted an example to show that, but who realizes it?
I guess we are talking about rescaling, not mpeg2-encoding. Or do you really want to know, how CCEs mpeg2-engine works?
ProRes uses 4:2:2, so it's not a "solution" for that problem. There's no solution, because all formats used by DVD, BluRay,... use chroma-subsampling. You have to live with that and to hope that end-consumer-playback-devices use proper chroma-filtering to transform pixelation to blurriness.

You need a codec which can handle RGB, and preferably 10-bit or higher. Prores is 10-bit, but yuv, not rgb. You'll note that I typically ask for QT PNG or Uncompressed. Both of these are RGB with alpha channel. I am still testing codecs, but right now am about to test 10 bit Avid.

I have so far not found any hint about Compresor and Lanczos. But thx very much, will turn my radar more towards Compressor, albeit I don't like its GUI.

Rescaling, which CCE MP, the plugin, does also.

Is there a table with the subsampling specs of all known codecs? Or would you say ProRes is kind of "the best compromise". QT Uncompressed leads to extremely large files...

Fraunhofer Institut suggests JPEG2000 for a comparable task. Any comments on that?

Yes sir, I am. Which apps are you talking about? Would love to test them on my Mac turned WinMachine;-)

Based on my results tonight, and over the past few months of testing, Avi's DNxHD is giving results darn near equal to uncompressed at a fraction of the size. ProRes isn't close since it cannot do RGB with Alpha channels (unless someone know's sommething I don't).

Looks like this is my new mastering format! Sweet!

And it comes in Mac and PC versions, and the codec is free. Can't beat that.

QT PNG > Win?
 

Is there a compatible codec to QT PNG in Winworld?
BTW, we are testing Avid Meridien uncompressed for file exchange between Macs and Wins (Liquid, Avid).

The program you want is Virtualdub. Its just a framework. You extend it by dropping "filters" into its folder. No install necessary. Just close the program and restart. The lanczos rescaler is built-in though. Just open your video, select the resize filter, type in the new size, choose lanczos, and convert.

It's pretty fast too.

I hadn't done a Jpeg2000 test in HD, so I tried it just now.

Basically, what I've tried are the following codecs

Uncompressed
BlackMagic 10 bit - Largest file besides uncompressed
Aja Kona 10 bit
Jpeg2000 (best quality) - Small files, performance not so great on timeline
HuffYUV - YUV so loses Chroma
Lagarith - Similar performance to HuffYUV
PNG Paeth - Nice encode but takes FOREVER to encode
Avid DNxHD 220 - Small files, good timeline performance.

I did not test DVCProHD because I wanted full raster
I did not test ProRes because I don't have a Mac.

For my money, the Avid Codec wins. Good performance, reasonably fast encode time, available on Mac and PC for FREE. I'm glad I have this solved as I have a huge project I am currently working on and needed something I could work with. But now I have to go back and remaster all the bit's I've done already!

But don't you have to edit in AVID to use it? How can you go from FCP to this codec?

Absolutely not. I edit in Vegas. If you install the quicktime Avid Codec, you should be able to render to it from FCP just like any quicktime file.