Incredible Artifacts disappear with Z1 Downconversion

[Ruslan Odintsov]

Originally Posted by Gino Do

Have you tried using CAPDVHS application? and then convert to the uncompressed format. Cineform is an intermediary Codec so it does some converting and manupilation of the video, on top of the original on camera compression. I think CAPDVHS just captures the original compression.

Gino, I don't own Cineform product, so I don't use it at all. I don't even use the intemediate crippled down Cineform built into the Premiere Pro. I can't even use it to read in most if the outside applications, such as older version of After Effects - it gives me a watermark. That's just wrong. Besides, Cineform seems to be playing with color space, because the re-exported M2T never looks the same color space-wise as the original, if I passed it through Cineform (I tried Aspect HD Demo 3.0). Never such color space problem when using MainConcept MPEG Pro plugin and then working with TARGA or some other lossless codecs as intermediate, and re-exporting back to M2T. Also, on a side note, MainConcept plugin is perfect for cut-editing: no recompression, no generation loss, pretty fast and no need for intermediate at all. Even laying a title or a similar 1st generation effect doesn't result in any visible loss of quality.

I always capture HDV with CapDVHS into M2T, and that's what I tried to downconvert in a few software solutions. So, direct MPEG2 to lossless AVI (or QT for that matter), still results in inferior downconversion of interlaced horizontal detail compared to Z1/FX1's downconversion.

Originally Posted by Kurth Bousman

one question amigos - I'm not at home or I'd try ( maybe next week ) but can you downconvert and record component to a dvcam deck like dsr20/40 ? or do you need a $2k video card ?

I'm sure if your DVCAM deck has component inputs you will be able to record downconverted component output from Z1/FX1. I you have such deck - I'm very jealous :)

[Douglas Spotted Eagle]

Ruslan,
I'm VERY interested in what you're seeing, because your comments in various threads are completely inconsistent with what the majority of users are seeing, and completely opposite any post I've ever read. (and I moderate several HDV fora)

Cineform is exceptionally color accurate. They were the FIRST ITU 709 accurate application in the "lower" end of the NLE world, and have only improved on that since day one. Premiere is only ITU 709 accurate when they are using the Cineform-authored avi files, transitions, and plugs.
That said, my own experiences with the Main Concept codec are less than stellar, and while that's my opinion, there are dozens of posts from users that express the same thing.
Are you by chance using a scope? If so, I'd very much appreciate a post of m2t vs avi as seen by a scope. When Sony Vegas 5 was the only option, I was testing a lot of images with scopes, and with the help of the Cineform folks, we released a preset for the Vegas color correction tools that offered color resampling accuracy to maintain the 709 colorspace. So, I've got a fair idea of what they're doing. And in Premiere, there is no substantive difference as relates to colorspace.
Since you demanded images from those of us that disagreed with you in other posts, I can only submit it's fair that we see this from you. We've all posted various images at some point, please do the same so we can get an idea of what you are truly seeing. With your consistent deriding the software conversions, I'd like to get a grip on why you have this opinion that is diametrically opposed to every experience I've had, and that of every post I've ever seen on the subject.

Kurth, you can easily take component out of the Z1 and go to an HDCAM or DVCam deck, and get a great image. As discussed deeply in other threads, you'll get the best image overall this way if you are staying HD, but you'll likely find your NLE can do a slightly better conversion in software, vs hardware.

[Steven White]

it's fallacy to think that a dedicated hardware solution is not good enough

My statment wasn't an opinion - it's math, and hence my use of the word "fallacy". "thinking" and "good enough" have nothing to do with mathematical truths.

Hardware is nothing more than algorithms implemented efficiently on a dedicated chip. The Sony hardware solution may well be the best solution currently available. However, if the algorithm was published, it would be possible to execute in software, and hence avoid noise introduced by digital-to-analog-to-digital conversions... and while this noise could in principle be negligible, or even enhance the perception of the image, it is degredation.

There are a lot of problems to be addressed in a HD to SD conversion:
- resolution change (1440x1080i to 720x480i)
- colour sampling change (4:2:0 to 4:4:4/4:2:2/4:1:1/4:2:0 all of which are possible)
- colour space change: YUV.709 to YUV.601
- upper fields to lower fields
- MPEG-2 artifacting

The path between HDV and SD of any flavour isn't obvious, and is completely subject to interpretation. When do you manipulate which part of the image? When do you de-artifact? Do you de-artifact? How do you convert the YUV spaces? How do you edit afterwards? Most people don't even want to think about this stuff - so most of it is hidden from view in "black boxes".

But like DSE just said above - the Cineform folk have been working on this problem almost exclusively. It will be interesting to see what falls out of it all.

-Steve

[Ruslan Odintsov]

Originally Posted by Douglas Spotted Eagle

Ruslan,
I'm VERY interested in what you're seeing, because your comments in various threads are completely inconsistent with what the majority of users are seeing, and completely opposite any post I've ever read. (and I moderate several HDV fora)

Douglas, before we go any further, did you compare those Canopus 576i clips on a quality PAL monitor? Please let me know if you've noticed any difference in field rendition. Somehow, nobody who defends the software downconversion (majority of people) have tried to answer to my challenge in that area. And it's pretty obvious. Just for comparison, everybody loves progressive scan these days, for no other reason, but the lack of annoying flickering lines that interlaced scan introduces. Getting back to interlaced, so far, every software solution seems to accentuate those annoying lines, while Z1's downconversion looks much more natural, without loosing sharpness (at least in the component output version). I mean, the software downconversion is almost as bad as what cheap $100-$300 PAL<->NTSC standalone converters do on horizontal lines. I was genuinely relieved when I got Z1 and tried downconverting, which looked so much better. I was doing many PAL<->NTSC conversions for awhile on a few software/hardware solutions, and believe me, I got a "trained" eye for those scaling artifacts. That's why the software downconversion from HDV to SD bothers me so much - I see those artifacts. It's like when you for the first time in your life notice MPEG blocky artifacts, you'll be seeing them and be annoyed by them forever since that moment. Same thing for me with those lines... Please, look at those Canopus clips and try to look for what I see in there.

As for colorspace tests with CF codec, I'll have to redo them when I get time, because I have deleted the original test files. But I remember that neither ProCoder nor Premiere Pro was able to render them into other formats and/or back to M2T with proper color space. Perhaps Cineform fixed those issues with version 3.1, I don't know. But as I said, I use MainConcept MPEG Pro for simple editing and output back to camera. Or render as Huffyuv or QT TARGA if I have to do some complex layering in After Effects, etc. That is truly lossless intermediate solution, unlike Cineform, which is "visually perfect", but in reality is lossy.

Another thing is, yes MainConcept MPEG Pro is not real-time, but it's lossless for simple editing, unlike Cineform. For that task it's also quicker than Cineform, because there's no need to render back and forth between M2T from CF. People should really give MainConcept 1.0.6 a try (they have a demo), as they fixed many bugs, also included the corrected AR for Premiere. Besides, I've never had any corrupted output with MainConcept, unlike what you see happening with Cineform, where many people report various blocky artifacts.

I'm not biased towards any particular solution, I simply report what I have observed, opening the can of worms (read: truth) on popular tools which unfortunately don't deliver the goods the way they should or promise. Unfortunately, many people believe what they'd like to believe, perhaps because of spending many $$$'s on software solutions, without willing to look at those solutions objectively. That of course would mean that they wasted their money on something inferior. I understand that, but it doesn't make the real facts any different.


P.S.: Again, with all the arguments against my observations, nobody still gave their opinion on Canopus clips from BBC UK thread (page 5). Please do that and try to be objective. If nobody's willing to explore that horizontal lines issue, why should I believe anyone who defends the software downconversion if they are so biased that they won't even objectively compare the results from all possible angles?

[Douglas Spotted Eagle]

I don't have the canopus clips, I didn't shoot them, haven't looked at them, and aren't really interested in them because I didn't have control over them. I do know that I've experienced, I do know what others have experienced, and do know that it makes sense that good resampling and scaling can do better than fixed hardware, albeit MUCH more slowly.
I'm also on travel where I don't have access to large files, so can't really look at much.
I do have a great deal of experience with the Canopus codec, and with their software, I developed and produced their HDV solutions training DVD which is just now starting to ship. And Sony Vegas, and the High setting in Cineform, are superior, IMO when it comes to resampling. I've resampled the same footage on all systems we've both commented on.
Again, please post YOUR images so that folks can perhaps get a glimpse of what it is you're seeing/describing.

[Thomas Smet]

Lets end this debate.

Somebody with a FX1/Z1 shoot something with a lot of different overlapping colors.

Capture a segment of the video down converted from the camera to component.

Next capture that same segment as a HDV stream and then down convert in any software or multiple software applications.

Post images to compare the results. (try to get the same frame)

Clearly the results could be different depending on what program you use to down convert but it would give all of us an idea of how good the hardware method is.



One other note.

Even using component will only give you 4:2:2. Software can give very close to 4:4:4. I say close because of the "0" in 4:2:0. Even though there are no jagged edges the chroma channels get interpolated in a different way than the luma since for filtering scaling has to look at the pixels next to the current pixel. This means the chroma doesn't match up pixel for pixel with the luma. Now for most things this will not matter. For the pixel re-mapping program I am writing I like to call it 4:4:2. Horizontal chroma is a perfect match but vertical has a small glitch every other line. The best solution for SD would be to capture 4:2:2 HD uncompressed and then down convert to a perfect 4:4:4 SD. That is a huge waste of time and money for SD in my opinion however.

[Steven White]

Or render as Huffyuv or QT TARGA if I have to do some complex layering in After Effects, etc. That is truly lossless intermediate solution, unlike Cineform, which is "visually perfect", but in reality is lossy.

It was my understanding that Huffyuv doesn't handle the .709 colourspace properly. Keep in mind that colour-space conversions are inherently lossy.

While I will not claim that Cineform is lossless in any sense of the word, it is a higher-bitrate and a less-compressed codec than HDV. I would be interested in seeing a demonstration of the following done in the native YUV.709 colourspace:

HDV vs. (Cineform compressed from HDV source)

and

HDV vs. (Cineform compressed from HDV source, re-compressed to HDV for output)

Ultimately I expect we're splitting hairs on colourspace and compression issues though, and that this aspect of the discussion is quite irrelevant.

I'm still interested in the field rendering from the camera - in particular the blurring and sharpening that I already mentioned from my tests with the posted *.png files. Has any one else done some experimenting with them? Does anyone have a software deinterlacing tool that doesn't result in said artifacts?

-Steve

[Kurth Bousman]

hey guys - are we splitting pixels - or what ? wow- this is just the reason many people are scared of hdv. All ( imho ) I can say about my z1 , is it's a far better dvcam camera than my pd100a. And I'll keep on shooting hdv and downconverting in the camera because ( maybe it is and maybe it isn't ) the best way but it's definitely good enough for me at the moment , and it's convenient and free . When my fcstudio comes in next week maybe I'll change , and maybe I won't , but whichever route i personally chose to get my images to monitor will pale in comparison in importance to the images and subjects I'm able to capture in the first place . When you experts ( and I'm being serious not derisive - 'cause I'm here to learn ) get this worked out and have some workflow worked out that is modest in cost and easy on the learning curve , I'll be waiting to give it a try . In the meantime , I gotta shoot some images.
One last thought - my experience thus far has been - get good exposures and don't move the camera too much and people will be very impressed. I think the camera steadiness issue , in terms of image breakdown , is a far more important issue . thanks Kurth

[Ruslan Odintsov]

Originally Posted by Steven White

I would be interested in seeing a demonstration of the following done in the native YUV.709 colourspace:

HDV vs. (Cineform compressed from HDV source, re-compressed to HDV for output)

This is what I had problem with: HDV->CF->HDV.
On the other hand, when I did HDV->Huffyuv/QT TARGA->HDV, I never had a single color space issue (in Premiere, because Vegas does weird things when you render to foreign codecs, so you have to use a filter to expand colorspace back to normal).

Originally Posted by Kurth Bousman

One last thought - my experience thus far has been - get good exposures and don't move the camera too much and people will be very impressed. I think the camera steadiness issue , in terms of image breakdown , is a far more important issue.

Kurth, I was shooting a low-movement theatrical play last week, naturally using a good tripod. I didn't even have to move the camera to see the fields related artifacts with software downconversion. Whenever an actor moves even slightly, you see those lines floating in their hair, on their forehead wrinkles, on their shirt details. So, it still doesn't work well, unless you shoot completely still pictures. But the in-camera downconversion looks very smooth.

I might post some stills, but when it's really noticeable is when you view the real video clip on a TV monitor. I notice it on both interlaced CRT's and progressive plasma.

Douglas, here are the direct links to Canopus files:

Sony Z1's downconversion version (34.6MB):
http://www.canopushdv.com/conversion.../sony_tram.m2v

Edius' downconversion version (34.6MB):
http://www.canopushdv.com/conversion...edius_tram.m2v

Yes I can try and find a web space and post my own files, but the above links provide a perfect example of what I get as well, so my results wouldn't be any different, except for a different footage.

[Douglas Spotted Eagle]

Originally Posted by Ruslan Odintsov

This is what I had problem with: HDV->CF->HDV.
On the other hand, when I did HDV->Huffyuv/QT TARGA->HDV, I never had a single color space issue (in Premiere, because Vegas does weird things when you render to foreign codecs, so you have to use a filter to expand colorspace back to normal).

Now your posts are making even LESS sense. Premiere, unless you are using the Cineform codec does NOT maintain correct colorspace, and Vegas 6 DOES maintain correct colorspace. Premiere still does not maintain correct colorspace, and likely won't until the next rev, because they rely on the HAL layer in Premier using the Cineform tools. So, I don't see how you could accurately see what you are suggesting above. Further, Huffy is ALSO NOT 709 compatible, so again, your post makes absolutely zero sense.

While I enjoy good discussion, something here is either exceptionally fishy, or you are simply not understanding the differences of colorspace.

[Ruslan Odintsov]

Originally Posted by Douglas Spotted Eagle

Now your posts are making even LESS sense. Premiere, unless you are using the Cineform codec does NOT maintain correct colorspace, and Vegas 6 DOES maintain correct colorspace. Premiere still does not maintain correct colorspace, and likely won't until the next rev, because they rely on the HAL layer in Premier using the Cineform tools. So, I don't see how you could accurately see what you are suggesting above. Further, Huffy is ALSO NOT 709 compatible, so again, your post makes absolutely zero sense.

While I enjoy good discussion, something here is either exceptionally fishy, or you are simply not understanding the differences of colorspace.

I don't know and honestly don't care whether Huffyuv is compatible or not with 709 colorspace, but the fact of life is:

If I use MainConcept MPEG Pro project in Premiere 1.5.1 to work on M2T files, then take a small portion of the timeline, "Make Movie..." into Huffyuv, put this Huffyuv AVI file into After Effects, add whatever fancy effects I want to add, then paste the final Huffyuv AVI back into MainConcept MPEG Pro project in Premiere in between pre- and post- M2T fragments, render it all as M2T, send it back to camera: I don't even see where the pasting has occured. In other words the color space is preserved perfectly.

And talking about Vegas, simply try exporting even a regular DV file into AVI uncompressed and see how Vegas screwes up the color space. Then put that screwed up uncompressed AVI onto timeline and re-render it again, and it will get even worse (colorspace shrinking). If you render it into Sony codec or Vegas' built in DV codec, there will be no problem. Any other codec or uncompressed gets mistreatment. Of course there are filters to correct this problem, but why is it happening in the first place?

There's nothing fishy and I do have a basic understanding of colorspace, because I quite often have to convert between formats and need to adjust the colorspace when some program messes it up.

It just seems that nobody wants to give my discoveries a fair hearing. Perhaps your theories should work as you explain, but the practice is the only real eye opener. I have my eyes open and I don't like what I see. I'm not trying to insult anybody, but many of you have decided to keep your eyes closed because the professionals told you that some particular way is the only way it can work. You feel comfortable with what you're getting because you got reassurance from an expert. Sorry, but sometimes you just have to think outside of the box.

P.S.: Speaking of the box. Nobody still gave their opinions on the Canopus files. Amazing!

[Douglas Spotted Eagle]

Ruslan, it's not a *theory.* Either the application supports 709 or it doesn't. Premiere doesn't. Huffy doesn't. Vdub doesn't. You use the term DV Codec from Vegas, that alone tells me you don't understand, because DV doesn't support 709 either.
How about reading a scope? Do you know how? Maybe there is something being lost in a translation here. Ain't NO WAY, NO HOW you are using DV, Huffy, Premiere (without CineForm) and keeping the 709 colorspace. Period.
So again, either there is serious confusion, and maybe it's on my part.
Either way, until you post images that demonstrate exactly what YOU are seeing, I'm done with this confusion. There is just too much wrong with what you're suggesting, and you're apparently unwilling to show specifically what you're doing, or seeing, but yet willing to malign applications that at this point are the only applications that support correct colorspaces, and you're making comments that fly in the face of what many websites have posted images and comparisons.
That alone seems pretty suspect.

[Richard Entwistle]

Originally Posted by Peter Rixner

the steadishot was off, when shooting that pictures.
But I really don't see much problems when it is on. Could be even an advantage if your tripod's paning isn't the best.

When I mount the Z1 on Steadicam I use steadishot as well.
Never had a Problem with it.

Thanks Peter.

Sounds like my PD150 learning and experiences are transferrable! Which is why I am happy to upgrade, even if HDV is only transitory. The Z1 will pay for itself, so no need to get too hung up on potential shortcomings.

Richard

[Ruslan Odintsov]

Originally Posted by Douglas Spotted Eagle

Ruslan, it's not a *theory.* Either the application supports 709 or it doesn't. Premiere doesn't. Huffy doesn't. Vdub doesn't. You use the term DV Codec from Vegas, that alone tells me you don't understand, because DV doesn't support 709 either.
How about reading a scope? Do you know how? Maybe there is something being lost in a translation here. Ain't NO WAY, NO HOW you are using DV, Huffy, Premiere (without CineForm) and keeping the 709 colorspace. Period.

Douglas, I used DV codec term in absolutely no relation to HDV. I was just sayig that Vegas changes things when saving to foreign codecs. As an example, I was once trying to convert DV PAL to DV NTSC using MainConcept DV codec inside Vegas. The 16-235 colorspace of the original file was shrinked even further and the file was very low contrast with lighter dark spots and darker light spots. Same thing happened when I tried converting to 24p (well 23.97) from DV NTSC file, using your exact description on how to do it in Vegas 6, except for either saving to MainConcept DV, Huffyuv or uncompressed. Uncompressed AVI export colorspace shrinkage is what surprised me the most. As I said before, the only times that doesn't happen is when you use codecs bundled with Vegas, otherwise I have to apply filter to stretch colorspace back to 16-235. This kind of thing never happened to me in any version of Premiere.

But back to HDV. OK I was able to squeeze about 5MB om my free web space so here it is. These are six JPEG 100% quality ImageReady re-saved files, all rendered in Premiere as BMP, using MainConcept HDV 1080i 29.97fps preset. I'm sorry, but I don't have any space to put up BMP's or PNG's, though these JPEG's at 100% are not visually any different. If you want BMP or PNG I can e-mail it to you. So:

1) 1-Flowers-HDV-HDV.jpg:
M2T re-rendered as M2T using MainConcept plugin, then 2nd generation M2t placed back on timeline and saved as BMP. This is the exact same quality as the original M2T, because MainConcept used "smart rendering", where no recompression was applied. If you'd like to see the first generation file that's fine, but it's exactly the same in quality as the 2nd.

2) 2-Flowers-HDV-Huffy.jpg:
M2T re-rendered as Huffyuv AVI, then Huffyuv AVI placed back on timeline and saved as BMP. This second generation Huffyuv loos exactly like M2T grab, grain by grain, color by color.

3) 3-Flowers-HDV-CF.jpg:
M2T re-rendered as Premiere's built-in Cineform AVI, then Cineform AVI placed back on timeline and saved as BMP. This second generation Cineform made into BMP looks exactly like M2T colorspace-wise, though the grain/noise is changed because Cineform uses lossy compression.

4) 4-Flowers-HDV-HDV-HDV.jpg:
M2T re-rendered as M2T using MainConcept plugin, then 2nd generation M2t placed back on timeline and again re-rendered as M2T. This is the exact same quality as the original M2T, because MainConcept used "smart rendering", where no recompression was applied. If you'd like to see the first generation file that's fine, but it's exactly the same in quality as the 3rd.

5) 5-Flowers-HDV-Huffy-HDV.jpg:
M2T re-rendered as Huffyuv AVI, then Huffyuv AVI placed back on timeline and again re-rendered as M2T. The third generation looks the same colorspace-wise, but has some slight changes in noise/grain due to recompression into MPEG2.

6) 6-Flowers-HDV-CF-HDV.jpg:
M2T re-rendered as Premiere's built-in Cineform AVI, then Cineform AVI placed back on timeline and again re-rendered as M2T. The third generation Cineform-made-into-M2T has colorspace noticeably changed, plus added even more compression related grain/noise changes due to MPEG2 recompression.

Therefore, where Cineform suffers is when there is a conversion back to HDV MPEG2. Other codecs don't have this problem. See it for yourself:

http://www.geocities.com/odintsov/HDVTests.htm

If the above link doesn't work, then wait for about an hour and try again. I have bandwith restrictions with a free web space. I will leave this stuff there only for a few days, so grab it now if you want to see it.

The defects best show on an LCD computer monitor, but you can even see them on CRT. For the best comparison, simply insert these files into the image editing software, put one layer on top of the other and keep enabling/disabling them to see the difference.

P.S.: Same Cineform colorspace issues during export I was also getting in Canopus ProCoder.

P.P.S.: Same Cineform colorspace issues I was getting even using Cineform's native preset in Premiere Pro.

P.P.P.S.: These images are still not what I was originally talking about. This is just about pure 1440x1080. My challenge about the fields downconversion in Z1 vs. software still stands open. And still there's a deep silence from everybody concerning those Canopus clips which prove the inferiority of in-software downconversion of interlaced material...

[Douglas Spotted Eagle]

I have now tried 4 times to download the images, only to find that only the first two open, the rest don't. I'm on a slow dialup in the Pacific at the moment, so as said before, I'm done with this. Too many things simply don't add up. Maybe someone else is willing to track down why you are diametrically opposed to what has been everyone else' experience, from the Sony team to every post I've ever seen.
You are absolutely incorrect in saying Vegas changes colorspaces when rendering to third party codecs. If the codec can't support 709, how can you blame Vegas? Premiere, Ulead, Avid, Pinnacle, Canopus....they'll all behave exactly the same way.
I've spent hours and hours looking at scopes with a wide variety of source and rendered media, and am very comfortable with what Vegas does and doesn't do.
On the other hand, you bring up Huffy and other codecs that *don't* support 709 and claim they are exactly the same, when they cannot be, so again, something is very, very fishy here.