HDSDI OUT is 420
A couple of days ago at IBC I asked the Japanese tech guy responsible for the EX1/3
|
For playout yes (4:2:0 8-bit padded to 10-bit); for live feed, it's 4:2:2 10-bit.
This subject is months old. There is this thread; see post #7, #11 en #39. In that last post Matt Jeppsen from FreshDV recontacted Bob Ott from Sony to address this very issue. Aparently Sony keeps giving out the wrong info at events. George/ |
George,
Today the Sony rep I talked to (might have been the same guy, but with the number of reps there it seems unlikely) confirmed Emmanuel's post. HD-SDI has a spec of 4:2:2 but the signal from an EX3 is nothing more than 4:2:0 due to the circuitry used. This apparently applies to both live feeds and recorded feeds. He went as far as saying that HD-SDI on the EX3 makes no sense at all from a color space perspective. |
A couple of days ago at IBC I asked the Japanese tech guy responsible for the EX1/3 cameras about the LIVE FEED of the HDSDI OUT
He replied, that although the signal is formatted as 422 in reality the quality is 420, because after it is captured from the CMOS, is processed in the 420 colorspace. He then proceeded to explain that in such a small camera a higher bitrate or larger color information would result in overheating. After I was challenged by other members of the forum that I had misunderstood I went back today. I asked more than ten times the same question in various ways so it will be clear to him and the answer was always the same: The signal is formatted as 422 but the quality is 420. I asked other Sony people in the booth but every one of them directed me to him as the most knowledgeable person on the EX1/3 series camcorders. Then I went to the person responsible to the XDCAM line and asked the same question about the F355 (the F335 was noticeably absent) . He said the same thing referring for the F355, that although the signal is formatted as 422 the quality is 420. So I asked, if EX1/3 are a similar case and he reaffirmed it. I know neither of them spoke the best English, but I was so insisting that its not possible that I had misunderstood them. I guess further clarification from Sony is due. |
No verification needed. It's filling all 10 bits 4:2:2.
This had been veirifed many times here. When blowing up EX1 4:2:0 images up against the 4:2:2, you can clearly see there's a difference. A while back, someone even posted an EX1 4:2:0 and the same frame from SDI 4:2:2 here: http://www.dvinfo.net/conf/sony-xdca...omparison.html |
I've updated my post with the references to the earlier discussion.
George/ |
Quote:
Hm... Are you not confusing things a bit now? That comparison is of MPEG 35 Mbit/s versus uncompressed HD-SDI. Wheather the HD-SDI is 8 or 10 bits is not possible to judge from these images. |
It would be helpful for and Engineer from Convergent to provide some factual confirmation one way 4:2:2/10-bit or the other 4:2:0/10-bit. Convergent is expending considerable resources to develop a product that is intended to make full us of the signal out of EX1/3 SDI port.
For my part, it's hard to believe that Convergent would be advertising SDI 4:2:2/10-bit potential, charge a good buck for this capability and then deliver something less. |
I believe Cineform has looked at the HD SDI out and confirmed all 10 bits contain data (no padding) and it is 4:2:2 color data.
George/ |
Yes,
Sorry... I meant to write 10bit 4:2:2. This was confirmed by Cineform for the EX1. |
Again I can confirm my live HDSI feed was 10bit 4:2:2 as well. From the EX1.
|
Here's an FYI for anyone who want's to test this themselves. Record a scene to both SxS and to a HD-SDI capture card on your computer. Take the same frame from both recordings and zoom in to close to pixel level. 4:2:2 will always have two pixels side by side of the same base colour. 4:2:0 will always have four pixels in a square of the same base colour when in progressive mode.
|
HDSDI OUT is 420
Hi,
sorry for being so much a neophyte, but I have watched the above thread and Emmanual's quote was: <<the answer was always the same: The signal is formatted as 422 but the quality is 420>>, so all the words to go and look that HD-SDI is 4:2:2 do not address the quality, but the format, and that wasn't the claim. So my question is, when people say "you can see the quality difference" are you sure that what is being looked at is truly a difference at the output or is it a difference produced post output? Hope this makes sense to you all, because, as I say, I am on the edge of my understanding of the two process chains here. Cheers, Ian |
Sorry guys, CineForm hasn't tested this. Anyone with an EX1/3 in San Diego, we would be happy to do so. Or capture a live HDSDI feed to an uncompressed (or CineForm) AVI a red object on a blue background in 60i mode and send me the data. This will be straight forward to determine 4:2:2 vs 4:2:0 up sampled. As for 10-bit vs 8-bit, I expect the precision is 8-bit in either case.
|
Oops... Is Convergent Design the next major player now to come and admit "they haven't tested this"? Could it be that everybody just hoped on Mr. Martinez words, because they liked them?
|
Quote:
If you can patch an old VHS camera to an HD-SDI port, the Nanoflash would faithfully record it in 4:2:2/10-bit HD. But the picture wouldn't look any better than the composite out from the old VHS camera. It's the same as capturing a composite feed from a VHS recorder and dropping it on an uncompressed 4:2:2 HD timeline. It would be in a 4:2:2 format, but there would not be any more image data than what came over the original composite wire during capture. |
Thanks David,
Sorry, I belive the discussion a while back was not regarding 4:2:2, but 10bit. Elton Barlow mentioned he sent you an SDI captured Sheer file a while back. DVXuser.com -- The online community for filmmaking - View Single Post - Sony EX1 market plan: typical Sony |
Quote:
EX3 & nanoFlash spec sheet: http://www.convergent-design.com/dow...20Brochure.pdf EX1 & Flash XDR spec sheet: http://www.convergent-design.com/dow...20Brochure.pdf Convergent is using the HD-SDI digital port and if you've been following the DVinfo threads on the Flash and nanoFlash then you'd also be aware of the comments made by the Convergent Engineering/Marketing folks with regards to working with the EX1/3. |
Quote:
Quote:
Quote:
|
The thing I find the most amusing about all of this is that nobody can really seem to tell if what they have is 4:2:0 or 4:2:2 but yet they complain about how bad the 4:2:0 is. If it was really that bad then more of you would notice it.
|
Quote:
And if Convergent is not willing to confirm the facts for whatever reason, then from the posting above, Cineform has come forward to answer the question. If someone would kindly provide them an uncompressed file captured off the EX1/3 Hd-SDI port. Cheers! |
I agree to a certain level. If you can't really determine a difference by close analyses of the footage, the option is questionable.
But in other circumstances, for example color correcting, the 4:2:2 space will make a huge difference. So, it;'s not just a question of "looks". |
Quote:
|
Hopefully George, you can help clear this up once and for all.
I'm really hoping that it is at least 4:2:2. A true 10-bit (non-padded) stream would be peachy too. |
Quote:
One of our local users can also confirm (through detailed image analysis) that the color sampling is indeed 4:2:2. We can not determine if the original sampling is 4:2:0 or 4:2:2 on our waveform monitor, since all HD-SDI is 4:2:2, but the original signal could have been generated from a 4:2:0 source and then upsampled. Just to be clear, Flash XDR and nanoFlash uses the Sony MPEG2 CODEC for compression/decompression. MPEG2 is, by definition, only 8-bit processing, but can be either 4:2:0 or 4:2:2. Also the compression allows either 1440x1080 or the full 1920x1080. On our Flash products, the MPEG2 compression uses the full 1920x1080 4:2:2 processing at bit rates of 50/100 Mbps Long-GOP or 100/160 Mbps I-Frame only. So, although the processing is only 8-bit, we do not suffer from horizontal subsampling artifacts (scaling 1920 -> 1440) or loss of color fidelity (4:2:2 -> 4:2:0). We should be able to soon upload some of the gorgeous airplane footage we showed at IBC (in QT format). The footage included smoke, water and high-motion without any visible artifacts whatsoever. This 7th generation Sony MPEG2 CODEC is simply brilliant, especially at the 50/100 Mbps level. |
Thank you Mike, this is indeed the independent confirmation that goes a long way to clearing up doubts, misunderstandings and possible bad feelings.
I also appreciate your comment about the source signal that it may be up-sampled to generate 4:2:2 and that there is no way of knowing. Perhaps that is what the Sony engineer at the IBC was referring too. What is truly important is your independent confirmation that the signal is genuine 10-bit and 4:2:2 color space. |
Mike, thanks for clearing these questions up.
|
Here's the practical greenscreen test.
EX1 recorded to SxS card and out HD-SDI to Blackmagic Decklink Extreme simultaneously. (Exposure: image was underexposed at acquisition with intention to manipulate it in post.) SxS footage was later converted by HDlink from MP4 to Cineform Prospect HD High. HD-SDI capture was live via HDlink to Cineform Prospect HD High. (All software is PC based.) Then both videos were placed into After Effects project, and same frame was exported as Photoshop files. I then exported from Photoshop to TIFF for more universal compatibility. See files below (I hope they'll post here fine. If not, see my own web site below. Each image is over 8Mb. Right-click to download.) http://primehd.com/sxs-sdi They are named sdi and sxs accordingly, and contain areas of both motion blur and sharp still image. Let me know how do you interpret the results. My own humble interpretation is that there's no discernible difference in levels (including individual channels) or visible quality of the image. At 200x magnification, you can see very slight difference, with SxS image actually looking a tiny bit sharper. In any case, differences seem to be very close to noise level of the image. Sadly, I have to conclude that, for all practical purposes, HD-SDI out of EX1 is the same as SxS in terms of visual quality and color space, which is 4:2:0. There's a frame in my video (not posted here) with higher motion, where SDI footage looks a tiny bit better at 200x magnification, but that's it. |
Perhaps what we should be talking about is chroma resolution rather than sampling.
If the chroma is being upsampled then little to nothing is gained by recording that signal over what can be done in post for where it matters e.g. keying. |
Quote:
These 4:2:0 optical block sampling claims also came about when the 330/350 were introduced. I've been told many times since then that if taken live, it's 4:2:2 sampling. -gb- |
Quote:
If true, it also raises the question: is the 50/100Mbps Convergent Design is offering, still worth it with this particular camera, less compression being the only benefit over the regular SxS? |
Quote:
|
Quote:
To me though, the discovery of HD-SDI out being same quality as internal recording was a bit of a relief. This means I don't feel obligated to capture live from HD-SDI to ensure the highest quality of the image... SxS cards will do just the same. |
I'd rather that the cam had HDMI out.
It'd make it so much easier to monitor signal with HDMI-enabled cheap monitors/HDTVs. HDMI could carry 1080p signal just fine as well. HD-SDI connector on this cam is probably a bit of Pro level marketing pretense for this camera that is not matched 100% by the real quality of the signal outputted. |
Well Alex, I must say that I would NOT be relieved to find out that the HD-SDI out is pretty much the same quality as the SxS cards as I was wanting to use the Convergent-Design nanoFlash recorder to obtain a higher quality image. Especially for compositing work and color correction.
Pitor makes a good point. Am I willing to spend the extra $$$ to ONLY have less compression? Hopefully it IS 4:2:2 10-bit. |
Sorry but it seems you can't win on compression artifacts either, judging by my tests.
|
Why doesn't somebody call Sony professional support and ask them directly?
If they don't know or give you an incorrect answer, then who would know? |
alex, those stills are surprisingly similar. although there does not seem to be significant chroma artifacting in either... maybe you could do a test with fewer steps? original->cineform->psd->tiff probably means the images are going through 8-10bit, YUV-RGB and codec conversions before we see the final tiffs.
im going through some greenscreen footage i shot and am seeing a bit more in the way of stepping when i isolate the color channels than alex's tiffs (maybe because the image is going through fewer conversions and my image has higher saturation/exposure), but its not that strong, certainly not 2x2 chroma pixels like one might expect to see from 4:2:0, so maybe its getting smoothed out in the YUV-RGB conversion. im really only seeing the stepping in the red channel as well. then again things have to get converted to RGB for any kind of program to display, but does anyone have any better ideas how to isolate R and B color channels from YUV for more accurate resolution analysis? ive tried photoshop and after effects but both only show me the channels after RGB conversion. Maybe there is a way to get it to do the RGB conversion after isolating one of the YUV chroma channels? |
oh yeah and note that i only have sxs, i dont have HD-SDI capture capability at the moment. I was just doing this test because if anything, SxS should have chroma stepping and we should be seeing if SDI does as well, but alex's image show no stepping from images acquired either way so there is something wrong with the process. seems like the best way to verify would be to find a method where we can see and analyze the stepping in SxS mode and then repeat it in SDI mode to see if it still exists.
|
I'm interested in Cineform only, because it is my current workflow, and after years of use I'm very happy with it.
I don't think Cineform conversion would bring in enough artifacts to muddy up the results in significant way. Stills: look to me exactly as they did in AE timeline on-screen. I'm not big on video signal theory... What is *stepping* that you are referring to? |
All times are GMT -6. The time now is 07:34 PM. |
DV Info Net -- Real Names, Real People, Real Info!
1998-2024 The Digital Video Information Network