Noise comparison: 35/4:2:0 vs. 180/4:2:2
 

Here is another - far from scientific - comparison; this time I wanted to compare the noise in the EX1 picture (35 Mbps, 4:2:0, to the left) with that generated by the nanoFlash at 180 Mbps Long-GoP, 4:2:2 (right side picture).

Please take a look at the two screen grabs below; the left one is from the EX1. Please, compare them at a full HD screen of at least 50", or enlarge on your computer monitor (e.g. so that you can just see the wooden barn area directly around the square window).

As can be seen, the nanoFlash image is much, much more noisy!

Now, I have deliberately chosen a back-lit scene like this, and a picture profile not designed to minimize the chroma noise, so that the results are more readily visible. But frankly, I never suspected the difference would be so great, and not in favour of the nanoFlash...

Somebody more knowledgeable, please explain:

1. Is my nanoFlash faulty (I don't think so)?
2. Is the EX1 using some noise-reduction, which can not be turned off and is active in all PP settings - regardless of detail, matrix or gamma settings?
3. Or, is the noise visible in the nanoFlash recording a side-effect of the deeper color resolution the nano is using?

PS After a second thought, I guess the answer to my dilemma is simple:

The more noise "resolution" (read: visibility) is just a price for the more color resolution of the nanoFlash. It almost looks like the macroblocking at 35 Mbps is "masking" the grain, visible as excessive noise in the 180 Mbps, nanoFlash picture...

Any other ideas?

Piotr,
The pictures are JPEG.
Not valid for a test.
Use PNG, Targa or TIFF to make any comparation.

I've made many-many tests comparing the Nano vs the SxS files, and no way the SxS beats the Nano.
Rafael

Anybody?

I must say I'm shocked with the above results. The EX1 camera - IMHO - has only one flaw (relatively speaking), and that's the noise when lighting is low. When buying the nanoFlash, I was secretly hoping this would be improved, as well. Since the very first use of my nano I already knew it is not (it cannot be) - but why is it made so much worse?!!

Dan, Mike and Tommy - could you comment, please? Is my observation consistent with your results of the many nanoFlash testing you preformed? If so, is my "theory" above true? I mean, this would best be described by the saying "Sh#t in, [even more] sh#t out"...

If not, could anything be wrong with my camera SDI output, the SDI cable, or... the nanoFlash?

Please comment!

I trust you meant the saying "Garbage in, garbage out," usually just shortened to GIGO.

Adam,

In the CAD/CAM/CAE industry, we use the SISO version :)

Dear Dan,

I realize this comparison might be somewhat embarrassing to CD (unless a pretty obvious explanation exists) - but please address it, nevertheless.

In case one cannot see the striking difference in grain when watching the whole 1920x1080 frames, I placed below small fragments of the grabs from the original post (EX1 - left, nano - right hand side).

I can see the difference now that you made the closeup. I am not CD, so this is just my opinion, but it seems to me the difference is between 4:2:0 and 4:2:2 and between the low and high bit rate. That is, as you said yourself, the EX3 does not perform that well under low light conditions, so all that noise comes from the EX3 to the nF and the nF gives you what it gets as it gets it.

But the EX3 does not show it all to you because it throws some detail away in 4:2:0 and a lower bit rate, which just happens to smooth out this particular image.

It is also possible that Sony knows about the low light problem (I mean, they have to know), so they may be compensating for it, cleaning it up in the firmware, but they still send out the uncompensated data to the nF, which, as a general device, has no way of knowing about the problem, so it just gives you what it has received.

So, yes, it's GIGO. But the EX3 may be giving you specific GICO (garbage in, compost out) because they compensate for the garbage.

Thanks Adam - you more or less confirmed all my "theories" and possible explanations. Including the one about some permanent noise reduction going on in the EX cameras, independent of PP settings and NOT applied to the SDI output - only used directly prior to the camera's own compression by its internal, 35 Mbps encoder (see http://www.dvinfo.net/forum/sony-xdc...ctive-ex1.html)...

BTW, a similar feature was overdone in the otherwise great little HDV camera from Sony - the V1 - where instead of noise/grain, excessive macroblocking and "oil pain effect" was visible, even without magnification.

Well, I guess every EX1/3 - nanoFlash combo user must simply keep in mind the "augmenting GiGo", or GiCo, rule :)

Piotr,

What were the settings on your EX1? Gain, PP detail, gamma?

Garrett

Garrett,

I didn't write about PP settings because I just wanted to point out that the nanofile can be actually more noisy at 180 Mbps and with 4:2:2 color, than the EX1 file encoded at 35 Mbps, 4:2:0.

But since you asked, here are the main settings:

- gain: -3 dB
- matrix: HiSat
- detail: on, at 0 (frequency, white and black limits at some 50'ties)
- gamma: Cine3

Actually, when I think more about it - and assuming there IS some noise-reduction circuitry on the EX1/3 which doesn't influence the SDI stream - it's quite possible that some PP settings exist that would optimize the signal for the nanoFlash to compress at its high datarates and color resolution, and that those settings might differ from those, optimized for the EX cameras' own encoder.

But with so many variables, starting another "PP settings for the nanoFlash" thread on the EX1 forum is beyond my imagination. Certainly not possible without a little help from out CD Friends (or at least recognizing the problem by them)...

If you look at the detial in the tree bark and the roof you'll see that the nanoFlash did a better job. I think you are right in your assumption that the nano records everything w/ more detail, including noise.

I'd rather have the extra detail; noise you can always try to get rid of in post.

P.S. Can you post uncompressed grabs? The jpg compression kind of throws another variable into the equation.

I would tend to believe in the theory that the NF is just recording more information. The noise that is being produced by the sensor is just not retained in the 35Mbps 4:2:0 capture. It is lost due to the compression. I do know that some of the Gamma settings produce more noise than others. I wonder if you'd get the same results if you set your EX1 back to factory default settings?

Garrett

Piotr,
You can not compare pictures using JPEG.
You must use PNG, Targa, TIFF or so.
rafael

Piotr,

On the EX1 there's a switch in Picture Profiles that does noise reduction. Are you sure that switch is not turned on? That would account for the relative lack of noise in the EX image.

David

Sorry David, but there is no "noise reduction" switch on the EX1 that I'm aware of.

Also, the issue in this thread is not so much the noise on the EX1, but its increase in high bitrate nanoFlash files.

Picture Profiles / Details / Crispening

Adjusts noise suppression levels.

If you have it on the processed EX image could appear less noisy than the unprocessed nF image.

Oh,, crispening is what you mean.

Yes, I do use it sometimes (dialed into negative or positive territories, depending on the quality of scene lighting and hence the anticipated level of noise vs. the high detail definition desired), but it was at zero in this particular test shoot.

But you must keep in mind, as well, that BOTH the EX1 original AND the nanoFlash recording ARE processed (the PP settings act on the SDI stream, as well). So even if the crispening was at some strange value, it would affect BOTH.

All I can offer is that from my experience the EX1 hooked up to a 17" Panasonic monitor via HD SDI looks noiser than what it records. One question, how does the video recorded by the EX1 compare to that recorded by the NF at the same bitrate.

Hi Bob,

Thanks for chiming into another of my measurebator threads. Do you recognize the picture from our V1E's "oil paint" discussion back in 2007? :)

Your above statement is very valuable as it would suggest there IS some moderate noise reduction going on the EX cameras just prior do the on-board compressing/encoding (as opposed to acting on the pre-compressed stream, that is subject to DSP according to the PP settings in power, and output from SDI)!

Will have to check in similar low-light situ, but if you take a look at the red flower close-up in another of my measurebating threads on this forum, where I was comparing color resolution of 35 Mbps 4:2:0 with that of the nanoFlash'es 50 Mbps 4:2:2, there is no such apparent noise increase in the nano picture.

"Your above statement is very valuable as it would suggest there IS some moderate noise reduction going on the EX cameras just prior do the on-board compressing/encoding (as opposed to acting on the pre-compressed stream, that is subject to DSP according to the PP settings in power, and output from SDI)!"

Or like you said earlier, the EX codec itself is providing some form of "noise reduction." I would like to hear CD take on all this. Hope they respond.

Personally I don´t think the HD-SDI output is noisier, I have been using my EX-1 to capture uncompressed 8 bit 4:2:2 video with a BlackMagic Decklink HD Extreme 2, under Windows and using Adobe Production suite and the results are Great, I don´t see a noiser Image than the one in the cards. even with color corection applied the final noise level is far better.

now my nanoflash arrives next month so I can not compare the difference with my actual wokflow, but whatever is generating the noisier images we all see I don´t think is the HD-SDI output.

greetings

Thanks Julio for this input.

Yes - definitely looking forward to some exhaustive explanation from Convergent Design!

Piotr,

Did you try emailing CD directly? They seem to be pretty responsive.

I'd be interested in their insight too.

Garrett

Yes Garrett - I did today morning, after having waited in vain for some response from Dan, who used to be so active in this forum.

I e-mailed both Dan and Tommy, but still no answer...

I honestly don't see any issue here. The nanoFlash is recording a much more detailed image, that includes the noise information. The lower bitrate of the on-board Sony codec is smoothing those noise artifacts right out of the image.

I think these two images are showing me exactly what I would expect to see.

This has been one of my *possible* explanations, as well - but a word from CD would be appropriate, none the less.

Also, Alister Chapman (whose knowledge can't be overestimated, and who is also using the EX/nanoFlash combo) stated on the EX forum he never noticed such increase in noise, so the issue is far from closed.

Last but not least: the increased noise at the highest nanoFlash encoding bitrate thwarts all advantages of "much more detailed image" - as you put it - by far. On my 50" plasma, the noise/grain in nanoFlash recordings is much more striking than any potential or factual benefits of using it (like the better color resolution, or lesser mosquito noise)...

Disclaimer:
On the other hand, I seem to be the only one complaining about it - so a possibility exists something might be wrong with my SDI connection or nano unit. Which calls for some comments by CD even more!

Several things concern me conceptually about the methods in the test.

The screen grabs are jpg compressed, from different streams and codecs. One is sourced from 35 mbps compressed the other sourced from uncompressed and then compressed to 180 mbps by the nanoflash, but the jpg file size of the displayed grabs are roughly equivalent in size, meaning the latter jpg is about 4-5x more compressed for display.

In other words, the extra chroma samples responsible in part for the 180 mbps bit rate, what's happened to them in the conversion to roughly equivalent jpg file sizes?

The other thing is the second set of grabs that are 1920x1080, but have been magnified. When I looked at the full frame images (the first grabs) at 100%, I could hardly see what Piotr was talking about, but now I do, that's why he asked to view them at 1:1 on at least a 50 inch monitor. It is hard for me to see the noise at 1:1 on a 17 inch monitor.

I think what we're missing here as well, is the intention of nanoflash is the improvement in noise reduction in motion, yet we're looking at static images that arguably can be played at much lower bitrates 8-10 mbps on an HDTV monitor with people having a hard time to distinguish the difference from even 35 mbps unless there is motion.

What I am absolutely certain about, is I've never tried the 180 mbps setting, just the 100 mbps is what I use (looks good to me), and I also don't know if Piotr's native stream was interlaced or progressive..

Thanks for posting the examples, Piotr and Luben. Appreciated. I posted a similar comparison with elephant seals at sunset and demonstrated an increase in noise recording in low-light as well.

So, what's the complaint? The Nanoflash is obviously recording more detail, including noise detail. The device is doing its job. If you're worried about noise, then software noise reduction will allow for a more selective.

I'm currently experimenting with shooting with shooting the EX1 at a higher gain level of +6dB to allow my lens to close to f/4 instead of the usual f/2.8 for my regular indoor sports shoot. This increases scene detail due to increased DOF, and I believe puts the lens at an aperture which delivers greater center sharpness. (Someone correct me if f/4 is not actually sharper than f/2.8) The higher bitrates of the Nanoflash would relax the stress on the video compressor. I'm shooting at 140 MBps LongGOP and will post images soon.

Agreed. Piotr, aren't you able to grab frames as uncompressed PNG?

EDIT: Apologies, Piotr. I see that you posted uncompressed PNG of the red flowers in another thread.

Hi Tom,

Thanks for the sound comments (as usually). This is of course a huge topic for discussion, and it contains immense number of variables - but let me just address a couple of your points (sorry for not being systematic about it - haven't got time right now). And, I did say this test isn't even pretending to be scientific :)

1. I can assure you that the increased noise is even more striking when in motion!

2. The recordings were 1080/25p (simultaneously to the nano and SxS)

3. Like you, I usually record at 100 Mbps. The 180 Mbps bitrate has been chosen here deliberately, in order to "test extremities", if you will. So was shooting the poorly lit object, rather than - say - a properly lit concert stage, or full sunshine landscape...

Here you go (I can even do TIFF for you, but that would be just wasting the bandwidth :)):

I thought the red flowers from the nanoflash were better in the other thread? Is there a problem just at the 140-180 mbps bitrates or does the 100 mbps show more noise as well?

I guess what's being said is that the Sony compression to 35 mbps HQ is superior to nanoflash 4:2:2?

It was always my opinion EX 35 mbps HQ was regarded unfairly, but what about those videos showing the showering of sparks from welding and grinding, and twisting motion and all that? It was clearly superior at 100 mbps? And the ocean surf, same thing? Now we are reporting nanoflash is noisy? A bit hard to understand. Maybe the 100 mbps bitrate is best overall? Or is the noise happening there too? I sure thought it looked clean.

I still think it's wrong to upscale a crop to 1920x1080? What's the point? The image is being interpolated. The correct thing to do is view crops at 100%. (Appreciate the efforts)

Tom, the only reason I upscaled crops to fullHD size is that - like you - most people cannot see what I'm talking about in the regular, full screen grabs because they watch them on small monitor screens.

If you have access to a 50-incher, just please A-B the two original full screen pictures in the first post of this topic, and you will see the same difference.

Understood.

Thanks Piotr.

The Nanoflash does a better job of preserving detail. Piotr's example using the cabin is a bit unfair because the color of the wood siding is nearly uniform and the increase in detail on the wood siding is perceived to be less valuable than the accuracy increase in capturing the noise. That is, some prefer a smoothed wood surface to a textured, noisy one.

I posted an example of elephant seals on the beach on this forum. The Nanoflash clearly renders the breaking ocean waves with less macro-blocking at 100 MBps. I did another test this weekend on a water faucet+closed hose that to create fine streams leaking from the seal, and the fine water streams showed much more detail on the Nanoflash.

CD should be posting more examples to demonstrate the Nanoflash advantage.

Using images seems a bit flawed to me as evaluation is somewhat subjective.
I'd suggest closing the iris / lens hood and recording black. Then compare the results using a waveform monitor.

In my previous comment it was the waveform monitor in the Panasonic BT-LH1760 that I was looking at with a seriously under exposed image. If I can find the time and the gear is available I'll try to do a more scientific test and take snapshots off the Panny's LCD.
In general though I have to agree with previous comments, the difference in chroma sampling and the way an mpeg-2 encoder works at different bitrates more than likely explain what is happening. That's why I suggested doing an apple to apple comparison, run the CF at the same bitrate as the camera and see what the outcome is.

Rafael,

Even though I believe that for showing a difference, just about any format is OK - I have repeated the two enlarged grabs in PNG (see post#32).

Look guys, it's clearly not an issue with rescaling a blow up or jpg compression. Piotr is seeing the issue and posting examples on the web to illustrate what he's seeing on his own computer w/ the raw footage. The posted images fairly represent what he's seeing, otherwise he wouldn't be wasting his time and everyone else's with this thread.

Okay, how to figure out what's going on.

1) Test the nano at different bit rates and long GOP vs I frame.

2) Test the nano in different lighting conditions.

3) See if where you bought the nano from would give you another unit to test.

4) Set up a static scene at home and record it to EX. Then send the component feed to a monitor. Compare the footage. While it's not looking at HD-SDI, I'm assuming you don't have an HD-SDI monitor, and the EX-1 does not have HDMI. If you see similar noise on the compenent feed, then you know it's the camera.

Just some ideas.

That all said, it would be GREAT for CD to reply here.