4:2:2 vs 4:2:0 example; 4:4:4 would be even nicer...
 

Just for comparison sake, I have blown up a very small fragment of the same scene, recorded both on my EX1 SxS card (using the 35 Mbps, 4:2:0 codec), and on my nanoFlash (50 Mbps, 4:2:2).

One can easily see how the vertical color resolution is doubled in the latter case. But looking at the picture, I've dreaming of how nice the full 4:4:4 color would look - while 4:2:2 removed all the vertical jaggies at the red flower edge, the horizontal edge is still ugly...

Well, it's nice to dream, isn't it :)

Red Camera Stills
 

Hi Piotr:
The Nano Flash recorded still looked significantly superior to the 4:2:0 in terms of both color and resolution. However, both images looked really bad ! Could you post uncompressed .Tiffs to this thread (Sorry, I don't know if this web site can handle Tiffs but it's worth a try). I don't know to what degree the .png stills are affecting the image quality.

EDIT: Piotr, what resolution was the RED camera shooting the flowers at ? 2K, 3K, 4K ?

Or maybe just show us a still of the full frame from either EX1 or Nano to get an idea how much you've blown this up?
Thanks for posting the pictures!

Piotr good discussion topic. Ultimately, topics like this can lead to better use of the nanoFlash.

Just on an anecdotal level (which all of our comments will be without using scopes and such) I find pros and cons in the image quality of both.

Focusing on the nanoFlash image, it should be emphasized that this is a 50Mbit image and as such it is dealing with more color (chroma) information in that bit range which could account for the clearly visible mosquito noise in the darker portions of the image.

I think it would be more instructive to do a series of shots at various bitrates. What this tells me, in a practical sense, as I do a lot of low light work that I should be using higher bitrates than 50Mbits if I expect that I will have to grade in a serious manner in low light situations.

It will be interesting to see if we can determine what factors go into creating mosquito noise.

-Andrew

Gentlemen,

In this little ad hoc example, I didn't want to compare bit rates, but color resolution. Perhaps it will suffice to say that both pictures are some 427x240 pixels fragments, blown up to the full 1920x1080 size.

Piotr I don't believe you can leave bitrates out of the discussion as the bitrate determines the amount of color mangling that occurs with the source. If you are using 50Mbit on 422 color and 35 Mbit on 420 then your result is going to be relatively close as the 50Mbit crunching has to deal with more source data than the 35 Mbit. Providing a higher bitrate image would allow a better "triangulation" to figure out what is playing into the various kinds of distortion that are being introduced.

-Andrew

Andrew,

True, but no bitrate will change the color subsampling method; with 4:2:2 only the vertical color resolution is doubled compared to 4:2:0 (see the graphics in http://dougkerr.net/pumpkin/articles/Subsampling.pdf).

Right, I'm sure that the higher bit rates look better. But this wasn't meant to be an exhaustive test, just an experiment to show how much 4:2:2 really helps.

Andrew and Peter,

You're absolutely right the darker portions of the 50/422 image show a little less mosquito noise around edges than the 35/420 version does. And this is where the higher bitrates, available with the nanoFlash, really shine!

Having said that, I have deliberately compared two images, created with more or less the same effective bitrate (50 Mbps at 4:2:2 isn't really much more than 35 Mbps at 4:2:0, considering it must compress larger amount of the color information). So it's good to know that even in this scenario, there is less mosquito noise in nanoFlash files - at 180 (280), it would probably be virtually invisible.

Dear Piotr,

I think this is a very interesting and informative discussion.

In your example, I calculate that this is 22.2222% of the entire image, or approximate a 450% zoon (if I calculated correctly). I base this on your reporting that the image area is 427x240 and the original image was 1920 x 1080.

Alister Chapman has posted on this blog, an analysis of Sony's 35 Mbps 4:2:0 versus 50 Mbps 4:2:2. He concludes that the compression ratio is about the same.

I think it be worthwhile to show what is possible with the tools that you already have.

Thus, I feel that one could record using I-Frame Only, at 220 Mbps or 280 Mbps I-Frame Only and then zoom in to check the quality. One could compare this to 35 Mbps 4:2:0, or 50 Mbps 4:2:2 or anything else they desire.

What About The Quantization ?
 

Hi Friends:
Very interesting discussion. What about the difference in *Quantization* between the way MPEG 2 @ 4:2:0 (Probably also 1440 x 1080 pixels ??) color space is being encoded versus Full Raster HD 1920 x 1080 pixels @ 4:2:2 color space MPEG - 2 at 50 Mbps ? I have seen 25 Mbps HDV Mpeg - 2 .m2t files with only 1440 x 1080 pixels look visually free of Mosquito noise and blocking artifacts, while I've seen 35 Mbps MPEG - 2 not look as good. Can we ascertain what the Quantization factors are from the two examples ? My point being there are other factors which can be colouring what we are seeing here. EDIT: Data rate isn't everything, but it certainly is an important factor of course.

Hi Mark,
Remember, there are 2 different 35 Mbps codecs - both 4:2:0.
The one for the EX cameras is 1920x1080, the one for the PDW-355 series disc cameras is 1440x1080.
Also, I believe both the 35 Mbps codecs are VBR, as opposed to the 25 Mbps HDV codec and 50 Mbps XDcam422 codec which are both CBR.
Just more variables for the equation...
At least we all seem to agree that the 100 Mbps (and higher) Nano recordings look better than the lower bit rate camera-based ones!

All the Variables !
 

...Hi Dave: Yes indeed, you'e absolutely right :-) Also, HDV MPEG 2 m2t file quality can vary in quality from camera to camera as well @ 25 Mbps encode rate. The one in the Canon XL H1 camcorder is truly excellent, but I understood it to be a VBR encoder, and it was for this reason that it was above average. Perhaps Dan or Chris Hurd would know what kind of MPEG - 2 encoder is in the XL H1 ? Dan I thought the Sony XDCAM HD 4:2:2 hardware encoder CD uses in the XDR & Nano Flash was a VBR encoder ? (I hope I'm not transversing private corporate information asking about this ?)

Dear Mark,

JVC used what has become to be known as HDV 1, 18.7 Mbps (I think)

Sony and Canon used 25 Mbps, known as HDV 2 (as far as I remember).

I do not know if HDV 2 is VBR or CBR.

I do know that our 50 Mbps 4:2:2 and above Long-GOP codecs are CBR (Constant Bit Rate).

Our I-Frame Only and 18 and 35 Mbps codecs are VBR.

Dan,
The HDV as used in Sony camcorders is 25 Mbps CBR.

Mark,
You're right that even the same standard (like the said 25 Mbps 420 HDV) can be implemented better or worse. One of the examples of very poor implementation of that codec has been the Sony's V1 camcorder, which - particularly in its 25p version - suffered from many artifacts (macroblocking, mosquito noise, and some peculiar phenomenon known as "oil pain effect"). It was also very easy to "break"...I owned that camera, and sold it away as soon as the EX1 was announced.

I'm very happy now with the excellent XDCAM EX 35 Mbps VBR codec of the EX1, but even more so - with the codec options offered in Convergent Design recorders!

CBR vs VBR
 

...OK. this explains allot of what I've been seeing in my testing, I have consistently noticed a superior result in MPEG 2 software encoding using the VBR method versus the CBR (Constant Bit Rate) scheme. I think it's a very good idea for CD to employ VBR encoding in lower data rate settings, since they will benefit the most from this superior type of encoding method. This also explains my consistent test results with I-Frame encoding at 280 Mbps vs Long GOP @ 180 Mbps. I find the i-Frame very, very, very clean and noticeably superior to the Long GOP. (When viewing results on my properly calibrated Sony broadcast HD-SDI monitor). If I color my camera output slightly by setting the XL H1 to a preset with cine gamma 1 setting, then I get a much more detailed film look in I-Frame 280 versus even 220 Mbps ! It looks more film like. (I recognize this to be a very subjective point)

Mark,

Frankly, I don't see why a VBR codec should always be superior to a CBR one, when the maximum bitrate of the former equals that of the latter.

All other things equal, they should give identical quality - and the CBR only disadvantage being more media space requirements.

Vbr & cbr
 

....Hey Piotr: OK., but why would you conclude that CBR & VBR yield about the same results quality wise ? A variable bit rate encode yields a file which is usually superior and most importantly able to adapt the encoding process for sections which require more processing (Excessive movement for example) than other sections, thus yielding a smaller file size as well. I don't think bit rate is everything.

True, but if we assume VBR of 25 Mbps MAX vs. a CBR of 25 Mbps CONSTANT - there is no margin in the VBR for the better quality. It's just more efficient size-wise, that's all.

VBR Encode's Margin for Error @ Low Bitrates
 

...Hey Piotr:
I'm not sure this to be strictly true. (??) Usually, you see VBR encoding does the most good, when you have a low encoding bit rate. I believe the encoding margin in VBR is still present as long as the footage being encoded doe not require an increase in encoding variable bit rate above the maximum ceiling of the target bit rate.

Hi Mark,

I guess you may be right, but only when the VBR encoding uses more than a single pass.

If I'm wrong - I'm all ears :)

It's my understanding that if you set a CBR of say 25Mbps you are doing the same as setting a VBR of Max, targeted average and minimum bit rate of 25Mbps. Therefore simply setting a max in the VBR setting of 25Mbps will yield inferior quality to a CBR of 25Mbps. That is because at certain times the bitrate will drop below 25Mbps and will never be above it. However, if you have an average rate of 25Mbps in VBR, with multipass encoding, you should have better picture quality if say you set the max bitrate at 30Mbps and the min at 15Mbps. Then it really depends on the algorithm used to determine the chosen bitrates used for each frame.

Garrett

You've got it nailed, Garrett.

Perhaps I just wasn't clear enough in my messages that I was comparing VBR MAX bitrate with CBR of that same (constant) bitrate.

If you compare VBR with the target, or average, bitrate being equal to that of CBR, of course you have the potential for a better quality from VBR - especially if it's multi-pass.

Mark, was this what you meant?

What Did I Mean ?
 

...Hey Piotr: Yeah-sort of. :-) I've noticed VBR encoding scheme (Wether it be single or double pass) seems to have more of a positive effect on encoding footage when used in low data rates.

i.e. A 25 Mbps encoding rate encoded using VBR instead of CBR looks better than using CBR - even if the VBR is only single pass. This is not to say that CBR encoding cannot also look very clean. I think the VBR will have the edge.

Does anybody know what kind of encoder is in the Canon XL H1 ? (CBR or VBR ?)

Mark and Piotr, you are both right: When comparing VBR with the maximum rate same as the regular CBR, CBR will keep more information, so the result will look better.

When, on the other hand, comparing VBR with the average rate same as the regular CBR, then VBR will most likely look better than CBR. I say most likely because it will still have portions of the video compressed with a lower rate than CBR. But if it makes good choices as to which areas can get by with a lower rate and which require a higher rate, then it will look better overall.

Mark,

When specifying the Data Rate for VBR you need to say if the number you are stating is the minimum, average or maximum rate.

The XL H1 records HDV which is CBR.

Garrett

What VBR Data Rate ?
 

...Hi Garret: Oh yeah, sorry. 25 Mbps is the *Average.* The CBR encoder in this camera is second to *none !*

Mark,

The H1 is a HDV, tape based camera - therefore its codec is CBR 25 Mbps. With CBR, there is no "average", "max" or "min".

Hi Mark,

I use to have an XL H1a and XH A1 and will agree that Canon's HDV encoding is excellent. But, I have to say that my current setup of Sony EX3 with a nanoFlash cranked up above 100Mbps gives a truly stunning picture.

Garrett

Nano & Flash XDR Yield Superior Results
 

Hi Garrett: I agree ! No one's arguing this fact. Although I consider the Canon XL H1 camcorder to have the finest HDV encoding quality I have ever seen, I find the image quality of my Flash XDR recording from my camera's uncompressed HD-SDI output @ even Long GOP MXF 50 Mbps to be truly superior. I find I-Frame Recording with my Flash XDR @ 280 Mbps to be utterly stunning ! By far the results look like someone changed out the internal circuitry of my XL H1 and put in very high end Sony HD camera cirucits instead !

Hi Mark,

I totally agree with you on the Canon HDV encoding. You would think that there wouldn't be much difference from one manufacturer to another but there is. I also think that Canon glass makes a difference.

I still have my trusty little HV20 and will soon try the nanoFlash with it just to see how it performs. I know I'll be limited by the small sensor but if I light the scene correctly I wonder how good I can get it to look.

Garrett

Garrett have you compared the XL, XH and EX pictures all using the nanoFlash? I'm assuming the EX looks a little better than the Caons, and the XL and XH look about the same. Is that accurate?

Thanks much!

P.S. Oh, I'd be interested to hear how the HV-20 looks as well. I wouldn't be all that surprised if the HV-20 w/ the nano looks better than the EX-1 w/ native recording.

Thanks again.

Hi Peter,

I no longer have my Canon Cameras, Had to sell both of them to get the Sony. I'll have to find some time to do the test with the HV20. I've got a bunch of shoots and edits over the next month so I may not be until after I'm done with them.

Unfortunately I doubt that the HV20 + nanoFlash will outperform the native EX3. The HV20 just doesn't have the resolving power in it's small sensor. Give good lighting I'm sure it will outperform many cameras costing the combined $3500 or so that the camera and NF would cost. The native EX3 produces one of the best pictures I've see, especially for a sub $10k cam. I actually was not a big Sony fan prior to the XDCAM EX series.

If I can find some time in I'll shoot some HV20 +NF Footage.

Garrett

Thanks Garrett :).