Long GOP Image Quality Considerations
 

Hi Rafael:
I have done extensive image quality testing with my Flash XDR recording @ all Long GOP data rates. As far as I can tell, this Sony XDCAM HD 4:2:2 hardware encoder used in both the Nano Flash and Flash XDR is optimized to produce high quality results @ only 50 Mbps to go inside XDCAM camcorders to record to disk. Therefore, in Long GOP the sweet spot cannot be anything other than @ Long GOP 50 Mbps. In I-Frame (Intra) recording, the sweet spot is 220 to 280 Mbps. Only the high end I-Frame data rates are actually superior to the Long GOP 50 Mbps data rate. IMHO, there's no such thing as a "sweet spot" at 100 Mbps, or higher. I think you may have been labouring under the false impression that your camera's output is preventing you from obtaining better results at higher data rates because of "so called" noisy video signal output. With all due respect to you, I find this explanation to be nothing short of an insult to my intelligence.

Dear Rafael and Dan,

It is a good feeling I'm not alone :)

On the other hand - while I share Rafael's opinions in 100% - I'm afraid the two of us (well, plus Gints perhaps) is not enough to make CD look into the problem seriously. If most customers seems to be satisfied, why should they care - am I right here, Dan?

This is not criticizing CD's attitude at all - just stating the obvious fact that when - in order to make just a couple of customers even happier - considerable resources would have to be involved, it wouldn't be a very pragmatic way of operating in the business environment. I'm sure Dan would confirm my thinking :)

Like you, Rafael, I also used to be very disappointed that the L-GoP picture quality is not at all scalable with the bit rates. But I guess I've already settled with my work-arounds...

Rafael, I'm just trying to be realistic here; with the large sensor cameras' advent, S/N of 60b and above is going to become standard. I haven't tried it personally, but I suspect that with such a low noise sources, even the 180 Mbps L-GoP picture from the nanoFlash looks gorgeous. Also, I agree that, as you suggest, a camera could be eliminated as the picture source for the sake of proper tests - but...and I'm going to be very frank with you now... not being able to afford a better camera than my EX1 in the foreseeable future, I'd rather not know how much better my video COULD look.

I guess this attitude has a lot to do with my age and health problems - I never used to compromise like that in the past. Nevertheless, I'll always back up your efforts and help you in testing things.

I hope you understand,

Piotr

Dear Rafael and Piotr,

I do not agree at all that 50 Mbps Long-GOP is the new sweet spot.
The compression ratio for 35 Mbps 4:2:0 and 50 Mbps 4:2:2 is approximately the same.

We have over 2,800 nanoFlashes in the field. Many are using 100 Mbps.
We have the results of the Video Clarity System.

We have had numerous comments from professional people that they can see a visual improvement when switching from 50 Mbps to 100 Mbps.

Piotr, in the past you have, at times, been reluctant to share the source files, camera settings, etc., that you used to make your judgments.

With the images that you have shared, they have tended to be in less than ideal lighting conditions, and you have not provided your camera settings.

Thus, we cannot independently confirm your findings or conclusions.

A video recorder cannot distinguish noise from video detail.
Lower bit-rate recordings will hide some of the noise.
Higher bit-rate recordings will show more detail and noise.

We have thousands of people who are happy with the image quality they are getting from their nanoFlashes and quite a few use 100 Mbps Long-GOP. I do not know how many, but I talk with many of our customers.

There are, of course, fewer people that use 140 Mbps, and 180 Mbps Long-GOP.

My own personal recommendations are to shoot 100 Mbps Long-GOP (unless you are using certain versions of Avid) and then use 140 Mbps Long-GOP if one expects an extreme amount of detail or motion.

The high bit-rate I-Frame Only modes are recommended also in special cases.
And I never recommend shooting 100 Mbps I-Frame Only.


Rafael, I highly appreciate your suggestion for a carefully controlled test.

I feel that, if such a test is performed, the details of the test should be public.

If a camera is used as the source, we should be able to see/access the source files as recorded in camera. We should have the camera details, such as Picture Profile, gain settings, aperture, lens used, lighting levels, etc. It should be an open test so that others can comment.

Ideally, we would have the HD-SDI recorded as uncompressed, then compared to the in-camera and nanoFlash files.

If we use carefully generated and clean (as noise free as possible) source, then the test can be more meaningful.

Sony spent untold millions of dollars developing the Codec module that we use in the nanoFlash.

We have control of various internal aspects of how it works.

But, please understand that we need to maintain compatibility with many different Non-Linear Editors.
We would be doing a grave disservice to our customers if we introduced something that did not work with a wide variety of NLE's.

We also like to be very responsive to our customer's wishes.

We fully support any testing that anyone wants to perform, but for the tests to be meaningful to us, we need lots of details.

And yes, technically there are settings that could be changed.

But, I suggest that proper, controlled tests be run first.

And I welcome others to comment on their own "Sweet Sport" with the nanoFlash.

Please understand that when we tell others about the "Sweet Spot", we are taking into consideration image quality and file size.

The nanoFlash does not care, and neither do we if someone wants to record at a different bit-rate or in a different mode. Many shooting situations and quality needs are different.

….Hi Dan:
In my testing and shooting use, I have found 50 Mbps Long GOP to be the real "sweet spot." I cannot see any appreciable difference on my Sony Broadcast HD-SDI HD monitor, which would justify the large file size of recording 100 Mbps files. However, I have found I-Frame 220 Mbps and 280 Mbps recording to be far superior to any Long GOP recording data rate. I can see a demonstrable difference between Long GOP 50 Mbps and I-Frame 280 Mbps.

Given the current quality fluctuation from I to P/B frames, I don't trust material at 140 MBps LongGOP. Step through some frames of detailed Long-GOP material and see if you notice this quality fluctuation. I saw a mushier frame roughly every six frames in Sony Vegas for 1080p30 video, though I don't have an explanation for the length of that interval. In the past, I described this as a sharp frame, but really, it's just a mushy frame in which details across the frame are softened. Is this is an artifact of SonyVegas? I just repeated this with 720p60 video at 140 MBps LongGOP. In that footage, the artifact was a mushy frame nearly every six frames. When I couldn't detect a mushy frame every 6 frames, the mushy frame showed up 12 frames later, but not every time.

EDIT : I barely see this quality fluctuation at 100 MBps LongGOP. An given that none of the footage (Piotr's or mine), is staged well, a good test at 35 Mbps, 100 MBps and 140 MBps LongGOP and a high I-Frame rate is necessary.
Long-GOP should provide a coding efficiency of 2-3x over I-Frame only. So, we should be seeing 100 MBps LongGOP with quality equivalent to I-Frame only 220. A ConstantQ implementation would fluctuate in bitrate and provide higher bitrates when needed. The value here is the preservation of the quality of the material, not so much the file size. A Constant Q implementation may also provide a method to estimate the bitrate needed for particular material. A real-time bitrate monitor in the case of VBR or even a bitrate recommendation function would be useful to help select the bitrate or quality needed for a scene - a lofty request indeed.

Dear Piotr,
I'm sure that if we show evidences that the NANO efficiency is not as expected, CD will have a look on that.
I have also an EX-1 as my main camera, so I have the same concerns than you, but if we want to know bout the NANO, we have to eliminate variants as cameras.
We don't need any special kind os test, but a workflow where the only possible added noise comes from the NANO.
Where we can compare a clean picture, with flats areas, gradients, fast moving areas, high detail areas and noisy areas. Easy to compare and easy to identify individual frames.
Basically we have been comparing SxS files with the NANO files. Those are very poor conditions to extract any valid conclusion.

Dear Dan,
I don't say that "50mbps" is the sweet spot. I'm very happy working at 100Mbps. That's the average data-rate I had in mind before buying the NANO.
I haven't had time to make any kind of half-serious test as I all ready said.
I'm very happy with the noise of my EX-1 and I have no reasons to think that the NANO introduces any noise.
Dan, the "Efficiency Curves " tests that you brought here, for my self, are most clarifying, but as camera/editors they result very "theoretical" for us.
We need to see and compare on the picture.
I don't think we need too complicated tests. Most of the times zooming just a bit is enough to find what you are looking for.
I have no power-horses to work on HD Uncompress, but I don't thin k that necessary.
I think that something like Prores HQ or AVID DNx would be more good.
The test will work if is:
- Easy to compare and spot differences.
- Identify single frames.
- Easy to share and compare on every body computer.
I'll post you a possible test picture.
rafael

Except for the fact that codec efficiency and advantages may decrease w/ data rate. That's why AVCHD looks so much better at low bit rates, but at high bit rates, MPEG-2 and other "old" codecs can look very good.

Long-GOP works great when frames are mostly the same. But as you give more data rate for the codec to use, you have more detail and more slight variation from frame to frame, so the efficiency advantage of Long-GOP diminishes. Adding data rate to the codec also improves the quality of the I frames, so the potential lower quality of P and B frames may become more objectionable. Which is why I know the proposal for changing the size ratio for I, P and B frames. But the easiest solution is to throw some more bandwidth at the problem and make everything I-frames. I don't see the big issue with that.

So I agree with Piotr (who's posts have been a great contribution to the nano's acceptance and real benefit to CD, even if it doesn't always feel like it for them) that this is an academic discussion. If possible, it would be a lot of work to save space that is getting cheaper all the time.

And when you throw in that NLE's will have to work with this modified Long-GOP format for it to be useful, I think it really becomes a non-starter. JMHO, but I think the rationale makes sense.

Thank you, Peter, for your correct interpretation of my posts' true intention :)

After a year of 100-140 Mbps LongGOP, going I-Frame is my conclusion as well, and the Nano 220-280 Mbps I-frame looks great. My major regret is that I thought moving from 100 Mbps to 140 Mbps LongGOP would yield better video, and due to the interframe quality fluctuation, it did not. Since I shot in parallel with an EX1 with 16 GB SDHC and Nano with 64 GB CF (2 x 32GB), I just figured I may as well use the CF space for 4x the EX1 35 Mbps data rate.

The bottom line then Gents, either use 50Mb L-GOP or 280 i - frame. I'm happy with that.

To get even better quality from my EX3, i'm going to have to upgrade the lens.

Now You've Got it ! :-)
 

…Hi Kris & Gints, & Piotr: Now you have the three maximum quality settings of the Nano Flash & Flash XDR.

As one of the seemingly few XLH1/nano users, at least compared with EX1/3s, and one who videos wildlife, mainly birds , what would be a recommended best setting? I'm guessing 220 Mbs I-frame.
(I use PhotoFast 533x 64gb cards)

Ron

Dear Ron,

Our 280 Mbps is our best quality.

Our 220 Mbps is our second best quality and it is very close to 280 Mbps in quality.

Thanks, Dan. One of my CF Cards has some footage taken at 100Mbps Long GOP. Can I continue using this card with a 220Mbps I-Frame setting?


Ron

Dear Ron,

Yes, you can, no problems. The nanoFlash can handle this fine.

You can actually mix and match any footage on the nanoFlash.

However, If you are playing back the footage on the nanoFlash, when one clip ends, and the next one is in a different format, such as changing from HD to SD or 1080 to 720, the picture will break up as the nanoFlash has to adjust to the next format on the fly.

When you import the clips into your editing system, all will be fine.

If you playback any clip, starting with one format, any clip on the card, it will play back properly.
The only problem is when the format of the clips changes on the fly.