Straining EX HQ codec at DVinfo.net

Piotr Wozniacki · October 26th, 2009, 07:06 AM

I probably am considered an extreme measurebator by folks here, but what the heck :)

I'd just like to share with you guys an interesting observation I've done recently. Well, to start with, ever since the EX cameras came into existence, all Picture Profile-related threads where people talked about Detail setting, recommended turning it off (or to dialing it down to some negative value). Myself, I usually kept Detail off to avoid those artificial edge enhancements (with the exception of when using my Letus adaptor with Canon FD lenses, which generally benefited from setting Detail to ON).

Now, I've performed some tests recently, with the goal of answering the following question:

- just what happens if I not only keep Detail on, but even crank it up considerably, like to e.g. +35 ?

Well, the results I got are interesting. It was shot with Detail at +35 on a pretty windy day, with the trees foliage moving vigorously... The right-hand side screen grab shows the entire, 1920x1080/25p frame, and it looks pretty sharp. But look at the left-hand side one; it's a 640x360 fragment of the original frame, blown up to fill the 1920x1080 screen. While those regular and static fragments are OK (low-rez of course due to up-scaling, but acceptable), the foliage is one hell of a mess! Not only is it terribly blocky, but I would even say it reminds me of the infamous "oil paint effect" the early V!E suffered in 25p mode!

No, am I right that it looks like the 35 Mbps EX HQ codec has been broken? What do you think?

And most importantly: would the NanoFlash / XDR 100 Mbps Long-GOP codec show its robustness in a scenario like this? Comments welcome, especially from those already using CD recorders with their EX cameras...

Alister?

David Issko · October 26th, 2009, 02:19 PM

Quote:

Originally Posted by Piotr Wozniacki

And most importantly: would the NanoFlash / XDR 100 Mbps Long-GOP codec show its robustness in a scenario like this? Comments welcome, especially from those already using CD recorders with their EX cameras...

Alister?

Piotr,

I can understand your curiosity to see what happens with high detail setting. It artificially changes things, and perhaps the setting range is too great. However, just like say the colour saturation setting on a monitor that has too great a range. It's there but you would pretty much never use it at its high end.

I can say that I have never had the detail setting on my EX3 up so high. It does not make sense to do so for any job unless you are demonstrating extremes & limits. Personally, I have my camera's detail on with just a little 'enhancement'.

I have not tried this experiment with my nanoFlash, so I cannot comment - sorry but I spent my time when I received my camera trying to get the best out of it and it never entered my mind to see what too high a level of detail might produce.

To me, what you recorded is an artificial image and I'm not sure that you 'broke' the codec.

Best wishes

Alister Chapman · October 26th, 2009, 02:30 PM

Difficult to tell from a frame grab, but it does look blocky. It could also simply be blur caused by the movement of the foliage. It does look suspiciously like the codec is struggling. Foliage in the wind is a hard test for any codec. You need a NanoFlash, it wouldn't struggle with this at 100Mb/s or higher ;)

Piotr Wozniacki · October 26th, 2009, 02:40 PM

Quote:

Originally Posted by David Issko

To me, what you recorded is an artificial image [...]

But of course David - it's very unnatural, and I did it for testing purposes only (you know, between some serious jobs, I still like to experiment with this marvellous camera settings).

On the other hand, *some* people really do like it sharp - and even unnatural look doesn't frighten them... I guess many people still don't quite get the difference between "high resolution" and "sharp" :)

Another strange thing I can see in those grabs is that it looks like the upper part of the frame is softer than its bottom. Haven't got a clue why; the only information I can share is that I was wearing a grad ND filter...

Ned Soltz · October 26th, 2009, 02:58 PM

I would second Alister's observation, having been working with an evaluation unit of the NanoFlash for an upcoming review. What you've basically done is break the 35 Mbps codec which is a very good thing for testing the limits. NanoFlash @ 100 Mbps will be considerably more robust.

Ned Soltz

Bob Grant · October 26th, 2009, 03:43 PM

Multiple things can happen in this test scenario. The detail enhacement logic looks for edges to enhance. Anything that's out of focus misses out as would anything affected by motion blur. It could be that much of what is being seen is happening before the image reaches the encoder.

A more controlled test would have been better. On a windy day it's quite possible for the camera itself to be shaken around affecting the whole frame. A test chart on a turntable would make for a more controlled test.

Bo Skelmose · October 26th, 2009, 05:03 PM

I once had the same problems in a DVCam recording - I interviewed a person with a a lot of leaves in the background - The leaves was moving in the wind and the persons face turned into big square pixels - yes that is the problem when you compress something that hard - guess it is not better to compress HD into 35 MbS.

Peter Moretti · October 27th, 2009, 12:59 AM

Piotr,

I have no idea how you can tell how much of the foliage blur is caused by the shutter speed versus the codec crapping out. But XDCAM EX is long GOP, so I have to think movement isn't its best friend.

I also have to think the NanoFlash at 100 would be an improvement.

Dean Sensui · October 27th, 2009, 04:12 AM

I shoot fish coming through propwash all the time. The entire frame consists of constantly changing pixel values.

Hadn't experienced a codec failure ever.

Piotr Wozniacki · October 27th, 2009, 04:50 AM

Quote:

Originally Posted by Dean Sensui

Hadn't experienced a codec failure ever.

Neither have I, Dean -- not with Detail at zero or below. This is my point.

Piotr Wozniacki · October 27th, 2009, 09:49 AM

Funny... The test was conducted using:
Detail = +35
Crispening= +40
Frequency = +40
Cine1 gamma.

You may believe it or not, but I repeated the test with both Crispening and Frequency back to zero, while maintaining the extreme Detail setting.

Result: No blockiness!

In order to try and make this thread as conclusive as possible, here is my kind request to Alister:

- since I've been following your suggestions about the Frequency and Aperture settings Which work fine with Detail off), what do you make of all this?

- also, could you please try and shoot some samples with Detail cranked up to +35 (or even 50), but recording it to the NanoFlash?

I'd be very grateful if you share your findings :)

Cheers

Piotr

Alister Chapman · October 27th, 2009, 10:25 AM

Turning up detail and frequency is never a good idea as you are significantly increasing or even adding lots of very fine edges into the image which will almost certainly increase aliasing. What we may actually be seeing is excessive aliasing hiding the detail in the foliage.

I am out of the office for the next couple of weeks doing presentations on the Televisual Roadshow in the UK so won't be able to shoot anything for a week or so.

Piotr Wozniacki · October 27th, 2009, 11:04 AM

Quote:

Originally Posted by Alister Chapman

Turning up detail and frequency is never a good idea as you are significantly increasing or even adding lots of very fine edges into the image which will almost certainly increase aliasing.

Alister, have you changed your mind since this post?:

http://www.dvinfo.net/forum/1019869-post17.html

I have endorsed your recommendations contained therein, and have been quite satisfied with it ever since :)

Quote:

Originally Posted by Alister Chapman

What we may actually be seeing is excessive aliasing hiding the detail in the foliage.

I don't think it's aliasing that's causing the blockiness. If aliasing was really that excessive, what I would see in the first place would be line twitter (like those fine horizontal lines between the wooden boards on the shed wall; this is what happened with the V1E when over-sharpened).

Again, what an amazing technology this is!

Tom Roper · October 27th, 2009, 11:10 AM

It's not a good idea to broadly generalize when there are cases to be made for it. For example, the BBC white papers for the EX1 list a setting of +30 for the Frequency. Sony gives us these adjustable settings so that we can meet the needs of our particular circumstances.

The Detail setting seems to evoke strong negative feelings, but my own experience with it has taught it can be manipulated through the various parameters to achieve a detailed image without obvious noise or edge ringing. A clean image and the detail On setting are not mutually exclusive.

Piotr Wozniacki · October 27th, 2009, 11:17 AM

Quote:

Originally Posted by Tom Roper

A clean image and the detail On setting are not mutually exclusive.

I certainly hope so, Tom. Will be experimenting further to find that "sweet spot" :)

October 26th, 2009, 07:06 AM	#1
Piotr Wozniacki Inner Circle Join Date: Aug 2006 Location: Poland Posts: 4,086	Straining EX HQ codec I probably am considered an extreme measurebator by folks here, but what the heck :) I'd just like to share with you guys an interesting observation I've done recently. Well, to start with, ever since the EX cameras came into existence, all Picture Profile-related threads where people talked about Detail setting, recommended turning it off (or to dialing it down to some negative value). Myself, I usually kept Detail off to avoid those artificial edge enhancements (with the exception of when using my Letus adaptor with Canon FD lenses, which generally benefited from setting Detail to ON). Now, I've performed some tests recently, with the goal of answering the following question: - just what happens if I not only keep Detail on, but even crank it up considerably, like to e.g. +35 ? Well, the results I got are interesting. It was shot with Detail at +35 on a pretty windy day, with the trees foliage moving vigorously... The right-hand side screen grab shows the entire, 1920x1080/25p frame, and it looks pretty sharp. But look at the left-hand side one; it's a 640x360 fragment of the original frame, blown up to fill the 1920x1080 screen. While those regular and static fragments are OK (low-rez of course due to up-scaling, but acceptable), the foliage is one hell of a mess! Not only is it terribly blocky, but I would even say it reminds me of the infamous "oil paint effect" the early V!E suffered in 25p mode! No, am I right that it looks like the 35 Mbps EX HQ codec has been broken? What do you think? And most importantly: would the NanoFlash / XDR 100 Mbps Long-GOP codec show its robustness in a scenario like this? Comments welcome, especially from those already using CD recorders with their EX cameras... Alister? Attached Thumbnails __________________ Sony PXW-FS7 \| DaVinci Resolve Studio; Magix Vegas Pro; i7-5960X CPU; 64 GB RAM; 2x GTX 1080 8GB GPU; Decklink 4K Extreme 12G; 4x 3TB WD Black in RAID 0; 1TB M.2 NVMe cache drive

October 26th, 2009, 02:30 PM	#3
Alister Chapman Inner Circle Join Date: Dec 2005 Location: Bracknell, Berkshire, UK Posts: 4,957	Difficult to tell from a frame grab, but it does look blocky. It could also simply be blur caused by the movement of the foliage. It does look suspiciously like the codec is struggling. Foliage in the wind is a hard test for any codec. You need a NanoFlash, it wouldn't struggle with this at 100Mb/s or higher ;) __________________ Alister Chapman, Film-Maker/Stormchaser http://www.xdcam-user.com/alisters-blog/ My XDCAM site and blog. http://www.hurricane-rig.com

October 27th, 2009, 12:59 AM	#8
Peter Moretti Trustee Join Date: Mar 2007 Location: Sherman Oaks, CA Posts: 1,259	Piotr, I have no idea how you can tell how much of the foliage blur is caused by the shutter speed versus the codec crapping out. But XDCAM EX is long GOP, so I have to think movement isn't its best friend. I also have to think the NanoFlash at 100 would be an improvement. __________________ Avid Media Composer 3.1.3. Boris Red and Continuum Complete. Vegas 8.0c. TMPGEnc Xpress Pro 4.0

October 27th, 2009, 04:12 AM	#9
Dean Sensui Inner Circle Join Date: Oct 2001 Location: Honolulu, HI Posts: 2,054	I shoot fish coming through propwash all the time. The entire frame consists of constantly changing pixel values. Hadn't experienced a codec failure ever. __________________ Dean Sensui Exec Producer, Hawaii Goes Fishing

October 27th, 2009, 09:49 AM	#11
Piotr Wozniacki Inner Circle Join Date: Aug 2006 Location: Poland Posts: 4,086	Funny... The test was conducted using: Detail = +35 Crispening= +40 Frequency = +40 Cine1 gamma. You may believe it or not, but I repeated the test with both Crispening and Frequency back to zero, while maintaining the extreme Detail setting. Result: No blockiness! In order to try and make this thread as conclusive as possible, here is my kind request to Alister: - since I've been following your suggestions about the Frequency and Aperture settings Which work fine with Detail off), what do you make of all this? - also, could you please try and shoot some samples with Detail cranked up to +35 (or even 50), but recording it to the NanoFlash? I'd be very grateful if you share your findings :) Cheers Piotr __________________ Sony PXW-FS7 \| DaVinci Resolve Studio; Magix Vegas Pro; i7-5960X CPU; 64 GB RAM; 2x GTX 1080 8GB GPU; Decklink 4K Extreme 12G; 4x 3TB WD Black in RAID 0; 1TB M.2 NVMe cache drive Last edited by Piotr Wozniacki; October 28th, 2009 at 02:16 AM.

October 26th, 2009, 02:58 PM	#5
Ned Soltz Major Player Join Date: May 2006 Location: Teaneck, NJ Posts: 659	I would second Alister's observation, having been working with an evaluation unit of the NanoFlash for an upcoming review. What you've basically done is break the 35 Mbps codec which is a very good thing for testing the limits. NanoFlash @ 100 Mbps will be considerably more robust. Ned Soltz

October 26th, 2009, 03:43 PM	#6
Bob Grant Trustee Join Date: Nov 2005 Location: Sydney Australia Posts: 1,570	Multiple things can happen in this test scenario. The detail enhacement logic looks for edges to enhance. Anything that's out of focus misses out as would anything affected by motion blur. It could be that much of what is being seen is happening before the image reaches the encoder. A more controlled test would have been better. On a windy day it's quite possible for the camera itself to be shaken around affecting the whole frame. A test chart on a turntable would make for a more controlled test.

October 26th, 2009, 05:03 PM	#7
Bo Skelmose Major Player Join Date: Apr 2008 Location: Denmark Posts: 495	I once had the same problems in a DVCam recording - I interviewed a person with a a lot of leaves in the background - The leaves was moving in the wind and the persons face turned into big square pixels - yes that is the problem when you compress something that hard - guess it is not better to compress HD into 35 MbS.

October 27th, 2009, 10:25 AM	#12
Alister Chapman Inner Circle Join Date: Dec 2005 Location: Bracknell, Berkshire, UK Posts: 4,957	Turning up detail and frequency is never a good idea as you are significantly increasing or even adding lots of very fine edges into the image which will almost certainly increase aliasing. What we may actually be seeing is excessive aliasing hiding the detail in the foliage. I am out of the office for the next couple of weeks doing presentations on the Televisual Roadshow in the UK so won't be able to shoot anything for a week or so. __________________ Alister Chapman, Film-Maker/Stormchaser http://www.xdcam-user.com/alisters-blog/ My XDCAM site and blog. http://www.hurricane-rig.com

October 27th, 2009, 11:10 AM	#14
Tom Roper Trustee Join Date: Jun 2004 Location: Denver, Colorado Posts: 1,891	It's not a good idea to broadly generalize when there are cases to be made for it. For example, the BBC white papers for the EX1 list a setting of +30 for the Frequency. Sony gives us these adjustable settings so that we can meet the needs of our particular circumstances. The Detail setting seems to evoke strong negative feelings, but my own experience with it has taught it can be manipulated through the various parameters to achieve a detailed image without obvious noise or edge ringing. A clean image and the detail On setting are not mutually exclusive.