DV Info Net

DV Info Net (https://www.dvinfo.net/forum/)
-   Sony NXCAM NEX-FS700 CineAlta (https://www.dvinfo.net/forum/sony-nxcam-nex-fs700-cinealta/)
-   -   Picture Profile Test (https://www.dvinfo.net/forum/sony-nxcam-nex-fs700-cinealta/511202-picture-profile-test.html)

Eric Aumen October 6th, 2012 12:04 PM

Picture Profile Test
 
Here is a quick picture profile test we shot to test out some different profiles we found on Abel Cine and Noah Yuan-Vogel’s blog. Aumen Media Sony NEX-FS700 Picture Profile Tests

Noah Yuan-Vogel October 8th, 2012 08:31 AM

Re: Picture Profile Test
 
Thanks for doing that test. I am going to update my nLog profile soon with a couple "color depth" tweaks since that is one of those odd settings that is complicated to master but you absolutely have to set them right for primary colored gradients to look right. You can see this in the do not enter sign in some of the PPs, where there are pieces of its out of focus shape that look over-saturated and represent a discontinuity in the red gamma curve.

Notice in AbelCine's NEXLOG how the highlight range turns pastel looking with a significant loss in detail contrast and that is not restored in grading since it is very difficult to restore properly. It comes from over-compressing the highest part of the highlight range into a luma range that is too small. It is exacerbated by the use of Standard Gamma curve which is known to have issues retaining color fidelity in the highlights. I used to use a similar knee value to the one in NEXLOG when I first got my camera and quickly found that I ran into the issues you see in Abel's PP, so my nLog PP represents an great improvement over those settings.

Eric Aumen October 8th, 2012 09:55 AM

Re: Picture Profile Test
 
Noah, thanks so much for the reply! Also, I would like to thank you for sharing this profile and your knowledge. I am learning so much from guys like you in the industry. Coming from a dslr background, I still have a ton to learn about these settings. I went back and reviewed the footage and I can see what you mean. One thing I recently on Allister Chapman's blog is that lowering the knee value might actually "reduce the recording signal range". Here is the blog post he did recently on a "log" like profile :XDCAM-USER.COM • View topic - FS-700 Flat - Near Log
What are your thoughts?

Alister Chapman October 8th, 2012 10:05 AM

Re: Picture Profile Test
 
The knee and cinegammas don't play well together. Cinegammas have their own built in highlight compression and adding additional knee can send the cinegammas into negative gain. The brighter the overload/overexposure the higher the knee gain correction becomes until it flips the top end of the cinegamma slope into a negative slope and this is really nasty looking.

The color depth settings are a nightmare because they don't actually work on chroma or saturation, they work on the luma gain of the selected color channel. Too much adjustment and strange things happen as the luma gain of highly saturated colours gets altered. Where you have two colours close together on the same object you can have two different luma levels.

Watch for PP's that waste data by using elevated pedestal (black level) settings. Raising the black level does not increase dynamic range, it just raises the pedestal level, everything below the pedestal level is just wasted data bits, with an 8 bit camera you really want every last available bit. With black level there is no gain adjustment only level, you need a gain change to alter dynamic range at the low end. The black gamma setting achieves this but only really if you use the low range setting as the mid and high range settings spread the same amount of gain over such a large range that the low levels remain largely gain free. Use the low range setting and all the gain is applied to the low end and this can bring a half stop DR improvement in deep shadows. Even this gain is marginal as your bringing up the noise floor as well as real picture information so whether you can actually extract meaning full picture information out of the noise is debatable. The 12 useable/13 measurable stops of Cinegamma 4 are probably all you want out of an 8 bit camera anyway.

In addition you also need to use the right amount of camera gain. The dynamic range window of the camera changes with gain/ISO so a picture profile set up at one ISO may not offer the best range at a different ISO.

Noah Yuan-Vogel October 8th, 2012 11:09 AM

Re: Picture Profile Test
 
It's true that you can use a slightly greater portion of the 8bit range with particular settings. And for those who are particularly obsessed about the limitations of 8bit per channel recording, that may an issue, but in my experience those people are usually have little evidence or experience to support their paranoia. And I don't mean to discredit those concerns because they are real and quantifiable but they are often blown out of proportion since a lot of people who know 12 and 10 bit recording formats exist who then hear they are only using 16-235 of their possible 255 8bit levels get really worried that their recordings are inferior without fully understanding why.

That said, I highly recommend anyone with control over their workflow that is doing their own post or with a strong relationship with their post people decide for themselves if they want to use an expanded recording range. On the other hand, for anyone who does not have control over their post workflow, I highly recommend keeping broadcast safe black and white levels by using CINE2 or using CINE4 with my tested knee settings and by using a black level of at least +5 or more. Getting a slightly expanded bit range is great, but it is not worth it if it means you have a good chance of your black or white levels getting clipped out.

I've worked with clients a few times who do their own editing in FCP and once very recently even with my nLog settings with black level set +10 I saw the my FS700 footage on their timeline still appear to have overly hard blacks at or near clipping. No doubt if I had recorded superwhites they would have been clipped too. And a lot of editors or post people don't seem that aware of the issue or don't care enough or have enough time to worry about reclaiming clipped highlight or shadow detail due to recording an expanded range.

Alister Chapman October 8th, 2012 02:07 PM

Re: Picture Profile Test
 
In my opinion based on my tests, Black level above plus 1 is a complete waste of data and gains absolutely nothing except an excessively high pedestal level which should then be corrected to a legal pedestal level in post.
Quote:

Originally Posted by Noah Yuan-Vogel (Post 1757448)
I've worked with clients a few times who do their own editing in FCP and once very recently even with my nLog settings with black level set +10 I saw the my FS700 footage on their timeline still appear to have overly hard blacks at or near clipping.

No surprises there as raising the pedestal clips blacks earlier as it limits the black negative excursions, so after grading and restoring the correct pedestal level, blacks may look more clipped/crushed than they would if the pedestal was at the correct level to start with. On my FS700 +10 raises the pedestal to around 7 IRE (bit 31) and causes blacks to get clipped 2 IRE earlier.

Full range recording is 16 to 254 = 238 bits.
Recording to only 100% = 16 to 235 = 219 bits.
Reduce by raising pedestal by 15 bits = bit range of 204 bits, a total reduction of 17% from whats possible

If you were using standard gammas or a limited dynamic range of 10 stops or less, that 17% would not be all that significant, but every extra stop of dynamic range you add without any compression doubles the amount of luma data you need to record. However, of course Cinegammas, the knee etc are a form of compression that allows use to squeeze extra range into the restricted bit bucket. When you start trying to squeeze 400% more data (Cinegamma 1 and 4) into that same highly restricted bit bucket, every last bit counts, especially if your hoping to pull meaningful picture information from your highlights. Most people expect the main consequence of using fewer bits to be banding, which is only the case in very extreme situations (main cause of banding tends to be compression and quantisation noise). The main consequence in most cases is the inability to extract gradients and textures from highlights simply because not enough data bits were available to record the already compressed signal. If more data bits are used then more subtle textures and gradients are recorded so for example bright clouds in the sky will look more natural and the scene will grade better.

If you are worried about keeping below 100 IRE then use Cinegamma 2 which re-maps exactly the same dynamic range as Cinegamma 2 into zero to 100 IRE. No need to use any additional knee.

Of course, as you say, the best thing with any workflow is to do your own tests, but make sure the tests include different options so you can really make a comparative assessment.

Eric Aumen October 8th, 2012 02:51 PM

Re: Picture Profile Test
 
2 Attachment(s)
Thank you both for your help. I did another test really quick to see how things looked between the nLOG and Allister's Flat. They both seemed to have their positives. The color looked better in the Flat profile and the highlight roll off looked better on nLOG. What do you guys think? The first attachment is nLOG and the second is FLAT. So could I take the color settings from FLAT and the knee settings from nLOG to create the best of both?

Alister Chapman October 8th, 2012 03:20 PM

Re: Picture Profile Test
 
Very close indeed.

Of course you can take the elements that work for you and create your own profile. That's what they are there for.

Noah Yuan-Vogel October 8th, 2012 03:27 PM

Re: Picture Profile Test
 
I wonder if somehow there is a difference between cameras, or maybe its a difference in testing or monitoring methods. I found that according to my waveform monitor connected through hd-sdi and hdmi, black level 0 reads shadow detail below 0 IRE and that a black level of +5 barely brings all detail mostly above 0 IRE. That is why I suggest a black level of 7-10 to avoid losing black detail. Personally I tend to be partial to a smooth image rather than a punchy one since I shoot narrative and commercial content rather than sports or news and would err on the side of low contrast since it is easier to add contrast in post than to remove it contrast and add back detail.

To my eye, if you end up with video with a bit low contrast and black/white levels dulled below the available range, worst case it looks a bit more like a film scan or something off an older RED camera firmware with kind of milky blacks but lots of range, whereas if you clip your black or white level it just starts having that hard clipped look that's over-contrasty and oversaturated like you might get out of a cheap camera with poor dynamic range. I always prefer too little contrast over too much contrast.

I understand that some people are simply more concerned with their bit depth than anything else, and that is up to them in the end, but I tend not to see that as the primary limitation in most scenarios.

Alister, I'm a little confused by your use of the term "bits", but I assume you mean base-10 values rather than the normal base-2 bit. We should also remember that when talking about 8bit recording, we are talking about 8bits per channel or 24 bits per pixel or 16.7million values per pixel. I don't want to scare people but this also means that the actual loss of range measured in bpp is even greater when what it looks like in bpc. For every 1% of the 8bit per channel range you don't use, you'd actually lose 3% of the full range. This does work against me since I'm advocating for using less range, but I think the practicalities of many post environments in my experience make it necessary to forget about bit depth and protect your image unless you have a very real and specific reason or symptom you are seeing that indicates your bit depth is not enough. Frankly I've never really had any posterization issues with 8bit recording formats, probably since a lot of delivery devices are only 6bit anyway and a lot of people cant tell the difference between 6bit with dithering and 8bit, plus a lot of live action video in 8bit tends to be dithered pretty well by video noise.

Noah Yuan-Vogel October 8th, 2012 03:59 PM

Re: Picture Profile Test
 
Quote:

Originally Posted by Eric Aumen (Post 1757488)
Thank you both for your help. I did another test really quick to see how things looked between the nLOG and Allister's Flat. They both seemed to have their positives. The color looked better in the Flat profile and the highlight roll off looked better on nLOG. What do you guys think? The first attachment is nLOG and the second is FLAT. So could I take the color settings from FLAT and the knee settings from nLOG to create the best of both?

The color just looks a little more green in the FLAT PP, because of the -2 color phase compared to nLOG's +3. I think the pinker skin tones look more natural, but obviously that can be very subjective and can vary depending on lighting conditions and scene content.

The shadows in the FLAT PP in that image have something really odd going on unless its my monitor or something. It looks like the shadows levels are a bit lifted but clipped harshly, so there are hard blacks but the slope of the black gamma is very high at the clip point so it looks discontinuous and unnatural. This seems to make sense given the +1 black level and +7 black gamma (and "low" range) which in my experience would have that effect. I would use a higher black level to avoid prematurely clipping blacks especially since the high black gamma gives the curve a high slope at that point, or you could use a lower black gamma value so the slope is less drastic at the point of black clip.

I'd generally avoid using black level values below +5 since it will tend to hard clip your blacks, which effectively loses some range and can increase the appearance of noise since if you hard clip shadows, it effectively creates a discontinuous (hard edge) at the bottom of your gamma curve with very high contrast at the point of black clipping.

In the second image, the color saturation appears too high on the right side of his face and too low on the left side, and I think that is because the gamma curve is doing something quite unnatural on the low end.

Alister Chapman October 9th, 2012 10:41 AM

Re: Picture Profile Test
 
My FS700 does not exhibit anything below the pedestal at any black level setting and I wouldn't expect it to, if yours is Noah then that's not right. The black level setting is a zero gain black bias adjustment so should have zero effect on dynamic range or clipping and should do nothing other than alter the black pedestal or setup level. Raising the pedestal should not reduce clipping, it should make no difference at all. I'm on the road at the moment, but when I get back later in the week I pull some images off the WFM. I assume you are only looking at Y and not composite, composite would show negative going excursions, but that's normal.

I think my Flat profile is capturing a lot more shadow information. When you grade the two images you can really pull out the texture of the darker bricks, while nLog leaves the bricks looking smooth and textureless. However nLog does appear to hold on to highlights better and you can see more branches against the bright sky (assuming there is no change in the cloud between shots)
Of course we are both splitting hairs over minutia, obsessing over the tiniest of details. Neither profile is right or wrong, just different approaches to a similar goal. Which one is right for you depends on many factors including whether you have a high or low key scene and your workflow.

Eric Aumen October 9th, 2012 10:52 AM

Re: Picture Profile Test
 
Thank you both VERY much for the help. I used a combination of your profiles to film an interview and some workout footage for a mini doc, and I was very pleased with the image! Also, I am learning a ton of great info about these settings. Thanks again for sharing your knowledge guys.

Noah Yuan-Vogel October 9th, 2012 03:37 PM

Re: Picture Profile Test
 
Quote:

Originally Posted by Alister Chapman (Post 1757637)
I think my Flat profile is capturing a lot more shadow information. When you grade the two images you can really pull out the texture of the darker bricks, while nLog leaves the bricks looking smooth and textureless. However nLog does appear to hold on to highlights better and you can see more branches against the bright sky (assuming there is no change in the cloud between shots)
Of course we are both splitting hairs over minutia, obsessing over the tiniest of details. Neither profile is right or wrong, just different approaches to a similar goal. Which one is right for you depends on many factors including whether you have a high or low key scene and your workflow.

It seems like all the detail you are seeing in the bricks is only more noise displayed with higher contrast. From what I can see, I think nLog is a more even distribution of tonal detail, whereas FLAT has very high contrast at a particular point which changes very quickly and leaves too little contrast in the range that seems to contain the detail in the left side of the mans face where things get unnaturally washed out and grey/desaturated compared to the darker tones in the bricks and hair.

That's why I don't really like the Black Gamma Range "LOW" setting, because lifting at only the lowest end of the curve is too drastic of a change to too small a portion of the gamma curve. When dealing with these coarse adjustments, I find delivers a better result to be extremely careful not to change the slope of the given gamma curve too drastically through any adjustment. I have found the HIGH setting tends to spread the gamma adjustment over a wider portion of the shadow range, giving a similar effect but with more natural results.

Alister Chapman October 9th, 2012 11:46 PM

Re: Picture Profile Test
 
No it's not just noise. It's real image detail, but yes it is noisy and that's just the expect consequence of trying to bring signals out of the noise floor to increase DR. The mid and high range adjustments make no appreciable difference to dynamic range. Yes they lift information that's already there in the image (including noise) but they do not increase DR, although with a bit starved recording they can help with grading as you will perhaps be more inclined to shift levels down in post to restore contrast.

The only way to gain more DR in the shadow and low key areas is to add gain at the very bottom of the cameras input range to try and squeeze the last bit of data out of the low end of the sensors output. The mid and high range gamma adjustments don't really do that, their gain is applied over such a broad range that it has almost zero effect at the bottom end. Again when home I'll pull off some WFM frames so you can see the extra DR gained. You really can pull about extra half stop out of the shadows, but only by using the low range, but it does of course come at the expense of noise. If you look at the guys hair you can very clearly see nLOG is hard clipping the blacks (probably because of the raised pedestal) and as a result loosing a lot of the finer texture and detail which is still clearly retained in FLAT.

Sami Sanpakkila October 25th, 2012 05:10 PM

Re: Picture Profile Test
 
Quote:

Originally Posted by Eric Aumen (Post 1757643)
Thank you both VERY much for the help. I used a combination of your profiles to film an interview and some workout footage for a mini doc, and I was very pleased with the image! Also, I am learning a ton of great info about these settings. Thanks again for sharing your knowledge guys.

Hi Eric! Would you mind sharing settings for this profile? I'd be keen to try. I'll be getting my camera in a week or two but already shot one project on a rental FS700. Was very pleased with the image but I think our picture profile was not setup optimally for the shoot.

PS. An offtopic question, how did you get the "Check out the blog post!" text and link to appear after your Vimeo video played? Is it coded on your webpage? Can't see it as an option in Vimeo settings.


All times are GMT -6. The time now is 03:03 AM.

DV Info Net -- Real Names, Real People, Real Info!
1998-2024 The Digital Video Information Network