NX1 Extended Dynamic Range? New Settings.

[KnightsFan]

In my experience, reducing RGB below 1.00 makes it so that you can't have anything actually be pure white. None of your test shots have pure white to begin with, so you won't see those negative side effects. Stick a lamp in the background behind the color checker and you should see the 0.75 getting clipped on the scopes.

One thing I do notice is the white balance shifting, which is an unfortunate feature of the NX1: color is not uniform across its dynamic range.

[mnewxcv]

12 minutes ago, KnightsFan said:

In my experience, reducing RGB below 1.00 makes it so that you can't have anything actually be pure white. None of your test shots have pure white to begin with, so you won't see those negative side effects. Stick a lamp in the background behind the color checker and you should see the 0.75 getting clipped on the scopes.

One thing I do notice is the white balance shifting, which is an unfortunate feature of the NX1: color is not uniform across its dynamic range.

Yes, I pointed out a few pages back that when set to 0.50, IRE is clipped at 92. I am trying to find how low you can go without clipping below 100, so I threw a dart basically and decided to try 0.75. If it doesnt work, I will go higher, and if it does work, I will go lower. My colorchecker shows 'pure white', based on what is lighting the scene, but I do not have anything brighter than that in the scene like you mention.  I just took some shots with light sources in them for the overexposure test. Stay tuned

[KnightsFan]

Well, I went to grab some pics to show what I mean, and I may have caused more problems for us. Turns out the clipping happens in Gamma DR but not Gamma C. These were done with RGB set to 0.7 ish. In Gamma C white still peaks, while in DR it has a hard clip.

But maybe we can use this info to estimate the processing order for the picture. My initial guess is that this shows that the RGB sliders multiply the R,G,B channels on a linearly encoded image before the gamma curve is applied. (That would make sense, because linear multiplication on a gamma-encoded image is a recipe for disaster!) Then, after the RGB multiplier, the gamma curve is applied. Gamma DR retains all/most dynamic range from the linear image, and thus you see the clipping, whereas Gamma C clips the highlights somewhere before the 0.7 value, so we don't see that hard clip on the scope.

More testing required, I guess!

 

1 minute ago, mnewxcv said:

Yes, I pointed out a few pages back that when set to 0.50, IRE is clipped at 92. I am trying to find how low you can go without clipping below 100, so I threw a dart basically and decided to try 0.75. If it doesnt work, I will go higher, and if it does work, I will go lower. My colorchecker shows 'pure white', based on what is lighting the scene, but I do not have anything brighter than that in the scene like you mention.  I just took some shots with light sources in them for the overexposure test. Stay tuned

As per my post above, which Gamma setting are you on?

In Gamma DR, I see sub-white clipping almost immediately. Zebras disappear entirely by 0.87. You can still see some noise above the hard clip on the waveform, but I think that's due to HEVC compression estimations and does not represent any meaningful highlight information.

[mnewxcv]

Those Gamma C results sure are a monkey wrench in the mix. I am shooting normal (no gamma dr/C) for my tests. I tested with a light source present and premiere measures it to be 255 (on a 0-255 scale) for the light source with 1.99RGB and 1.00RGB, and shows 252 for 0.75RGB. I am going to do a few more tests to see the lowest point it still measures 255. 

 

HOWEVER!

 

There may still be cases in which it is better to shoot with 0.75RGB and clip your highlights in order to expose the subject better (better shadow detail). Not that I would suggest taking a photo like this, but an example may be taking a photo of a person with a light source behind them. 

[KnightsFan]

Are you doing photos or video?

 

It seems to me that the RGB sliders essentially do the same thing as an ISO adjustment, but after the actual ISO adjustment clips something. I've seen a lot of claims that the NX1 doesn't use analog gain ("ISO invariant"). So ISO adjustments only adjust digital gain--essentially a multiplier, probably in linear space.

If that is the case, then, assuming we avoid clipping, the RGB sliders have no effect other than tricking the camera into thinking it's at a different ISO, which could change the automatic noise reduction and sharpness. That was sort of what my conclusion was, way back last time I posted here. Basically, I can shoot at ISO 400 with the boost, and get an ISO 800 exposure without such aggressive noise reduction.

IF my analysis is correct, then lowering your RGB to 0.75 will both lower your highlights and shadows, pre-gamma curve. So you gain some highlight DR, and lose some shadow detail--basically, the same thing as lowering your ISO. We can test this by measuring like you did in Premiere on their 0-255 scale, but instead we transform the image into linear space first. My prediction is that every pixel will drop in value by a factor of 0.75 at that point. Thus, the >255 values drop down into range, and values near 0 will have a lot of noise.

[mnewxcv]

2 minutes ago, KnightsFan said:

Are you doing photos or video?

 

It seems to me that the RGB sliders essentially do the same thing as an ISO adjustment, but after the actual ISO adjustment clips something. I've seen a lot of claims that the NX1 doesn't use analog gain ("ISO invariant"). So ISO adjustments only adjust digital gain--essentially a multiplier, probably in linear space.

If that is the case, then, assuming we avoid clipping, the RGB sliders have no effect other than tricking the camera into thinking it's at a different ISO, which could change the automatic noise reduction and sharpness. That was sort of what my conclusion was, way back last time I posted here. Basically, I can shoot at ISO 400 with the boost, and get an ISO 800 exposure without such aggressive noise reduction.

IF my analysis is correct, then lowering your RGB to 0.75 will both lower your highlights and shadows, pre-gamma curve. So you gain some highlight DR, and lose some shadow detail--basically, the same thing as lowering your ISO. We can test this by measuring like you did in Premiere on their 0-255 scale, but instead we transform the image into linear space first. My prediction is that every pixel will drop in value by a factor of 0.75 at that point. Thus, the >255 values drop down into range, and values near 0 will have a lot of noise.

Please feel free to do that test, as I don't know how you would make it into linear space. All of what you say though makes sense. The curve doesn't change, but the value for each point on the curve shifts. 

 

Also, 0.85 will produce something registering at 255 on a 0-255 scale, 80 and lower will not achieve this (only 252-253 max). My test was for overexposure to begin with, so I will have to retest with normal exposure to see what else is happening, but with 1 stop of overexposure, my tests with values of 1.00 and lower register the light source in the scene as well as the white on my color checker both at the same value (255). When shooting at 1.99, the light source is 255, while the white on my color checker is 254, so a slight difference in highlight range even at 1 stop over. 

 

Shooting video only BTW.

[KnightsFan]

So I did some tests and the results did not line up exactly with my prediction, however, I think there is a reasonable explanation. All were shot in normal DR using Custom 1 profile. All clips are 1/50 shutter, f6.3, and ISO 800, except the last one. I chose 0.5 for the RGB boost because halving is really easy to evaluate.

RGB1.0toLinearMult0.5: Shot with sliders set to 1.0. It was transformed from Rec.709 to Linear space, and then multiplied by 0.5.

RGB0.5toLinear: Shot at RGB 0.5 and then converted to linear. I expected this to be exactly 1/2 the value at each point. Obviously, this is not the case, most noticeably in the highlights. One possible explanation is that, like most cameras, the NX1 doesn't shoot a strictly Rec.709 image, but implements a soft highlight rolloff to make the image more pleasing. Unfortunately, I don't know any way to eliminate this source of error.

RGB0.5DoubleISOtoLinearMult0.5: This is the interesting result. I shot at ISO 1600 to compensate for the RGB drop to 0.5. I converted from Rec.709 to Linear, and then multiplied by 0.5 the same way as the first shot. Notice how it is almost identical to the first shot, except that it clips sooner. This is what I'd expect from an ISO invariant camera: higher ISO simply reduces highlight headroom without providing any exposure benefit (except that digital gain is done pre-compression which is a HUGE benefit for an 8 bit compressed image). As a side note, the reason we don't have the massive fluctuation in highlight shown in test 2 is that the recorded image is essentially the same exposure (half RGB, double ISO). I think this is more evidence that the RGB boost occurs before the gamma curve, which is where the rolloff would be implemented.

This correlates with my findings many months ago, which is that providing an RGB boost closer to 2.0 makes a DR benefit, simply because the camera is tricked into using a lower ISO and thus keeps that headroom, at the expense of shadow detail. What I have not tested is how much the shadow quality differs, since that's pretty subjective and I don't know how to test it objectively. By eye, I don't see any shadow detail loss, but again, that's really subjective and anecdotal. It's worth mentioning that my tests from way back when were in Gamma DR. It may be possible to extend DR in other modes using an RGB <1.0 since they clip at a different point. But if I was right earlier about Gamma C clipping before 0.7, then you may as well just shoot in Gamma DR if you want to maximize DR.

 

[mnewxcv]

Using the histogram in premiere shows more info in the highlights when boosting and more info in the shadows when going 0.85 (what I consider the lowest without going below 100IRE). I also looked at the different gammas and see gamma C carries a lot if information in the mid range. I've never shot anything in GammaC but  may have to try and see how it comes out. \

I'm thinking when in a situation where you cannot avoid highlights being blown out, it may be worth going to 0.85 and getting more shadow detail since the highlights are lost anyway. This doesn't apply to gradual highlights like skies, but rather like an indoor shot with a blown out window where it wouldn't make sense to expose for highlights. 

 

I think in this case the color checker is a limited resource. I will fill in as the subject for the next tests to see how real world situations are affected. Good stuff so far.

[KnightsFan]

8 minutes ago, mnewxcv said:

Using the histogram in premiere shows more info in the highlights when boosting and more info in the shadows when going 0.85

How are you judging highlight and shadow info from the histogram?

 

[mnewxcv]

18 minutes ago, KnightsFan said:

How are you judging highlight and shadow info from the histogram?

 

based on how much info there is (intensity on histogram) and the curve from mids to high or low with saturation turned to 0. Shown below is 1.00 vs 1.99. Look at the curve blending mids with highs and lows. A question in my mind right now is is there any benefit to shooting 0.85, exposed for highlights so theyre not blown out, and lifting in post to reveal hidden shadow info, or is there any benefit to shooting 1.99, exposing for shadows, and dropping in post to reveal highlight info. Perhaps it's just a wash. 

[KnightsFan]

Does one image have some motion blur? The word x-rite is much sharper in the bottom image. It looks to me as though the lighting changes a little bit, too. The balance of brightness on the bottom right frame of the color checker vs. the center hinge changes considerably between the images.

But it seems clear from the waveform that patches of every brightness are reduced in value by what appears to be a common factor. The apparent bump in shadows on the histogram I think is due to the lower bound changing from -.081 to -.131 (at least, that's what I assume those numbers are)

[mnewxcv]

2 minutes ago, KnightsFan said:

Does one image have some motion blur? The word x-rite is much sharper in the bottom image. It looks to me as though the lighting changes a little bit, too. The balance of brightness on the bottom right frame of the color checker vs. the center hinge changes considerably between the images.

But it seems clear from the waveform that patches of every brightness are reduced in value by what appears to be a common factor. The apparent bump in shadows on the histogram I think is due to the lower bound changing from -.081 to -.131 (at least, that's what I assume those numbers are)

not really sure. No motion blur but perhaps something moved a little (I rest my grey card on the color checker between shots to set exposure). It may be 1/3 of a stop difference though looking at it. In terms of the histrogram though, I'm more referring to the fact that the top highlight roll off is basically a round curve, while the second one (1.99) starts dropping and halfway juts out and is more of a line than a curve.

[KnightsFan]

13 minutes ago, mnewxcv said:

not really sure. No motion blur but perhaps something moved a little (I rest my grey card on the color checker between shots to set exposure). It may be 1/3 of a stop difference though looking at it. In terms of the histrogram though, I'm more referring to the fact that the top highlight roll off is basically a round curve, while the second one (1.99) starts dropping and halfway juts out and is more of a line than a curve.

Ah, I see what you mean now. But could that be from changing light? I think that spike at the top end of the histogram is the color patch, and everything above that is the reflective bits of the the color chart borders. Looking at the waveform, on the top image there is nothing above the rightmost color patch, whereas on the second image there is some vague noise much higher on the waveform than the rightmost patch.

[mnewxcv]

19 minutes ago, KnightsFan said:

Ah, I see what you mean now. But could that be from changing light? I think that spike at the top end of the histogram is the color patch, and everything above that is the reflective bits of the the color chart borders. Looking at the waveform, on the top image there is nothing above the rightmost color patch, whereas on the second image there is some vague noise much higher on the waveform than the rightmost patch.

I would say that is evidence of more detail in the highlights. The white on the color checker is supposed to be the brightest thing in your shot, but when using 1.99 it seems to pick up possible reflections from the light on the plastic casing of the color checker which since they are more reflective than the white show up. I believe this information is normally lost in the curve, but consists of especially bright highlights. Here is another example (this time I got the exposure right on both). You can clearly see the histogram extends further into the highlights and with a more gradual curve (almost linear at the end) and on the waveform you can see on the first clip IRE reaches maybe 102/103, while on the second clip it is probably closer to 107-108 where the light bulb is.You can also see the opposite happening in the shadows, especially on the histogram. The middle of the shot is very similar, but the ends show the biggest difference. So it shows that that info up above 103 IRE just was not present in the other footage. It is new information in the video. 

[KnightsFan]

Very interesting! It looks a lot like the results I got with 1.00 vs 1.99. How did you match exposure? Did you open the aperture a stop on the 1.00 version?

[mnewxcv]

2 minutes ago, KnightsFan said:

Very interesting! It looks a lot like the results I got with 1.00 vs 1.99. How did you match exposure? Did you open the aperture a stop on the 1.00 version?

Yes, I changed aperture 3 1/3 increments (so 1 stop), both shots were ISO100, 1/50.

[KnightsFan]

1 minute ago, mnewxcv said:

Yes, I changed aperture 3 1/3 increments (so 1 stop), both shots were ISO100, 1/50.

Nice, so ISO is consistent so any improvements in DR are from the boost itself. I guessed aperture because I can see some difference in sharpness, especially in the light in the background.

However, one thing I notice now is that there is a bright column in the background on the right that is only visible in one shot, and in the same way there is more pure white in the lamp. Since a Histogram totals a constant value, then adding white will necessarily change all the other bars as well. So I'm focusing more on the waveform.

It certainly appears that the lamp's patch raises a little bit farther past 100, which is interesting. Also it's got those noisy tendrils shooting way up to 110 almost. I wonder if it's possible that since the lamp is in sharper focus, the NX1's sharpening kicks in and adds some more local contrast to the lamp, sending a few pixels higher into superwhite? I mean it's not like the camera is actually resolving detail brighter than the center of the lamp. Though of course that doesn't explain the patch overall being taller, just the noise up to 110. Out of curiosity, are you using 0-255 of 16-235 here?

[mnewxcv]

23 minutes ago, KnightsFan said:

Nice, so ISO is consistent so any improvements in DR are from the boost itself. I guessed aperture because I can see some difference in sharpness, especially in the light in the background.

However, one thing I notice now is that there is a bright column in the background on the right that is only visible in one shot, and in the same way there is more pure white in the lamp. Since a Histogram totals a constant value, then adding white will necessarily change all the other bars as well. So I'm focusing more on the waveform.

It certainly appears that the lamp's patch raises a little bit farther past 100, which is interesting. Also it's got those noisy tendrils shooting way up to 110 almost. I wonder if it's possible that since the lamp is in sharper focus, the NX1's sharpening kicks in and adds some more local contrast to the lamp, sending a few pixels higher into superwhite? I mean it's not like the camera is actually resolving detail brighter than the center of the lamp. Though of course that doesn't explain the patch overall being taller, just the noise up to 110. Out of curiosity, are you using 0-255 of 16-235 here?

16-235. I just tried 0-255 as well in some random tests as well as messing with master black. The rabbit hole is getting deeper. You know how we said if you go negative on the RGB values it starts to clip highlights? Well that's true and it isn't. For instance, if you shoot at .50RGB, highlights clip at about 92 IRE. But if you expose to have your highlights under 92IRE, they don't clip. So my idea of shooting <1.00RGB values and raising in post could possibly be done. It seems the lower you go the less DR you get at a certain point, but I think .50 is still roughly the same number of stops. Also I did a shot of the bulb at 0.05RGB and the bulb had nice exposure, though everything else in the shot was black at base ISO. As we know boosting ISO would be useless since at that low of a rgb value the IRE cap would be very low as well. Taking the same shot with RGB boost at base ISO, image is still under exposed, but it can come up somewhat (there are limits obviously, I was shooting 1/4000 and base ISO to expose bulb, the rest of the footage was uselss, but there was some info there). However, the difference between -15 and +15 master pedestal was obvious when doing this extreme test. Lifting the shadows on the -15 MP shot, colors on the color checker were distorted and wrong. While MP +15 lacked contrast and DR, the colors were present and much more accurate. 

[mnewxcv]

I am going to rest my brain from this for a bit, but I think so far it makes sense to use 1.99 whenever lighting is good to excellent. It doesn't love the shadows in low light footage, in which case using 1.00 works just fine. Perhaps .85 is good for low light detail without many bright lights, though the jury is still out.

[KnightsFan]

11 minutes ago, mnewxcv said:

Well that's true and it isn't. For instance, if you shoot at .50RGB, highlights clip at about 92 IRE. But if you expose to have your highlights under 92IRE, they don't clip. So my idea of shooting <1.00RGB values and raising in post could possibly be done.

I thought I addressed that with my theory that the RGB slider operates on a linear image before the gamma curve is applied, which explains why it clips almost immediately in Gamma DR but not in Gamma Normal or C. But I expect that the RGB slider multiplies shadows as well, so not sure how much usable DR you would gain as it basically just lowers exposure overall. Certainly something to test if you want to follow the rabbit hole!

I think the only way we could meaningfully get a DR increase is if we could apply the RGB multiplier to the top half of values, but not the bottom. Most of what I've seen so far, both with my tests and yours, is really minimal, if at all present.

13 minutes ago, mnewxcv said:

However, the difference between -15 and +15 master pedestal was obvious when doing this extreme test. Lifting the shadows on the -15 MP shot, colors on the color checker were distorted and wrong. While MP +15 lacked contrast and DR, the colors were present and much more accurate.

As far as I can tell, MP doesn't really add DR with a positive value, it just raises the blacks up a bit. At most you gain back a little bit of noise. With negative values yeah, it seems to lose shadow detail. So I usually leave it at 0.

2 minutes ago, mnewxcv said:

I am going to rest my brain from this for a bit, but I think so far it makes sense to use 1.99 whenever lighting is good to excellent. It doesn't love the shadows in low light footage, in which case using 1.00 works just fine. Perhaps .85 is good for low light detail without many bright lights, though the jury is still out. 

Haha yeah, I just spent a good part of my day on this, and the funny part is I doubt I'll ever even use my NX1 for a project again! For my next project we're using the director's XT3. I did shoot a big project on my NX1 using RGB boost last year, which I might detail here once it's released. For me it was all about finding a general setting that worked for everything, since I wasn't going to fiddle with settings on set even if I knew wanted. So I used RGB values near 1.99 in Gamma DR, built some LUTs that worked with those settings, and left it that way. Worked pretty well in hindsight.