NX1 Extended Dynamic Range? New Settings.

[lucabutera]

1 hour ago, ReinisK said:

I've done some tests today. The camera was set to UHD, 25p, Gamma DR, -5 contrast, -7 sharpness, 16-235, sRGB. I compared different ISOs with and without RGB set to max (1.99). 

 

So what I found out:

All RGB channels at 1.99 increase exposure by 1 stop (or very, very close to it);

Therefore ISO 1600 with RGB at 1.99, is the same exposure as ISO 3200 with RGB at 1;

It positively affects the inbuilt noise reduction. Seems like the NR really is triggered by ISO value and not the actual sensor gain;

It doesn't have any visible gains in highlights.

 

Here's the test. The password is - rgb .

Here you can download the edited test file, if you would like to take a closer look - https://failiem.lv/u/xvgh44ks

 

Can you repeat the test with exposure compensation?

[Kisaha]

My impression is that NR, aggressively, kicks in above 1600ISO. To be honest, that is the most I go, but 3200 should be possible for such a modern and capable camera.

It seems here that this work-around increances light intensity(?) somehow (something to do with the very unique and capable BSI sensor I guess), so you can gain some light through this.

I am not sure about the codec, and how it can handle the settings, it may increase color noise and push the codec to its limits, but it can help with the silly processing above 1600ISO.

I am 100% positive that 2 Samsung engineers (1 hardware/1 software) could gain easily a stop or two from this sensor!

[mnewxcv]

13 minutes ago, Kisaha said:

My impression is that NR, aggressively, kicks in above 1600ISO. To be honest, that is the most I go, but 3200 should be possible for such a modern and capable camera.

It seems here that this work-around increances light intensity(?) somehow (something to do with the very unique and capable BSI sensor I guess), so you can gain some light through this.

I am not sure about the codec, and how it can handle the settings, it may increase color noise and push the codec to its limits, but it can help with the silly processing above 1600ISO.

I am 100% positive that 2 Samsung engineers (1 hardware/1 software) could gain easily a stop or two from this sensor!

if you look at the last 2 photos in my last post, the picture wizard with no changes other than RGB boost, sharpness contrast and sat all the same, iso 1600 with RGB boost is noticeably noisier than picture wizard off at 3200iso. 

[keessie65]

Impressive, I have to do a test soon. I normally colour the film in post with Adobe. I use mostly a lut for the extra bite. Wondering how that will look. 

My DR settings I use normally are sat -2, sharpness -10, contrast -3, master black +10, lum 16-235

[MKSN]

4 hours ago, mnewxcv said:

if you look at the last 2 photos in my last post, the picture wizard with no changes other than RGB boost, sharpness contrast and sat all the same, iso 1600 with RGB boost is noticeably noisier than picture wizard off at 3200iso. 

What if rgb boost is actually boosting each channels iso somehow? ? Should it be tested also with negative values and try to shoot 6400iso or something like that... 

Example Red x1.99 is actually iso3200 x 1.99 = 6368iso for red channel...

I try to do my own tests also tomorrow. This is interesting.

 

[mnewxcv]

also seeing a green tint even with RGB set to 1.99, 1.90, and 1.99 respectively. Will need to adjust hue slightly it seems. It is odd, using RGB boost and iso100 was cleaner than no rgb boost and iso 200, but rgb boost with 3200iso was worse than no rgb boost 6400. 

[keessie65]

the NX1 reimagined. Now "NX1s"

[Pavel Mašek]

Thank you all especially @Happy Daze  for investigation. I have concluded for myself that boosting RGB values is similar to highering ISO values but it makes sense that this way can help reduce noise reduction that is probably higher on ISO 1600 and above ( @ReinisK  video).

But I have test it and found out that details in shadows are higher with lower RGB values but downside is the highlights are clipped because of longer shutter speed (unfortunately it does not work as digital ND filter). So maybe RGB x1.0 is still best compromise. But I still do not understand it... 

NOTE: Images grabbed from video directly in NX1

0-255, ISO100, RGB x1.99 on the left; RGB x0.50 on the right (just longer exposure to compensate loss of brightness).

You can see more details on the ground and clipped highlights on the sky on the right (OK, I know it seems to be clipped on the left, but there completely no details on the right - just solid grey...)

 

0-255, ISO100, RGB x1.00 on the left; RGB x0.50 on the right (look on the bush and grass - they are more detailed). On the otherside white wall looks worse to me on the right (RGB x0.50) image. But maybe it is not so bad... I do not know...

So maybe lower RGB values means more details in dark and complex areas but as downside - DR is is lower.

 

[ReinisK]

Wow, the detail gain is really noticeable. Was it 4k or 1080p?

[Pavel Mašek]

28 minutes ago, ReinisK said:

Wow, the detail gain is really noticeable. Was it 4k or 1080p?

4K30p, 160Mbps... and ignore histogram, it seems they are completely off (at least in one case)

Maybe is noise reduction mapped on RGB gain?

I am curious if it is somehow possible to maintain normal DR with benefit of more detailed image of lower RGB... but I guess it is not possible. It can useful in some situations where wide DR is not needed and there is enough light

I will check if change to 16-235 or GammaDR could help but I doubt about it. Maybe RGB x0.75 could be compromise.

[Mokara]

On ‎2018‎-‎04‎-‎03 at 12:26 PM, MKSN said:

What if rgb boost is actually boosting each channels iso somehow? ? Should it be tested also with negative values and try to shoot 6400iso or something like that... 

Example Red x1.99 is actually iso3200 x 1.99 = 6368iso for red channel...

I try to do my own tests also tomorrow. This is interesting.

 

I would guess that the sensor is returning something like 14 bits of data per pixel, but the processor is only actually using some lesser amount, such as 10 bits. When you increase or decrease the color channels you adjust exactly which 10 bits of the 14 gathered are used. So by moving everything up two notches and then reducing exposure by a corresponding amount, you are effectively increasing ISO without increasing gain. If NR is kicking in at a particular gain setting the net outcome is that for all practical purposes you could get an extra two stops of exposure before that happens using the OP's method.

I would guess that you can do the opposite as well, essentially adding an internal ND function to the camera if you had need for that.

The downside however is that data collected at the top and bottom of those 14 bits is likely to be less accurate, and that may cause other unanticipated issues, such as WB or color cast problems (since in camera correction of those properties requires some headroom in the color channels).

[mnewxcv]

I found the color to have a green tint or less red. It is hard to describe it was very subtle, but it turned brownish maroon colors into regular dark brown. It can be seen on dark cherry colored furniture. As for the histogram, it must get its information pre boost as it does not change with or without boost. 

1 hour ago, Mokara said:

I would guess that the sensor is returning something like 14 bits of data per pixel, but the processor is only actually using some lesser amount, such as 10 bits. When you increase or decrease the color channels you adjust exactly which 10 bits of the 14 gathered are used. So by moving everything up two notches and then reducing exposure by a corresponding amount, you are effectively increasing ISO without increasing gain. If NR is kicking in at a particular gain setting the net outcome is that for all practical purposes you could get an extra two stops of exposure before that happens using the OP's method.

I would guess that you can do the opposite as well, essentially adding an internal ND function to the camera if you had need for that.

The downside however is that data collected at the top and bottom of those 14 bits is likely to be less accurate, and that may cause other unanticipated issues, such as WB or color cast problems (since in camera correction of those properties requires some headroom in the color channels).

Holy crap. Can we drop the rgb boost to keep 1/60 shutter speed without an ND? Now THAT is worth exploring! 

[IKSLIM]

13 hours ago, mnewxcv said:

Holy crap. Can we drop the rgb boost to keep 1/60 shutter speed without an ND? Now THAT is worth exploring! 

I gues no. At least i failed. The shutter angle remains the same, even in manual mode M only ExpComp affected. ISO already locked by minimum value in this case.

But negative vector trick definitely deserve to investigate...

[mnewxcv]

3 hours ago, IKSLIM said:

I gues no. At least i failed. The shutter angle remains the same, even in manual mode M only ExpComp affected. ISO already locked by minimum value in this case.

But negative vector trick definitely deserve to investigate...

I will see what I come up with today. I was hoping the rgb values had something to do with the back side illumination. 

[Happy Daze]

21 hours ago, Pavel Mašek said:

4K30p, 160Mbps... and ignore histogram, it seems they are completely off (at least in one case)

Maybe is noise reduction mapped on RGB gain?

I am curious if it is somehow possible to maintain normal DR with benefit of more detailed image of lower RGB... but I guess it is not possible. It can useful in some situations where wide DR is not needed and there is enough light

I will check if change to 16-235 or GammaDR could help but I doubt about it. Maybe RGB x0.75 could be compromise.

You are correct Pavel, as far as negative RGB is concerned highlight control is a problem. The histogram shows that the highlights are well within range but they are in reality clipping at a lower level than 255 more like 235/245 that's why everything above these levels is a flat grey.

[Pavel Mašek]

So I have tried 16-235 directly in camera.

Potrait style, RGB x1.00 (with some tweaks on each channel) on the right

Vivid style,  RGB x0.50 (with some tweaks on each channel) on the left

So it this case it works like ND filter (you can see different exposure times, aperture is same indeed) and I am quite surprised that DR is almost identical BUT it all depends on scene. So if there less highlights you can lower RGB values/set longer exposure more (I have achieved 1/500s -> 1/200s)

[mnewxcv]

4 hours ago, Happy Daze said:

You are correct Pavel, as far as negative RGB is concerned highlight control is a problem. The histogram shows that the highlights are well within range but they are in reality clipping at a lower level than 255 more like 235/245 that's why everything above these levels is a flat grey.

Are you sure the highlights are clipping? The histogram is pre boost. As long as you expose for the scene it should be OK. Also, if you have overexposure guide on aka zebras they are still working properly. 

 

Also, good to hear about the ability to use as an ND filter. That is actually an amazing possibility. Set different picture wizards for different ND levels! 

[KnightsFan]

@mnewxcv Yes, the highlights are simply clipping early.

Here's a screenshot in Resolve with the waveform for reference. This was with RGB sliders set to 0.5. You can see even without the waveform that the lightbulb does not reach full white.

For comparison, here is the same scene with the RGB sliders at 1.0.

[Kisaha]

Can anyone sum it up a bit?

We started from 1.99 and now we are to 1 and 0.5 and 0 and everything in between?!?

@BTM_pix did a very interesting tool for Panasonic cameras on another recent thread to search for optimal settings more systimatically and methodically. We may need something similar.

[mnewxcv]

18 minutes ago, Kisaha said:

Can anyone sum it up a bit?

We started from 1.99 and now we are to 1 and 0.5 and 0 and everything in between?!?

@BTM_pix did a very interesting tool for Panasonic cameras on another recent thread to search for optimal settings more systimatically and methodically. We may need something similar.

going to 1.99 does not improve exposure without adding noise. The noise added is very similar to going an ISO level higher. Also, I found it changes the color tint somewhat. 

 

We have been experimenting with going lower now, as it *may* act as a sort of digital ND filter, however, as shown above, highlights are getting clipped somewhat. I will go ahead and get some testing done in the next day or so (maybe tonight) at different levels and show what is happening with highlights. 

42 minutes ago, KnightsFan said:

@mnewxcv Yes, the highlights are simply clipping early.

Here's a screenshot in Resolve with the waveform for reference. This was with RGB sliders set to 0.5. You can see even without the waveform that the lightbulb does not reach full white.

For comparison, here is the same scene with the RGB sliders at 1.0.

did you adjust exposure to account for the lower brightness, as in decrease shutter, open aperture, or boost iso? Auto exposure does not work when adjusting the RGB scales, without compensation. So if you set auto exposure and turned down the rgb scales, it will be *under exposed*, just as if you boost the scales without compensation, the image will be *over exposed*.