Mini 3 and dual native iso

[lukaspeca]

Hi everyone,
Can someone tell me at what settings and modes does mini 3 use dual native ISO ? Is it used on both photo and video recording ? I've read (somewhere on FB mini 3 page) that dual native iso is only activated when using 3200 ISO. Is this true or it is applied on all settings?
Couldn't find much info online or in manual

 

[Derlisz]

Second native iso is at 500. Sadly you since you can only do a full stop values the lowest iso using second circuit is ISO 800.

 

[Ralph thompson]

Have either of you determined if there’s a noticeable benefit of dual native iso? I think it’s only available in night mode? Which I don’t use because I find dlog provides much better results in post. I have the BMPCC4K with dual native iso, I can’t really notice any benefit although it could be subtle and beneficial under certain conditions

 

[Derlisz]

Have either of you determined if there’s a noticeable benefit of dual native iso? I think it’s only available in night mode? Which I don’t use because I find dlog provides much better results in post. I have the BMPCC4K with dual native iso, I can’t really notice any benefit although it could be subtle and beneficial under certain conditions

Mini 3 has no night mode. You can set the iso to 400 and shoot some video (in 50/60 fps) and the. Switch to just over 500. There will be a noticeable difference in ammount of noise. If you leave noise reduction on where will be a noticeable change in retained detail.

 

[Ralph thompson]

OK thanks Derlisz

 

[Droning on and on...]

Second native iso is at 500. Sadly you since you can only do a full stop values the lowest iso using second circuit is ISO 800.

Yes, but for stills, you can get around this a bit by choosing shutter priority, then adjusting shutter speed to get the iso you want. When auto, ISO isn't constrained to full-stop values.

 

[Ralph thompson]

I don’t think dual native ISO functions in still mode, just video But you do seem to be able to take stills at a very slow shutter speed, quite remarkable stability in the gimbal.

 

[Droning on and on...]

I don’t think dual native ISO functions in still mode

Two questions: Why wouldn't it, and why do you think it doesn't?

 

[Ralph thompson]

I've been a photographer for many years, I use professional Nikon DSLR's. I've followed the development of digital photography since its inception and never heard of anyone implementing dual native ISO (more correctly dual gain) in still photography. Nikon D6 ISO goes to 102,000 and expandable to dizzy 3,280,000 and they've never mention dual native ISO or gain. Its not mentioned in the Spec. Not sure why, maybe the gain is adjusted progressively, no idea but I've only seen it used for video, notwithstanding multiple social media forums unsubstantiated claims about dual gain in still photography. Video technology is quite different to digital still photography. I've followed Blackmagic's development and I have used dual native ISO in the pocket cinema 4K camera for a few years. I'm not convinced it is quite as beneficial as some state. If all else is done correctly for the environment, then there is a small visible benefit, but usually other factors are determining your settings for video like required FPS, shutter speed and aperture then lastly you tweak ISO to get exposure right on.

 

[Droning on and on...]

I'm not a photographer, but an engineer. Someone needs to explain to me how the sensor behaves differently capturing a video frame, vs. a still of the exact same scene.

Otherwise, why wouldn't the same reasons for video (lower noise in low light) apply to a still?

 

[Ralph thompson]

The fact that the manufacturers don't use it as a selling feature is telling.
Dual gain requires there to be two connections from the sensor, ie additional analogue circuitry is added to accommodate greater amplification, a higher gain applied to the sensor signal. An added cost and complexity, especially in a small consumer camera like the DJI Hasselblad system.
Video is quite different than still, video is not multiple stills. There are lots of computation in video, including DLog capture thats not required in still capture. Even RAW capture in video is not multiple stills. Still capture works in RGB whereas video works in YCC (a detailed luminosity channel and two highly reduced colour channels) to minimize data size. Being technically oriented you will know that video capture takes clever advantage of the fact that our eyes are very sensitive to luminosity (gray scale) and quite insensitive to colour.
Camera manufacturers are very careful not to give away too much detail on how they handle the science but I believe for digital stills there are other ways to minimize digital noise and still maintain highlights and shadows throughout the ISO range.
In many respects a camera that shoots stills & video is like two different technology cameras packaged into one box. They share a lens, sensor, storage and display but thats about it.
You should be aware that "dual iso" is not a panercea, I haven't seen a DJI technical explanation but by pumping up the gain for high ISO you substantially clip highlights in order to get more shadow detail.In your setup you have to decide if highlights are important or shadows because if you move to the higher "native" gain (higher ISO) then you clip the capture and loose several stops of highlights.
I think the "dual native iso" in the M3 is used in their new night mode, the reason being that the higher native ISO gives a lot more bandwidth (space below middle gray) in the shadows
One fact to remember, whether it's day or night, whether you set high ISO or low ISO, absolutely nothing changes from a capture point of view at the sensor. The sensor captures X amount of light regardless of ISO setting. If you work in RAW, the same image captured at night with high or low ISO, roughly speaking can be corrected in post and look similar. You either amplify the analogue signal from the sensor or you amplify the signal digitally in post, pretty much the same thing.

 

[Ralph thompson]

One example that might help to understand the difference between the way the ISO technology works in stills and video is as follows.
In Montreal right now we have lots of snow on the ground and it's a nice sunny day. Generally speaking, if I take a still of a snowy ski slope, I'd minimize ISO (ISO 100 for my D850 or D6), set optimum aperture (f8) then adjust shutter speed to ensure snow texture is visible. I'd probably use a circular polarizing filter. If I were videoing the same scene with the BMPCC4K, I'd start by upping the ISO (sounds counter intuitive) to a point where I get a lot more stops available above middle gray (ISO 400 or higher) then set shutter speed probably to the 180 rule and adjust aperture to the optimum (f5.6). I'd then have to add ND16 or 32 filters to prevent over exposure. The approach for shooting at night without the bright textured snow would be quite different again.

 

[Droning on and on...]

The fact that the manufacturers don't use it as a selling feature is telling.
Dual gain requires there to be two connections from the sensor, ie additional analogue circuitry is added to accommodate greater amplification, a higher gain applied to the sensor signal. An added cost and complexity, especially in a small consumer camera like the DJI Hasselblad system.

This is seriously at odds with my understanding of design and operation of dual native sensors.

As for video, yes, it's different from stills downstream of the sensor. The sensor doesn't do any of the processing to compress the stream, creating I, P, and B frames from the raw, full resolution image coming from the sensor for each frame. All that happens in a dedicated video processor SOC.

Every single frame in a video is either a full resolution image (I-frame, identical to a raw still image), or difference data between the last frame, and again a subsequent full resolution raw still image, and then compressed using a lossy algorithm.

So my question stands: How is the operation of the SENSOR different capturing a still with say, 1/60 shutter, and iso 800, compared to the operation of the SENSOR in capturing the image data for a video frame at 1/60 sec iso 800?

Finally, video is, in fact, simply a series of still images played in sequence fast enough that it looks continuous, but this has nothing to do with video images being different than still images. Take a look at an old still image for use in a slide projector, and a strip of movie film.

What's changed is how we manage the information in those images.

 

[Droning on and on...]

@Ralph thompson, here's a decent explanation about dual native ISO w.r.t. still photography. The reason for my original question is demonstrated starting around 9:00.

I'd be mighty pissed if my camera had dual iso capability, and didn't use it for still images, instead producing noisier stills at low light than is necessary.

I do not know, but I'd be surprised if the dual iso capability in the M3P for stills is disabled, given the better results in low light.

 

[Ralph thompson]

Dual native ISO is a misnomer. To implemented dual gain in video cameras (so called dual native ISO) the sensor output is redirected to a higher gain amplifier in the analogue circuitry. I'm sure they do other clever things in the processing but thats irrelevant in this discussion. The sensor output signal is identical regardless of ISO setting. Dual gain/native ISO is not implemented in most digital still cameras as far as I know because simply amplifying the RGB signal gives you nothing. Increasing the signal in analogue or digital is effectively the same thing. If you have a good digital still camera, shoot a RAW photograph in a normally lit room with all manual settings, 1/60th, f4.5 at say ISO 100 (this will be very dark) then take the same photograph at say ISO 3200 (should look like a reasonable exposure). Don't change any other settings. Take the two RAW files into Lightroom and increase the exposure setting on the ISO 100 photograph. This is almost the max. increase possible in exposure in Lightroom. Attached are exactly this example I just took in my studio with a Nikon D810. 1/60th sec, f4.5. No noise reduction in lightroom. First image shot at ISO100 and the 2nd shoot at ISO 3200. These are small jpgs, I can't attach the NEF's but on my computer (iMac) the NEF (RAW) files are very similar even at 400% zoom, I cannot see much difference except a slight colour shift in the pumped up image. WB was very slightly different.
BTW the D810 is quite an old camera, newer technology sensors have lower noise but it doesn't change the argument. Maybe a simpler way to look at is a DSL/Mirrorless still camera capturing RAW keeps all the data off the sensor within the limitations of its bit depth. Most video cameras only use part of the data off the sensor and use clever tricks like DLog, dual native gain, compression etc to allow for constant streaming.

 

[RogerDH]

I'm not a photographer, but an engineer. Someone needs to explain to me how the sensor behaves differently capturing a video frame, vs. a still of the exact same scene.

Otherwise, why wouldn't the same reasons for video (lower noise in low light) apply to a still?

If the sensor (or DSP) automatically does the ISO switch, yeah, I'd expect videos and photos to be the same, but maybe the controlling software does it, but only in some modes?

 

[Droning on and on...]

If the sensor (or DSP) automatically does the ISO switch, yeah, I'd expect videos and photos to be the same, but maybe the controlling software does it, but only in some modes?

Don't know. Wasn't the issue. However, for what reason would they disable it for stills, resulting in much noisier low light images? That would be monumentally foolish, when there's no benefit.

The issue is, does "dual native iso" feature operate when taking stills as well as video? It has been said here that no, it doesn't because video is different in terms of capture, which is false. At the SENSOR, video is nothing more than the sensor being read over and over again at the frame rate.

With iso-priority now available in Pro mode, we can test it. Anyone know what the ISO crossover point is?

@Ralph thompson, didn't read past your first line, as I don't have the patience for a semantic debate. Also, you're simply incorrect about the hardware/software in the image pipeline, a subject that is in my baliwick.

 

[Ralph thompson]

I’ll checkout after this, but as said above the simple way to understand the technical approach is that a DSLR/Mirrorless still camera processing RAW captures everything on the sensor & outputs it as a raw file. There is no more detail on the sensor, you have all the data in yr raw file to the limit of the bit depth. Changing ISO changes absolutely nothing on the sensor, if u shoot raw then its just a tag in the metadata. Sensors have one ISO. Most video cameras have to use tricks to output smooth streaming video at a reasonable size. They use YCC, sub sampling, reduced bit depth & compression then recover detail with log curves, dual native gain etc to overcome the shortcomings. The video is a reduced version of what came off the sensor with some clever tricks to make it look good. That is the answer to yr original question.

 

[Droning on and on...]

the sensor output is redirected to a higher gain amplifier in the analogue circuitry. I'm sure they do other clever things in the processing but thats irrelevant in this discussion. The sensor output signal is identical regardless of ISO setting.

Okay, I went back and read your post. There are a lot of things incorrect that are clearly so because you don't know how these sensors work, lack an understanding of noise, both analog and digital (i.e. quantization noise), and how this is managed in the circuit.

Without explanation (I'll explainin a follow-up), increasing the analog amplification for low level signals (i.e. low light) decreases the quantization noise. When you see noise in dark areas of an image at high iso, this is quantization noise as a result of the ADC.

Your assertions in the above post indicate to me you are not very familiar with these concepts, and how the play together in a digital image capture system.

 

[Droning on and on...]

Consider an analog voltage that can take on any value between 0 an 1 volts. In a dark image, the output signal among the pixels in that scene vary between 0.05V and 0.17V.

This voltage enters a 2-bit ADC (not actual – few bits to simplify the explanation), that quantizes the analog value into 4 levels, 0-0.25V being 00, 0.25-0.5V 01, etc.

When the pixels from the image are quantified, they all convert to 00 – black – because they are all below 0.25V. Even though there is all sorts of variation – detail – in the image.

Now, suppose we amplify that analog signal by a factor of 5 before we digitize it. Now the input values to the ADC range from 0-0.85V. After passing through the ADC, instead of all black pixels, with no detail, we will have pixels at 4 different levels, showing detail.

The digital "amplifier" is now simple mathematical scaling. To restore the right level in the digital domain, we simply divide all the digital values by 5.

But wait! This will simply result in all zeros again. This is where we get clever and apply a log transform to the values, disproportionately devoting more digital levels to smaller values than larger ones and voila! We've preserved detail in low light (low signal voltage) areas, while tossing detail for bright pixels, which generally isn't an issue for low-light scenes.

The important part you're missing is the ADC and the fact that it is a lossy conversion. With signal bias and math, you can adjust the characteristics of what sensor signal is preserved, and what is discarded.

You've got to stop thinking of a digital sensor as behaving like film. Film is analog, and therefore continuous. Digital values are discrete, and therefore lose information in the conversion from continuous analog signals. Smart design can give control over what information is discarded/ignored. In the case of low intensity images, we set up the sensor to ignore (actually, overexpose) bright pixels, that aren't really present much in a dark image.

Bottom line: Contrast for bright pixels is sacrificed for detail in the image, a worthy trade-off in a low-light image.