SG9665GC High Contrast/Dynamic Range Flaw and other Discouragements

[Dashmellow]

and sensor, the AE table is something all cameras use though and the science behind it is reasonably universal, still wouldn't mind having a look at their AE table if anyone has had a poke around the firmware

Yes, the sensor too! My hunch is that the sensor may turn out to be a bigger part of this problem than the DSP but of course, it must come down to how those two components interact.

 

[jokiin]

Yes, the sensor too! My hunch is that the sensor may turn out to be a bigger part of this problem than the DSP but of course, it must come down to how those two components interact.

there is something in that, the better the sensor performs in low light the more care you need with handling the bright light

 

[arcticfire]

Been playing about further and come across something that could be interesting. It looks like the dashcam is allowing the IRE level to go to 110 rather than the standard 100 and this could be what's contributing all the clipping at the upper end of the tonal curve as it's allowing too much to go over IRE100 and could be confusing/messing with the AE table.

Is it possible that if the camera thinks that IRE110 is the point of overexposure then it will also think that IRE100 is not overexposed (where it actually is) and therefore this messes with the rest of the exposure?

Check out these screenshots with the luma waveform, you can see the areas which exceed IRE 100 (the horizontal position in the waveform directly relates to the horizontal position in the footage) and then at the bottom I've posted up one of the 0826 and you can see that it only allows it to go to IRE 100.



The white pickip is overexposed





Building on right has overexposed areas as does a few other sections



Building on right cropped out




Car roof



Lots of values above IRE100




Here's the Mini 0826, notice how it only goes to IRE100.

 

[Dashmellow]

and sensor, the AE table is something all cameras use though and the science behind it is reasonably universal, still wouldn't mind having a look at their AE table if anyone has had a poke around the firmware

Why not talk with Rayman.Chan about the AE tables on the NanoQ/903 where they seem to have mostly avoided this issue altogether on the same platform? (from my current understanding, anyway). In fact, there are a number of things the NanoQ/903 accomplishes with the NT96655 DSP and the Sony IMX322 sensor that are related to the "other discouragements" I've held off bringing up in this thread for the time being so we could focus solely on the dynamic range issue, which I felt was the primary concern.

 

[jokiin]

Why not talk with Rayman.Chan about the AE tables on the NanoQ/903 where they seem to have mostly avoided this issue altogether on the same platform? (from my current understanding, anyway). In fact, there are a number of things the NanoQ/903 accomplishes with the NT96655 DSP and the Sony IMX322 sensor that are related to the "other discouragements" I've held off bringing up in this thread for the time being so we could focus solely on the dynamic range issue, which I felt was the primary concern.

I've seen examples of the same problem on their camera

 

[Dashmellow]

I've seen examples of the same problem on their camera

So far the samples on I've seen have looked much better than the GC but it certainly doesn't surprise me to hear you say this. I've suspected this was a platform wide problem for some time, one way or the other.

 

[Dashmellow]

Been playing about further and come across something that could be interesting. It looks like the dashcam is allowing the IRE level to go to 110 rather than the standard 100 and this could be what's contributing all the clipping at the upper end of the tonal curve as it's allowing too much to go over IRE100 and could be confusing/messing with the AE table.

Is it possible that if the camera thinks that IRE110 is the point of overexposure then it will also think that IRE100 is not overexposed (where it actually is) and therefore this messes with the rest of the exposure?

Check out these screenshots with the luma waveform, you can see the areas which exceed IRE 100 (the horizontal position in the waveform directly relates to the horizontal position in the footage) and then at the bottom I've posted up one of the 0826 and you can see that it only allows it to go to IRE 100.



The white pickip is overexposed
View attachment 22558




Building on right has overexposed areas as does a few other sections
View attachment 22559


Building on right cropped out
View attachment 22560



Car roof
View attachment 22562


Lots of values above IRE100
View attachment 22563



Here's the Mini 0826, notice how it only goes to IRE100.
View attachment 22564

Not quite sure I understand this. IRE is an analogue voltage measurement that is different to how these levels are measured in digital video.

 

[jokiin]

there are strengths and weaknesses to every platform/hardware combination, none are best in every situation, don't think there ever will be, we will improve it though regardless

 

[2000rpm]

So far the samples on I've seen have looked much better than the GC but it certainly doesn't surprise me to hear you say this. I've suspected this was a platform wide problem for some time, one way or the other.

'Over-exposure' (that's what I had labelled it in my head) in certain very bright conditions was something I assumed was a an unresolved issue with all IMX122/IMX322 cams and a trade-off for better night performance.
In my usage, possibly due to only moderate sun strength in the UK, the 'over exposure' hasn't caused an issue severe enough that it would render any video evidence useless, and most of the currently available IMX122/322 cams aren't great at picking up details anyway due to them all having a wide field of view which works against detail gathering.

 

[Street Guardian USA]

'Over-exposure' (that's what I had labelled it in my head) in certain very bright conditions was something I assumed was a an unresolved issue with all IMX122/IMX322 cams and a trade-off for better night performance.
In my usage, possibly due to only moderate sun strength in the UK, the 'over exposure' hasn't caused an issue severe enough that it would render any video evidence useless, and most of the currently available IMX122/322 cams aren't great at picking up details anyway due to them all having a wide field of view which works against detail gathering.

SGZC12SG has a super wide angle lens but we used a medium wide angle lens on the SG9665GC to keep the detail higher.

 

[jokiin]

'Over-exposure' (that's what I had labelled it in my head) in certain very bright conditions was something I assumed was a an unresolved issue with all IMX122/IMX322 cams and a trade-off for better night performance.

that true to a point, we have some things going on that are blowing it out more than it reasonably should though

 

[arcticfire]

Not quite sure I understand this. IRE is an analogue voltage measurement that is different to how these levels are measured in digital video.

I don't know enough about it to be honest but I think the software is doing some form of conversion so as to be able to show it as the IRE equivalent. As it shows the mini 0826 to be correctly cutting off at 100 I'm guessing it's accurate enough.

 

[Dashmellow]

I don't know enough about it to be honest but I think the software is doing some form of conversion so as to be able to show it as the IRE equivalent. As it shows the mini 0826 to be correctly cutting off at 100 I'm guessing it's accurate enough.

I know little about this as well. I remember first hearing the term IRE from a technician when I had to have a Sony Trinitron CRT TV calibrated years ago.

A little internet searching offers some enlightenment.

Analogue Video
In analogue video, all levels are represented by voltages, and measured in IRE or “Institute of Radio Engineers” units. In North America, the range of values for video luma levels is from 7.5 IRE (darkest black reference level) to 100 IRE (brightest white reference level). The rest of the world uses a 0 IRE to 100 IRE range.

Digital Video
In digital video, all levels are represented by the binary language of computers, but for our ease, we shall use decimal equivalents. Instead of the continuously variable voltage of analogue video, digital video uses a set of discrete values to represent the video levels. Each allowable video level in this system can be given a code number. In an 8bit system there are 256 (2 to the power of 8) allowable codes, numbered from 0 through to 255.

Interesting subject worth further exploration. I'm not clear how your method works or applies here but it does show a difference.

 

[2000rpm]

SGZC12SG has a super wide angle lens but we used a medium wide angle lens on the SG9665GC to keep the detail higher.

The SG-GC is still fairly wide, though.
On my field of view width scale the SGGC scores 9/10. The SGSG, SGRC and Panorama score 10/10 for fov but trade off some of the detail gathering. But as I've mentioned a few times: the Cobra 840 is the ultimate detail gatherer, although at the expense of unremarkable fov and night performance.
Three screenshots of a van which 'cut me up' last week:
1. SGRC (fov = 10/10)
2. Mio 538 (fov = 7.5/10)
3. Cobra 840 (fov = 5/10)

-

 

[jokiin]

I know little about this as well..

join the club

Interesting subject worth further exploration. I'm not clear how your method works or applies here but it does show a difference.

that might be more relevant perhaps if it were two different cameras based on the same hardware though, different hardware platforms have different characteristics already so a different end result is not unexpected I would think

 

[jokiin]

that's a very narrow field of view on the Cobra, probably one of the narrowest I've seen used on a dashcam

 

[Dashmellow]

join the club



that might be more relevant perhaps if it were two different cameras based on the same hardware though, different hardware platforms have different characteristics already so a different end result is not unexpected I would think

Probably since we are dealing with digital video cameras we should stick with using digital video measuring conventions rather than older analogue ones regardless of which platform we are talking about.

 

[jokiin]

Probably since we are dealing with digital video cameras we should stick with using digital video measuring conventions rather than older analogue ones regardless of which platform we are talking about.

yeah I agree, what I mean to say is different hardware have different characteristics regardless, even with IQ tuning you can't take an IMX322 and the same lens and make it look exactly the same on Ambarella, Novatek, Zoran etc, they each have their own style about them, throw a different sensor into the mix then of course the result will be different between different products, just as the lens assembly plays its part, they're each a sum of their parts, like for like hardware comparisons are more useful if wanting to compare IQ between products

 

[arcticfire]

It's a digital pro NLE using digital video files so it must be doing a conversion to represent IRE, from what I've read it will most likely be dividing the IRE range by 256 steps (possibly 235). It's probably to represent the conventional method for editors from analogue times.

 

[arcticfire]

Here's a few more grabs from different sources if it helps.




This is my Sony Bridge camera which is going to IRE110. Poor exposure with anything in the highlight area under fairly challenging bright beach conditions. You can see how much is sitting between IRE100 and 110.







SG9665GC at -2EV.





Sony CSC going to IRE110, but handles exposure fine, probably due to good algorithms and a decent dynamic range? Very little sitting between 100 and 110.





A Mini 0805 I had only going to IRE100. Although the sky is quite blown the white car is fine.