SG9665GC High Contrast/Dynamic Range Flaw and other Discouragements

AE table can be applied any way you please really, WDR is applied to dark areas across the whole image even if your AE table is restricted to a smaller specific field
 
Ah! So there we could gain something.
Rewriting the table so that it will shorten the exposure in bright areas when it detects them should do the trick.

Not surprised you're already doing this! ;)
 
Here are a couple of screenshots from a Panorama (IMX322) and Mio 538 (AR0330).
Notice motion blur on the road sign, cloud detail and detail of darker areas.

-



 
And here's a night shot from a Mio 528 (same as Mio 538 apart from not having GPS) (top picture) and a CF-100 (IMX122 - predecessor of the IMX322 in the SG/Panorama cams) (bottom picture).
Can you see the pedestrian looking down at their phone and about to step out into the road? (one-quarter of the way from the right edge)

-

 
2000rpm said:
Here are a couple of screenshots from a Panorama (IMX322) and Mio 538 (AR0330).
Notice motion blur on the road sign,
Click to expand...

I have noticed the same thing on SG when try to capture the incoming car's license plate.
Both sensors are using electronic rolling shutter. So this won't be the reason. I have notice GC use GOP setting of M=1,N=15 for H.264/AVC. encoding. Wonder what setting Mio 538 used.
Also, shutter speed may have play here as well.
 
painkiller said:
I have noticed the same thing on SG when try to capture the incoming car's license plate..
Click to expand...

Generally speaking, a cam doesn't easily capture details when it has a wide viewing angle. On the other hand a cam with a narrow viewing angle doesn't see what's happening at the sides of the car.

One pair of pictures summarises that (Panorama II S vs Cobra 840E):

-



 
I've never had the complete washout but I do get the highlight clipping with the SG9665GC. I still feel that the dynamic range is leaning towards exposing for shadows hence the highlights clipping so easily. I thought I'd run some footage through the video editor and play about with Shadow and Highlight adjustment and take some screen grabs. I'm not making adjustments to screenshots from the footage but adjusting the actual footage itself. Obviously I'm working from compressed footage out of the dashcam but even with that I've noticed that I can recovery very little from the highlights and nothing when it's completely white, yet I can recover a lot of the shadows even when they are near pitch black. The highlights are definitely being clipped and it seems quite early also.

It even seems I can recover the shadows with pretty much no impact on the highlights which makes me wonder if the dynamic range can be shifted more to expose better for highlights so they don't clip and perhaps there's still enough detail there to increase the gain in the shadows to allow for an overall good exposure? I'm not sure quite how much processing power would be require to do this realtime or if it's even possible at all. I may try running at -1EV or even -2EV and see how that effects the highlights and how much shadow I can pull back in software. I get the feeling that the full capability of the sensor is not being used here, or perhaps firmware, processing power etc is restricting such? Not sure, just a guess.

Looking forward to the firmware updates though. Here's the screen grabs:



1 Standard.jpg
1 100% Highlight Recovery.jpg
1 100% Shadow Recovery.jpg





2 Standard.jpg
2 100% Highlight Recovery.jpg
2 100% Shadow Recovery.jpg






3 Standard.jpg
3 100% Highlight Recovery.jpg
3 100% Shadow Recovery.jpg
 
You can't recover something that was never there is the first place. It is the concept of GIGO - Garbage In - Garbage Out. If the highlights are not there in the first place in the original footage they can't be recovered.

As discussed over a month ago in this thread, the top third or so of the tonal range seems to get lost completely when the SG9665GC encounters conditions that push it over a threshold where it can no longer handle the dynamic range it is presented with.
The problem appears to be influenced by exposure but is very much a tone curve issue.

tonechart2.jpg tonechart.jpg

Here is an interesting example of a Mobius encountering a challenging lighting situation that might be comparable to having a dark hood taking up much of the bottom of the FOV (moreso even) or perhaps driving under an overpass. The Mobius manages to handle the dynamic range in a situation where the GC would not. The highlight detail is all still there, even if it is beginning to show some strain at the upper end of the tone curve.

Mobius-dark-bottom.jpg
 
Last edited:
Dashmellow said:
Here is an interesting example of a Mobius encountering a challenging lighting situation that might be comparable to having a dark hood taking up much of the bottom of the FOV (moreso even) or perhaps driving under an overpass. The Mobius manages to handle the dynamic range in a situation where the GC would not. The highlight detail is all still there, even if it is beginning to show some strain.

View attachment 22553
Click to expand...

Based on this screenshot I can tell you with absolute certainty that the GC would suffer enormously in this situation and that is a direct result of the current AE table, it is completely wrong for this situation, I've spent the past two days with the engineers going through a huge list of changes that are happening, our next beta is probably going to take around two weeks as there's a lot of work to do
 
jokiin said:
Based on this screenshot I can tell you with absolute certainty that the GC would suffer enormously in this situation and that is a direct result of the current AE table, it is completely wrong for this situation, I've spent the past two days with the engineers going through a huge list of changes that are happening, our next beta is probably going to take around two weeks as there's a lot of work to do
Click to expand...

I'm very encouraged to hear that you and your engineers are working on this. Thanks!

Yeah, I've been thinking of posting that screen shot for a while now. Of course, it is a different platform but the Mobius is consistently far more capable of handling challenging lighting conditions. I believe one reason for that is the Mobius' heritage of RC and FPV use where challenging lighting was always par for the course from the beginning and which is probably behind the constant skillful attention to firmware tweaking that the Mobius developer is known for.
 
This is the clipping part I'm talking about and as said was discussed previously, but what I'm now trying to demonstrate is that even though the SG9665GC's dynamic range seems to be leaning towards exposing for the shadows (hence the early clipping of the highlights) there is still a fair bit of detail in the shadows which can be recovered. So it's making me feel like it's not using it's full dynamic range which is it capable of perhaps due to the AE table (which Jokiin is addressing) and therefore it doesn't need to shift the dynamic range towards exposure for the shadows in the first place because it should be capable for exposing the highlights correctly and still be able to pull a decent amount of details in the shadows also. I'm pretty confident that a revised AE table should be able to solve this. Although I'm not getting nearly as bad results, I do feel that the exposure can be better hence my interest in this :)
 
Dashmellow said:
I'm very encouraged to hear that you and your engineers are working on this. Thanks!
.
Click to expand...
thanks to you also, a great deal of the sample files we went through was material you have supplied, I know it has taken longer than what anyone would have liked but a million email conversations couldn't get through what I was able to do in person in just two days, it's much different when you can do this sort of thing one on one

thanks also to @DT MI as we finally found a GPS bug we've been working on and he provided sample material for that
 
arcticfire said:
This is the clipping part I'm talking about and as said was discussed previously, but what I'm now trying to demonstrate is that even though the SG9665GC's dynamic range seems to be leaning towards exposing for the shadows (hence the early clipping of the highlights) there is still a fair bit of detail in the shadows which can be recovered. So it's making me feel like it's not using it's full dynamic range which is it capable of perhaps due to the AE table (which Jokiin is addressing) and therefore it doesn't need to shift the dynamic range towards exposure for the shadows in the first place because it should be capable for exposing the highlights correctly and still be able to pull a decent amount of details in the shadows also. I'm pretty confident that a revised AE table should be able to solve this. Although I'm not getting nearly as bad results, I do feel that the exposure can be better hence my interest in this :)
Click to expand...

I understand what you are trying to say but in my experience with this issue it is often the upper end of the tonal range that gets lost while the shadow detail and mid range is fine. In many circumstances, the exposures are perfect but the camera fails when the dynamic range overwhelms the scene that is encountered. Hopefully the AE table reworkings should help.
 
I remember my 0826 always seem to handle the exposure well. Could we nick it's AE table? lol
 
Dashmellow said:
I understand what you are trying to say but in my experience with this issue it is often the upper end of the tonal range that gets lost while the shadow detail and mid range is fine. In many circumstances, the exposures are perfect but the camera fails when the dynamic range overwhelms the scene that is encountered. Hopefully the AE table reworkings should help.
Click to expand...

in that screenshot you posted above based on the current AE table it may as well be pitch dark, it would over drive the sensor like crazy, no wonder the upper end of the range was clipping
 
You must log in or register to reply here.
Back
Top