Originally Posted by gwgill
That's interesting. By default targen will order display patches so as to minimize the change in color between the patches, so as to minimize any display settling time. This is probably opposite to an order that will minimize display auto power regulation effects. Unfortunately I don't have access to the types of displays that have this type of characteristic, so really haven't had the opportunity to develop algorithms to minimize such effects.
Sure - can you outline what your code attempts to do ? (Perhaps email me directly).
I'll ping you an email outside of here.
From what I can see this isn't auto-power regulation. All LEDs have the property that as they heat up both the amount of light they emit and the wavelength of that light changes, and this is what is happening here. So an ideal patch set would do two things:
1) Doesn't cause large swings in average power consumption of the display which will knock into relatively large swings of luminance and to a lesser extent colour shift.
2) mimic the average power consumption over time of the actual content the end user watches - if the patch sequence results in overly hot or cold screen area you'll end up correcting an image that doesn't match the real content's profile
I'm not attempting to address 2) but I'm surprised it hasn't actually got any attention from folk yet. I don't have an OLED display either but it is an obvious flaw when folk are scrabbling over half a dE, to also keep an eye on not profiling a false use case (unless you like watching patch sequences... some folk here do!
For 1) I dealt with it in the script by:
1) Converting the patch values to estimate light output (more proportional to power) for a perfect gamma (2.2 was suggested to me as OLED starting point)
2) Making something up to estimate WRGB values from RGB. So I just assumed perfect mixing and that W is R + G + B. So W is (min (R,G,B) and R,G,B are now minus W. This leaves only 3 subpixels ever active (supposedly how these displays behave)
3) Working out for the patch sequence the average power
4) Taking each patch in the list, and trying to put it somewhere else in the list, looking for improvement in the rolling average of the list. (invented some metric for better).
5) Doing 3) until it seems the patch set doesn't get any better.
Anyway, the algo is basic as I understand the idea but my math / comp-sci fu is rusty.