Pavel Machek
Posts
2157Following
96Followers
198
Pavel Machek
pavel
@dos I wanted to do high-quality debayering, which turns 13Mpix into 52Mpix image if I want to pass it to existing shaders. Doing high quality debayer in the shaders would solve that. Really advanced debayer is probably too hard to do on GPU, but it should be possible to do better than current debayer-and-downscale.
1
0
0
Pavel Machek
pavel
@dos I tried to do something similar -- using good, external debayer, then attempting to use GPU pipeline. I attempted to use texture mode than 8200 pictures wide, and GPU did not like that. Rest can be seen in tui/ccam. Pass --cmd:convert, but clearly more work needs to be done there. If you have your code somewhere, I can try to play with it.
1
0
0
Pavel Machek
pavel
Clicks machine (tui/ccam) can now do phase-detection auto focus. Unfortunately, for good focus, you need to integrate phase-detection and contrast-detection. Plus, you need patched kernel. Fun :-).
0
0
2
Pavel Machek
pavel
OSSEU: koho uvidim v Amsterdamu a koho potkam v letadle? :-).
OSSEU: who will be there?
OSSEU: who will be there?
0
0
0
Pavel Machek
pavel
@dos @martijnbraam @linmob Martijn, if you agree with the analysis, would it be worth it to somehow "retract" that blog post / publish updated one?
1
0
0
Pavel Machek
pavel
@dos @martijnbraam @linmob Yes, agreed, effect may be from differences between the pixels. If it was bigger, it may be worth it to try to work with it, but it is likely too small to exploit.
0
0
0
Pavel Machek
pavel
@dos @martijnbraam @linmob Aha, thanks for this. Agreed, there's nonlinearity at the top, but we should not need to correct for it as it is really small.
1
0
0
Pavel Machek
pavel
@dos @martijnbraam @linmob Can you get, dunno, 20% too bright source in R, G and B? I believe you'll be able to distinguish between 100% and 120% in the averages. And that's the nonlinearity... Aha, I see the new post :-).
0
0
0
Pavel Machek
pavel
@dos @martijnbraam @linmob So... you are right only center pixels should be considered. And it surely looks like it is linear from 0 to 95% or so. I assume there would be similar effect at the 5% near saturation on red and blue ... and that surely is non-linearity, but may not be important enough to be worth correcting.
1
0
0
Pavel Machek
pavel
@dos @martijnbraam @linmob Do you have some data to back up that linearity claim? I believe there's consensus that it will not be linear near saturation. OTOH I don't believe there's consensus of how important that effect is :-(. https://www.mdpi.com/1424-8220/24/6/1841
1
0
0
Pavel Machek
pavel
@linmob @martijnbraam Yep, sensor calibration can be a lot of fun, too. There's enough work to be done all over the place. (And first step -- linearize data from sensor -- is something I'll need to investigate for Clicks Machine).
1
0
0
@linmob @martijnbraam (And yes, you could / should modify Clicks Machine so that it can do debayer + ISP on files on demand.) (And yes, eventually this should be all integrated into libcamera).
1
0
1