Thanks for posting the Pi 4 numbers. Comparing to a Pi 5, they seem mostly "as expected" to me, I think.Obviously all this stuff isn't intended for any kind of real time operation, so there are only certain applications where it will be useful. Remember too that you don't have to run it on a Pi - you can copy the images for processing over to your squillion core Threadripper!
The question of how best to denoise is an interesting one. It's definitely worth messing around a bit to see if you can find something you like better - after all, hardware denoise is "free". On a Pi 4 you have spatial denoise, with a pair of simple tuning parameters described in section 5.6.4 of the tuning guide. When you adjust denoise, you may well want to adjust the sharpening too (section 5.13.4 this time). You do have a runtime control for "sharpness" too, but it gives you less fine control.
A Pi 5 is worth considering too. Not just because the spatial denoise is more capable, but because you have temporal noise. This makes a big difference is cleaning up images in many cases. If you can let the camera mode run for ~10 frames before capturing, this could be a huge benefit.
I'm probably digressing a bit here, so if anyone wants to continue the "hardware denoise" discussion, perhaps we can move to a new thread!!
The question of how best to denoise is an interesting one. It's definitely worth messing around a bit to see if you can find something you like better - after all, hardware denoise is "free". On a Pi 4 you have spatial denoise, with a pair of simple tuning parameters described in section 5.6.4 of the tuning guide. When you adjust denoise, you may well want to adjust the sharpening too (section 5.13.4 this time). You do have a runtime control for "sharpness" too, but it gives you less fine control.
A Pi 5 is worth considering too. Not just because the spatial denoise is more capable, but because you have temporal noise. This makes a big difference is cleaning up images in many cases. If you can let the camera mode run for ~10 frames before capturing, this could be a huge benefit.
I'm probably digressing a bit here, so if anyone wants to continue the "hardware denoise" discussion, perhaps we can move to a new thread!!
Statistics: Posted by therealdavidp — Mon Jun 09, 2025 10:03 am