I had missed the original post so I had to give the RGB denoiser a go first. I always found the built-in denoise modes from rpicam a little too aggressive for my taste and eventually settled to using the vaguedenoise filter from ffmpeg combined with a low gain. Though one thing I never explored was using the tuning file to see if the denoising could made more conservative.
So far this one seems to blow the built-in denoisers out of the water, especially when using high gain values.
Performance is another story. Below are some numbers from a Pi4 (4GB) using the 12 megapixel camera module v3.
RGB denoise: 23.5 seconds per megapixel
RAW denoise: (small): 4.5 seconds per megapixel
RAW denoise (large): 13.6 seconds per megapixel
Almost 5 minutes for a single image is a bit on the high side but I guess expected considering the modest computing resources available on the Pi. Investing on a Pi5 would help here of course![:)]( "Smile")
Except for performance, I'm not seeing a huge difference between the small and large models for RAW although this is with a sample size of one. Need to run more tests.
So far this one seems to blow the built-in denoisers out of the water, especially when using high gain values.
Performance is another story. Below are some numbers from a Pi4 (4GB) using the 12 megapixel camera module v3.
RGB denoise: 23.5 seconds per megapixel
RAW denoise: (small): 4.5 seconds per megapixel
RAW denoise (large): 13.6 seconds per megapixel
Almost 5 minutes for a single image is a bit on the high side but I guess expected considering the modest computing resources available on the Pi. Investing on a Pi5 would help here of course
Except for performance, I'm not seeing a huge difference between the small and large models for RAW although this is with a sample size of one. Need to run more tests.
Statistics: Posted by Hourglass — Fri Jun 06, 2025 10:12 pm