[PATCH v2 16/17] ipa: rkisp1: awb: Take the CCM into account for the AWB gains calculation

Kieran Bingham kieran.bingham at ideasonboard.com
Mon Mar 31 19:56:27 CEST 2025


Quoting Stefan Klug (2025-03-19 16:11:21)
> The AWB measurements are taken after the CCM. This can be seen by
> enabling debug logging on AWB, disabling AWB (stats will still be
> processed) and manually chaning the CCM.
> 
> This means that the estimated colour temperature and the corresponding
> CCM also lead to changed rgbMeans which in turn leads to oscillations.
> Fix that by applying the inverse transform on the rgbMeans.
> 

I'm happy to see this one!

> Signed-off-by: Stefan Klug <stefan.klug at ideasonboard.com>
> 
> ---
> 
> Changes in v2:
> - Improved commit message
> - Added comment in the code
> ---
>  src/ipa/rkisp1/algorithms/awb.cpp | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/src/ipa/rkisp1/algorithms/awb.cpp b/src/ipa/rkisp1/algorithms/awb.cpp
> index 324bc14b42a0..e8b04d03748d 100644
> --- a/src/ipa/rkisp1/algorithms/awb.cpp
> +++ b/src/ipa/rkisp1/algorithms/awb.cpp
> @@ -412,6 +412,12 @@ RGB<double> Awb::calculateRgbMeans(const IPAFrameContext &frameContext, const rk
>                 rgbMeans = rgbMeans.max(0.0);
>         }
>  
> +       /*
> +        * The ISP computes the AWB means after applying the ccm. Apply the
> +        * inverse as we want to get the raw means before the colour gains.
> +        */
> +       rgbMeans = frameContext.ccm.ccm.inverse() * rgbMeans;
> +

So 'simple' yet so powerful thanks to the helpers before.

Reviewed-by: Kieran Bingham <kieran.bingham at ideasonboard.com>


Sounds silly, but I would have written this as :

	rgbMeans = rgbMeans * frameContext.ccm.ccm.inverse();

But I (very much) hope that's identical, but reads as 'multiply the rgb
by the inverse ccm' to me, (as opposed to 'multiply the inverse CCM by
the rgbMeans).

Anyway, it shouldn't make a difference so it's fine either way to me.


>         /*
>          * The ISP computes the AWB means after applying the colour gains,
>          * divide by the gains that were used to get the raw means from the
> -- 
> 2.43.0
>


More information about the libcamera-devel mailing list