[libcamera-devel] [PATCH v4 31/32] ipa: rkisp1: awb: Freeze AWB when means are too small

[Kieran Bingham]

Quoting Laurent Pinchart via libcamera-devel (2022-09-08 02:41:59)
> When the RGB means are too small, gains and color temperature can't be
> meaningfully calculated. Freeze the AWB in that case, using the
> previously calculated values.
> 
> Signed-off-by: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
> ---
>  src/ipa/rkisp1/algorithms/awb.cpp | 17 ++++++++++++++++-
>  src/ipa/rkisp1/ipa_context.cpp    |  3 +++
>  src/ipa/rkisp1/ipa_context.h      |  1 +
>  3 files changed, 20 insertions(+), 1 deletion(-)
> 
> diff --git a/src/ipa/rkisp1/algorithms/awb.cpp b/src/ipa/rkisp1/algorithms/awb.cpp
> index 5f2535688c93..404ad66e6953 100644
> --- a/src/ipa/rkisp1/algorithms/awb.cpp
> +++ b/src/ipa/rkisp1/algorithms/awb.cpp
> @@ -31,6 +31,9 @@ namespace ipa::rkisp1::algorithms {
>  
>  LOG_DEFINE_CATEGORY(RkISP1Awb)
>  
> +/* Minimum mean value below which AWB can't operate. */
> +constexpr double kMeanMinThreshold = 2.0;

Will this be a tuning parameter? Is it something to calibrate later? or
better document the expected ranges that should be used here?

What has led to the choice of 2.0, - is it simply that we can't estimate
a colour temparature from a near black / dark image?

If so - should we have a representation of overall brightness?



> +
>  Awb::Awb()
>         : rgbMode_(false)
>  {
> @@ -263,7 +266,17 @@ void Awb::process(IPAContext &context,
>         greenMean /= frameContext.awb.gains.green;
>         blueMean /= frameContext.awb.gains.blue;
>  
> -       frameContext.awb.temperatureK = estimateCCT(redMean, greenMean, blueMean);
> +       /*
> +        * If the means are too small we don't have enough information to
> +        * meaningfully calculate gains. Freeze the algorithm in that case.
> +        */
> +       if (redMean < kMeanMinThreshold && greenMean < kMeanMinThreshold &&
> +           blueMean < kMeanMinThreshold) {
> +               frameContext.awb.temperatureK = activeState.awb.temperatureK;
> +               return;
> +       }
> +
> +       activeState.awb.temperatureK = estimateCCT(redMean, greenMean, blueMean);

Aha - good, here's the active state view of the colour temperature.


While I have questions about how the threshold is determined, I think
this is still good to get in.


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

>  
>         /*
>          * Estimate the red and blue gains to apply in a grey world. The green
> @@ -290,6 +303,8 @@ void Awb::process(IPAContext &context,
>         activeState.awb.gains.automatic.blue = blueGain;
>         activeState.awb.gains.automatic.green = 1.0;
>  
> +       frameContext.awb.temperatureK = activeState.awb.temperatureK;
> +
>         LOG(RkISP1Awb, Debug) << std::showpoint
>                 << "Means [" << redMean << ", " << greenMean << ", " << blueMean
>                 << "], gains [" << activeState.awb.gains.automatic.red << ", "
> diff --git a/src/ipa/rkisp1/ipa_context.cpp b/src/ipa/rkisp1/ipa_context.cpp
> index 335cb32c538d..9cac05bec79b 100644
> --- a/src/ipa/rkisp1/ipa_context.cpp
> +++ b/src/ipa/rkisp1/ipa_context.cpp
> @@ -156,6 +156,9 @@ namespace libcamera::ipa::rkisp1 {
>   * \var IPAActiveState::awb.gains.automatic.blue
>   * \brief Automatic white balance gain for B channel
>   *
> + * \var IPAActiveState::awb.temperatureK
> + * \brief Estimated color temperature
> + *
>   * \var IPAActiveState::awb.autoEnabled
>   * \brief Whether the Auto White Balance algorithm is enabled
>   */
> diff --git a/src/ipa/rkisp1/ipa_context.h b/src/ipa/rkisp1/ipa_context.h
> index 53787320e565..d0bc9090a8f1 100644
> --- a/src/ipa/rkisp1/ipa_context.h
> +++ b/src/ipa/rkisp1/ipa_context.h
> @@ -68,6 +68,7 @@ struct IPAActiveState {
>                         } automatic;
>                 } gains;
>  
> +               unsigned int temperatureK;
>                 bool autoEnabled;
>         } awb;
>  
> -- 
> Regards,
> 
> Laurent Pinchart
>