[libcamera-devel] [PATCH v4 2/2] ipa: rpi: common: Handle AEC/AGC flicker controls
Kieran Bingham
kieran.bingham at ideasonboard.com
Tue Jul 18 13:16:02 CEST 2023
Quoting David Plowman (2023-07-18 10:13:42)
> Hi Kieran
>
> On Tue, 18 Jul 2023 at 09:48, Kieran Bingham
> <kieran.bingham at ideasonboard.com> wrote:
> >
> > Quoting David Plowman via libcamera-devel (2023-07-18 09:24:24)
> > > Hi Jacopo
> > >
> > > Thanks for the comments.
> > >
> > > On Tue, 18 Jul 2023 at 08:09, Jacopo Mondi
> > > <jacopo.mondi at ideasonboard.com> wrote:
> > > >
> > > > Hi David
> > > >
> > > > On Thu, Jul 13, 2023 at 04:12:18PM +0100, David Plowman via libcamera-devel wrote:
> > > > > We handle the flicker modes by passing the correct period to the
> > > > > AEC/AGC algorithm which already contains the necessary code.
> > > > >
> > > > > The "Auto" mode, as well as reporting the detected flicker period via
> > > >
> > > > If so you shouldn't register
> > > >
> > > > { &controls::AeFlickerMode, ControlInfo(controls::AeFlickerModeValues) },
> > > >
> > > > otherwise AeFlickerMode::FlickerAuto will be exposed to applications ?
> > >
> > > Indeed, it might be better not to expose the auto value!
> > >
> > > >
> > > >
> > > > > the "AeFlickerDetected" metadata, are unsupported for now.
> > > > >
> > > > > Signed-off-by: David Plowman <david.plowman at raspberrypi.com>
> > > > > ---
> > > > > src/ipa/rpi/common/ipa_base.cpp | 64 ++++++++++++++++++++++++++++++++-
> > > > > src/ipa/rpi/common/ipa_base.h | 6 ++++
> > > > > 2 files changed, 69 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/src/ipa/rpi/common/ipa_base.cpp b/src/ipa/rpi/common/ipa_base.cpp
> > > > > index f40f2e71..81d65ea5 100644
> > > > > --- a/src/ipa/rpi/common/ipa_base.cpp
> > > > > +++ b/src/ipa/rpi/common/ipa_base.cpp
> > > > > @@ -61,6 +61,8 @@ const ControlInfoMap::Map ipaControls{
> > > > > { &controls::AeConstraintMode, ControlInfo(controls::AeConstraintModeValues) },
> > > > > { &controls::AeExposureMode, ControlInfo(controls::AeExposureModeValues) },
> > > > > { &controls::ExposureValue, ControlInfo(-8.0f, 8.0f, 0.0f) },
> > > > > + { &controls::AeFlickerMode, ControlInfo(controls::AeFlickerModeValues) },
> > > > > + { &controls::AeFlickerCustom, ControlInfo(100, 1000000) },
> > > > > { &controls::Brightness, ControlInfo(-1.0f, 1.0f, 0.0f) },
> > > > > { &controls::Contrast, ControlInfo(0.0f, 32.0f, 1.0f) },
> > > > > { &controls::Sharpness, ControlInfo(0.0f, 16.0f, 1.0f) },
> > > > > @@ -97,7 +99,7 @@ namespace ipa::RPi {
> > > > >
> > > > > IpaBase::IpaBase()
> > > > > : controller_(), frameCount_(0), mistrustCount_(0), lastRunTimestamp_(0),
> > > > > - firstStart_(true)
> > > > > + firstStart_(true), flickerState_({ 0, 0s })
> > > > > {
> > > > > }
> > > > >
> > > > > @@ -812,6 +814,66 @@ void IpaBase::applyControls(const ControlList &controls)
> > > > > break;
> > > > > }
> > > > >
> > > > > + case controls::AE_FLICKER_MODE: {
> > > > > + RPiController::AgcAlgorithm *agc = dynamic_cast<RPiController::AgcAlgorithm *>(
> > > > > + controller_.getAlgorithm("agc"));
> > > > > + if (!agc) {
> > > > > + LOG(IPARPI, Warning)
> > > > > + << "Could not set AeFlickerMode - no AGC algorithm";
> > > > > + break;
> > > > > + }
> > > > > +
> > > > > + int32_t mode = ctrl.second.get<int32_t>();
> > > > > + bool modeValid = true;
> > > > > +
> > > > > + switch (mode) {
> > > > > + case controls::FlickerOff: {
> > > > > + agc->setFlickerPeriod(0us);
> > > > > +
> > > > > + break;
> > > > > + }
> > > >
> > > > Do you need braces ?
> > >
> > > Sure, I can remove them!
> > >
> > > >
> > > > > +
> > > > > + case controls::FlickerCustom: {
> > > > > + agc->setFlickerPeriod(flickerState_.customPeriod);
> > > > > +
> > > > > + break;
> > > > > + }
> > > > > +
> > > > > + default:
> > > > > + LOG(IPARPI, Error) << "Flicker mode " << mode << " is not supported";
> > > > > + modeValid = false;
> > > > > +
> > > > > + break;
> > > > > + }
> > > > > +
> > > > > + if (modeValid)
> > > > > + flickerState_.mode = mode;
> > > > > +
> > > > > + break;
> > > > > + }
> > > > > +
> > > > > + case controls::AE_FLICKER_CUSTOM: {
> > > > > + RPiController::AgcAlgorithm *agc = dynamic_cast<RPiController::AgcAlgorithm *>(
> > > > > + controller_.getAlgorithm("agc"));
> > > > > + if (!agc) {
> > > > > + LOG(IPARPI, Warning)
> > > > > + << "Could not set AeFlickerCustom - no AGC algorithm";
> > > > > + break;
> > > > > + }
> > > > > +
> > > > > + uint32_t customPeriod = ctrl.second.get<int32_t>();
> > > > > + flickerState_.customPeriod = customPeriod * 1.0us;
> > > > > +
> > > > > + /*
> > > > > + * We note that it makes no difference if the mode gets set to "custom"
> > > > > + * first, and the period updated after, or vice versa.
> > > > > + */
> > > >
> > > > True, but what if the app never provides a value for
> > > > controls::AE_FLICKER_CUSTOM ? Should you initialize customPeriod to
> > > > some sensible default ?
> > >
> > > In our implementation we use the value 0 to mean "no flicker
> > > cancellection", so setting the mode to custom/manual without a flicker
> > > period will cause it to do nothing (until you set one). I'm OK with
> > > this behaviour unless anyone objects, though I don't feel strongly!
> > > I'd rather document that setting custom/manual mode without a period
> > > leads to undefined behaviour rather than mandating a particular value,
> > > I think, as not every PH may wish to do the same. What do you think?
> >
> >
> > I had wondered if we could get this down to a single control by having
> > '0' as no flicker cancellation - but we'd have no way to express 'auto'
> > so I think we do still need two, and it lets us report in the metadata
> > if the value was determined automatically or explicitly I guess? (Though
> > the app should know that anyway as it would know if it set the manual
> > value?)
>
> Yes, I think we do still want a separate control.
>
> >
> > Throwing paint on the bike shed I feel like 'AeFlickerPeriod' would be
> > better suited for the control name as then it would be usable to report
> > the period back in the metadata when it's detected with 'Auto' mode.
>
> I agree that 'AeFlickerPeriod' is a better name, particularly if the
> 'custom' mode becomes the 'manual' mode.
>
> >
> > What limits will be imposed for the period ? I think setting
> > AeFlickerMode to 'Custom' (or 'Manual'?) and not setting an
> > {AeFlickerCustom,AeFlickerPeriod} control should be the same as not
> > changing anything, but I expect that maybe the ControlInfo might have
> > limits that would preclude setting a period of '0' ?
>
> I guess implementations get to choose their own limits though, for the
> most part, I see no reason not to allow them to be very wide.
> Extremely short or extremely long periods are going to be increasingly
> unlikely IRL, of course.
The part I was wondering about is if we 'defined' 0 as being equivalent
to controls::FlickerOff, but that could make it harder to validate the
control if it means any value from 0...MAX is 'acceptable'.
But yes, I think this can all be handled by the pipeline handler/IPA.
> > It might be reasonable to say setting AeFlickerMode to Manual/Custom
> > will only take effect when the corresponding AeFlickerPeriod control is
> > applied ?
>
> Yes, we could say that setting the mode to manual has no effect until
> a period is also set. Though of course I assume the period is
> "sticky", so once you've set it, you can go to and fro to manual mode
> without re-sending the said period.
<Time jump, now that I've writen the below> Does this mean you expect to
go from AeFlickerMode=Manaul(20000) to AeFlickerMode=Auto(Detected
30000) and then go to AeFlickerMode=Off(0) and when you finally go back
to AeFlickerMode=Manual(no period applied) you expect period to be
'remembered' as 20000?
>
> I guess we have to ask what happens if you go from auto to manual
> mode, having never set a period. What happens then? TBH, I think I'd
> be quite happy to say that selecting manual mode, without ever setting
> a period, is the same as being "off". Thoughts?
I suspect this might be something we should consider for all
'multi-control' settings, for how they react if they are only partially
set. As long as the behaviour is documented I think it's fine though.
I think I would consider setting the period to be 'sticky' as well so if
you did:
1) AeFlickerMode = Off
2) AeFlickerPeriod = 10000
3) AeFlickerMode = Manual
The Period would be defined at 10000
I think for switching from Auto to Manual it would remain at whatever
the autodetected value was...
1) AeFlickerMode = Off
2) AeFlickerMode = Auto #Metadata reports AeFlickerPeriod as 20000
3) AeFlickerMode = Manual # AeFlickerPeriod now 'locked' at manual 20000
Hrm... but that seems to differ with your suggestion above of setting it
to 'off'
To me 'sticky' to last actual value (where off = 0, auto == detected
value) seems easiest to codify and explain? But maybe I'm missing
something obvious.
And now I feel like we should have helper classes in libipa to codify
the behaviour of all these controls in a common way for each PH!?
(later)
--
Kieran
>
> David
>
> >
> >
> > --
> > Kieran
> >
> >
> > >
> > > Thanks!
> > > David
> > >
> > > >
> > > > > + if (flickerState_.mode == controls::FlickerCustom)
> > > > > + agc->setFlickerPeriod(flickerState_.customPeriod);
> > > > > +
> > > > > + break;
> > > > > + }
> > > > > +
> > > > > case controls::AWB_ENABLE: {
> > > > > /* Silently ignore this control for a mono sensor. */
> > > > > if (monoSensor_)
> > > > > diff --git a/src/ipa/rpi/common/ipa_base.h b/src/ipa/rpi/common/ipa_base.h
> > > > > index 39d00760..22823176 100644
> > > > > --- a/src/ipa/rpi/common/ipa_base.h
> > > > > +++ b/src/ipa/rpi/common/ipa_base.h
> > > > > @@ -116,6 +116,12 @@ private:
> > > > > /* Frame duration (1/fps) limits. */
> > > > > utils::Duration minFrameDuration_;
> > > > > utils::Duration maxFrameDuration_;
> > > > > +
> > > > > + /* The current state of flicker avoidance. */
> > > > > + struct FlickerState {
> > > > > + int32_t mode;
> > > > > + utils::Duration customPeriod;
> > > > > + } flickerState_;
> > > > > };
> > > > >
> > > > > } /* namespace ipa::RPi */
> > > > > --
> > > > > 2.30.2
> > > > >
More information about the libcamera-devel
mailing list