| Age | Commit message (Collapse) | Author |
|---|
|
The black level obtained from the tuning file in software ISP is
retrieved in init (because this is the standard algorithm method with
access to tuning data) and stored into context. But the context gets
reset in configure and the black level is lost and never applied.
Let's store the black level from the tuning file into an algorithm
instance variable and put it into the context only later in configure.
This is similar to what rkisp1 IPA does with the values obtained from
the tuning file.
Fixes: 41e3d61c745153779ed5a38634d7266bffd60d61 ("libcamera: software_isp: Clear IPA context on configure and stop")
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Tested-by: Robert Mader <robert.mader@collabora.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
On my setup, since commit fb8ad13d ("libcamera: software_isp: Move exposure+gain
to an algorithm module"), at start camera output stays very dark for dozen
of seconds, and then later slowly gets to normal. This is because existing
sensor exposure+gain settings are not used at start. We save initial
values in frameContext but in the agc algorithm we use IPA context.
Fix the problem by using in frameContext sensor values, since we already
use those in blc algorithm and change exposure type to int32_t to
unnecessary castings.
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
Tested-by: Robert Mader <robert.mader@collabora.com>
Tested-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
This patch allows obtaining a black level from a tuning file in addition
to the camera sensor helper. If both of them define a black level, the
one from the tuning file takes precedence.
The use cases are:
- A user wants to use a different black level, for whatever reason.
- There is a sensor without known gains but with a known black level.
Because a camera sensor helper cannot be defined without specifying
gains, the only way to specify the black level is using the tuning
file. Software ISP uses its fallback gain handling in such a case.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The black level in software ISP is unconditionally guessed from the
obtained frames. CameraSensorHelper optionally provides the black level
from camera specifications now. Let's use the value if available.
If the black level is not available from the given CameraSensorHelper
instance, it's still determined on the fly.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Tested-by: Robert Mader <robert.mader@collabora.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
A minor wording improvement suggested on refactoring review.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The black level is likely to get updated, if ever, only after exposure
or gain changes. Don't compute its possible updates if exposure and
gain are unchanged.
It's probably not worth trying to implement something more
sophisticated. Better to spend the effort on supporting tuning files.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
This is the last step to fully convert software ISP to Algorithm-based
processing.
The newly introduced frameContext.sensor parameters are set, and the
updated code moved, before calling Algorithm::process() to have the
values up-to-date in stats processing.
Resolves software ISP TODO #10.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
It's more natural to represent color gains as floating point numbers
rather than using a particular pixel-related representation.
double is used rather than float because it's a more common floating
point type in libcamera algorithms. Otherwise there is no obvious
reason to select one over the other here.
The constructed color tables still use integer representation for
efficiency.
Black level still uses pixel (integer) values, for consistency with
other libcamera parts.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
After black level handling has been moved to an algorithm module, white
balance and the construction of color tables can be moved to algorithm
modules too.
This time, the moved code is split between stats processing and
parameter construction methods. It is also split to two algorithm
modules:
- White balance computation.
- Gamma table computation and color lookup tables construction. While
this applies the color gains computed by the white balance algorithm,
it is not directly related to white balance. And we may want to
modify the color lookup tables in future according to other parameters
than just gamma and white balance gains.
Gamma table computation and color lookup tables construction could be
split to separate algorithms too. But there is no big need for that now
so they are kept together for simplicity.
This is the only part of the software ISP algorithms that sets the
parameters so emitting setIspParams can be moved to prepare() method.
A more natural representation of the gains (and black level) would be
floating point numbers. This is not done here in order to minimize
changes in code movements. It will be addressed in a followup patch.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Acked-by: Umang Jain <umang.jain@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The black level determination, already present as a separate class, can
be moved to the prepared Algorithm processing structure. It is the
first of the current software ISP algorithms that is called in the stats
processing sequence, which means it is also the first one that should be
moved to the new structure. Stats processing starts with calling
Algorithm-based processing so the call order of the algorithms is
retained.
Movement of this algorithm is relatively straightforward because all it
does is processing stats.
The comment about getting black level from the tuning files is dropped.
The black level will be taken from CameraSensorHelper if available and
that will be implemented in one of the followup patches.
Black level is now recomputed on each stats processing. In a future
patch, after DelayedControls are used, this will be changed to recompute
the black level only after exposure/gain changes.
The black level depends on the sensor used, the computed value can be
reused in a followup capture sessions with the same sensor. Thus it is
sufficient to (re)set the initial value in BlackLevel::init.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
We are ready to introduce algorithms now. First, let's create
algorithms. The algorithms are not called yet, calls to them will be
added in followup patches.
The maximum number of contexts is set to the same value as in hardware
pipelines.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
Software ISP image processing algorithms are currently defined in a
simplified way, different from other libcamera pipelines. This is not
good for several reasons:
- It makes the software ISP code harder to understand due to its
different structuring.
- Adding more algorithms may make the code harder to understand
generally.
- Mass libcamera code changes may not be easily applicable to software
ISP.
- Algorithm sharing with other pipelines is not easily possible.
This patch introduces basic software ISP IPA skeletons structured
similarly to the other pipelines. The newly added files are currently
not used or compiled and the general skeleton structures don't contain
anything particular. It is just a preparation step for a larger
refactoring and the code will be actually used and extended as needed in
followup patches.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
