[v3,00/16] Some rkisp1 awb improvements

23116 mbox series

Hi all,

This series contains a few fixes that greatly improve and stabilize the
AWB handling on rkisp1.

After the feedback from the first version, some parts were rewritten and
Laurent added a generic matrix inverse function.

Patches 1-8 add utility functions to Vector and Matrix.

The rest fixes some bugs and generally improves the AWB on rkisp1.
Please see the individual patches for details.

Best regards,
Stefan

Changes in v2:

- Replaced the matrix cast() and inverse() functions by better alternatives
- Added a manual CCM control
- Some small fixes (see logs in individual patches)

Changes in v3:
- Included feedback from review (See individual changelogs)
- Split patch 10 into 10 and 11
- Dropped patch 13 (Use gains from color temperature curve) because it
  was actually doing the wrong thing when bayesian awb was active and
the benefit was not that big anymore, now that bayesian awb is
available.
- Dropped patch 15 (Set rgb mean limits to 180) because it was not
  really necessary and the description was not sufficient.


Laurent Pinchart (2):
  libcamera: matrix: Extend multiplication operator to heterogenous
    types
  libcamera: vector: Extend matrix multiplication operator to
    heterogenous types

Stefan Klug (14):
  libcamera: matrix: Replace SFINAE with static_asserts
  libcamera: matrix: Make most functions constexpr
  libcamera: matrix: Add a Span based constructor
  libcamera: vector: Add a Span based constructor
  libcamera: matrix: Add inverse() function
  test: Add minimal test for Matrix
  ipa: rkisp1: Refactor automatic/manual structure in IPAActiveState
  ipa: rkisp1: algorithms: awb: Fix wrong colour temperature reporting
  ipa: rkisp1: ccm/lsc: Fix CCM/LSC based on manual color temperature
  ipa: rkisp1: Implement manual ColourCorrectionMatrix control
  libipa: awb: Make result of gainsFromColourTemp optional
  ipa: rkisp1: Damp color temperature regulation
  ipa: rkisp1: awb: Take the CCM into account for the AWB gains
    calculation
  ipa: rkisp1: awb: Avoid division by zero

 include/libcamera/internal/matrix.h |  71 +++++++----
 include/libcamera/internal/vector.h |  17 ++-
 src/ipa/libipa/awb.cpp              |   2 +-
 src/ipa/libipa/awb.h                |   3 +-
 src/ipa/libipa/awb_bayes.cpp        |   4 +-
 src/ipa/libipa/awb_bayes.h          |   2 +-
 src/ipa/libipa/awb_grey.cpp         |   6 +-
 src/ipa/libipa/awb_grey.h           |   2 +-
 src/ipa/rkisp1/algorithms/awb.cpp   |  64 ++++++----
 src/ipa/rkisp1/algorithms/awb.h     |   2 -
 src/ipa/rkisp1/algorithms/ccm.cpp   |  66 ++++++++--
 src/ipa/rkisp1/algorithms/ccm.h     |   6 +
 src/ipa/rkisp1/algorithms/lsc.cpp   |   6 +-
 src/ipa/rkisp1/ipa_context.cpp      |  36 ++++--
 src/ipa/rkisp1/ipa_context.h        |  15 ++-
 src/libcamera/matrix.cpp            | 181 +++++++++++++++++++++++++++-
 src/libcamera/vector.cpp            |  13 +-
 test/matrix.cpp                     |  53 ++++++++
 test/meson.build                    |   1 +
 19 files changed, 455 insertions(+), 95 deletions(-)
 create mode 100644 test/matrix.cpp