Black Box Camera IDT Generation


Starting a thread to document where I am, which is not further away than two weeks ago as per this writing. I have been busy on Colour trying to get closer to the 0.4.0 release which some of the IDT work depends on. The current notebook is here: Google Colab

I need to implement extrapolation and potentially fitting via a function, e.g. the piece-wise camera log function from CLF. Then the colour matrix maths which we discussed a bit with Josh and Nick in the background.





I have updated the notebook, it should work with Colour 0.4.x and the latest Colour Checker Detection 0.1.3. The 1D LUT generation supports:

  1. Linear interpolation at the bottom
  2. Logarithmic extrapolation at the top
  3. Smoothing of the data
  4. Trimming of clipped data + logarithmic extrapolation

Camera Samples Resulting from Segmentation

Filtered Camera Samples (Steps 1, 2, 3)

Filtered Camera Samples 2nd Order Derivatives

Trimmed Camera Samples (Step 4)

So much things left to do but it is getting there.



1 Like


I changed the extrapolation approach to something more solid than computing the derivatives which works rather well for clean cameras such as the Panasonic GH-5 but not so much for the iPad Pro…

The idea is to compute the log slope of the center portion of the samples, it is a straight line that can then be fitted. I then blend between the current data set and that line and a constant extrapolated version of the samples. The blend is fully achieved at the end of the samples. It works rather well, a few images: