In some cases a pair of inverse / forward LUTs don't get optimized away by OCIO but instead combined in a single half-domain LUT. This seems to happen when there are a number of other operators in between, like a pair of MatrixOp to handle domain scales.
For context, the example I saw this in our config was in the case of an ACEScg to ARRI LogC4 -> CDL (identity) -> ARRI LogC4 to ACEScg conversion, the LogC4 LUT was designed to preserve -1 to 500 range. Here is the Group transform from the Processor for reference. Note that the ExponentOp clamp is ignored when optimizing.
In some cases a pair of inverse / forward LUTs don't get optimized away by OCIO but instead combined in a single half-domain LUT. This seems to happen when there are a number of other operators in between, like a pair of MatrixOp to handle domain scales.
For context, the example I saw this in our config was in the case of an ACEScg to ARRI LogC4 -> CDL (identity) -> ARRI LogC4 to ACEScg conversion, the LogC4 LUT was designed to preserve -1 to 500 range. Here is the Group transform from the Processor for reference. Note that the ExponentOp clamp is ignored when optimizing.