OpenCV 4.5

As of OpenCV 4.5, the TIFFDecoder does not support 12 bits (integer).
It looks like a deliberate choice since decoding 8 and 16 are supported with optimized icv conversion functions.

Would it be accepted to add a "slow" unpacking step for non-standard bit depths ?

There are at least two problems :
-it would be logical to automatically scale the resulting cv::Mat (here by 1<<(16-12)), but the caller would not be aware that such a conversion was done
-TIFF is so complex that it's hard to ensure that all and every packing scheme is supported