On 5D3 and probably all other recent cameras, latest mlv_dump requires --black-fix=2048, regardless of ML build
(since this post).
With latest ML bulid from the Experiments page AND latest mlv_dump, there should be no need to change the black level
(since this post).
Checklist:
- with 10-bit files, the black level from MLV (visible with mlv_dump -v) should be 128 (127 is wrong)
- with 14-bit files, the black level should be close to 2048 (small variations are OK)
- the black level from the DNG (always upconverted to 14 bits) should be close to 2048 (e.g. 2050 is fine, 2032 is wrong)
- there should be no color difference between 10 and 14 bit files in shadows
If any of the above is not true for you with the latest experimental build and latest mlv_dump, please report.
For authors of other postprocessing tools: to avoid color casts at lower bit depths, conversion to a higher bit depth must be done by adding 0.5 LSB to the input signal (
explained here). Black level scaling requires no trickery (e.g. 128 -> 2048).
Relevant commits:
e3ef8c7 and
a31228c.