Specify Output Bit Depth about dnglab

[kmilos]

If I did though, it currently would be unusable because the bit depth is being resampled to 16bit from whatever the source is (most cinema dngs i have are 12bit, most stills are 14bit).

A bug report for the above

This is not a bug per se, but how it works in a simplest implementation. Uncompressed samples >8b must be stored as at least 2 bytes (16bits). They are not "resampled" in any way, just easiest to be stored that way (LSB aligned with some MSBs wasted/zeroes). One could do some bit packing to avoid wastage when this is then no longer true, but this is not mandatory (although it is recommended by the TIFF spec, so this could be reported as a feature request/enhancement).

It's the WhiteLevel tag that determines the actual bit depth of the sample after it is loaded as a uint16.

I'm pretty sure such files work ok w/ any Adobe SW, LibRaw-based SW (e.g. ART, rawpy, etc.), and RawSpeed-based SW (e.g. darktable), or vkdt using rawler. For any other apps not working with these simplest uncompressed files, there should be a bug filed with them.

[cytrinox]

As lossless compression in DNG is widely supported, I'v not implemented any optimization for uncompressed output. Do you have a usecase where lossless compression is unusable for you?

[jedypod]

I don't really have a use-case where I would want to be outputting uncompressed dngs. If I did though, it currently would be unusable because the bit depth is being resampled to 16bit from whatever the source is (most cinema dngs i have are 12bit, most stills are 14bit).

So I guess my rambling post above is twofold:

1. A bug report for the above
2. A request to specify the output bit depth. This would actually be quite useful for me I think. Taking a 12bit uncompressed dng source and re-encoding to 10bit dng with lj92 compression as 10bit with a nonlinear transfer function would be very useful. Currently I don't believe that is possible with dnglab?

Hopefully that clarifies things.

[cytrinox]

If lossless compression is selected, output bit depth is usually matches source bit depth (with few exceptions).
For uncompressed output, dnglab only supports 16 bit unpacked data for now.

Bit depth may change if source contains a linearization table - as linearization is applied during decoding process.