Floating Point Pixels with C# Half

[LaurentInSeattle]

My little photo app is currently using ImageSharp with a pixel type of RgbaVector. That's 4 Single floats for a total of 128 bits per pixel.

Since most image processing are 'happy' with half precision and dynamic range, is it possible to have a pixel type as a Vector of Half floats ? This would drastically cut memory requirements for large images provided by modern DSLR's.

Ideally it should work exactly like RgbaVector providing R G B properties as C# half instead of float.

[JimBobSquarePants]

New in v4.1.0

https://docs.sixlabors.com/api/ImageSharp/SixLabors.ImageSharp.PixelFormats.RgbaHalf.html

[LaurentInSeattle]

Great ! I am going to give it a shot ASAP!

What else do we get with version 4.1.0 ?

[LaurentInSeattle]

NVM: Found it here: https://github.com/SixLabors/ImageSharp/releases/tag/v4.1.0

[JimBobSquarePants]

You should use NuGet!!

[LaurentInSeattle]

I do use Nuget :) And conversion from RgbaVector to RgbaHalf is almost done.

[LaurentInSeattle]

Completed the conversion from RgbaVector to RgbaHalf in my app...

The Good :

Everything works.
Significant gain in memory footprint. (No speed up was expected.)
Conversion was easy.
Now able to handle 50 Mega-pixel images on my "old" computer.

The Bad :

Took me a while to test again everything

The Ugly :

Blame on Microsoft: Half is a half-baked type. No literals, missing intrinsic functions, need to cast and cast again... In many places I ended up converting half pixels in float pixels to execute most of my custom algorithms.