Re: [Openexr-devel] DWA compression

[Karl Rasche]

I *think* I described how that parameter is working at the giant comment in ImfDwaCompressor.cpp. If it's vague that can probably be remedied.

The standard jpeg 1-100 sort of range is a bit annoying, because there is large swath of the range that isn't terribly useful (like < 50), and not a ton of control where high quality image folks might want there to be (> 90). 

If I was going to hand wave about building a slider for DWAx, I'd probably put one of end of the slider at 45, and then debate over where to put the other end. There's a range of values that is probably good for production images (maybe < 150) and a range of values that's probably useful for HDR vacation snaps (< 500, lets say). For something rather arbitrary, how about 2000 for the far end. 

Then, to give a little more control for the production folks, I might pick a third point that's at the top end of what could be useful for production. Use those three points for fit a quadratic.

Maybe:

   1.0  -->   45

   0.75 -->  125

   0.0  --> 2000

I doubt that a linear mapping would be terribly satisfying for control, but I could be wrong.

Karl

On Sun, Aug 17, 2014 at 10:18 AM, Brendan Bolles <address@hidden> wrote:

On Aug 16, 2014, at 11:35 PM, Sean Y Chen wrote:

> In Brendan Bolles's demo video, it seems that 100000 is the limit for a couple of the images, though one of them goes to 250000 before reaching file size 1.

Here's the code for the little program I wrote to generate those sequences.  I'm basically doing compression = pow(10.0, frame / 25.0), with frames going from 0-149.  I wouldn't say it's perceptually linear exactly, although if you look at the MtTamWest example where I show difference mode, the diff seems to come in pretty linearly.

Karl's has a nice write-up on the algorithm at the top of ImfDwaCompressor.cpp.