i wrote a little primer/basic intro to HDR images. Hopefully it clears up a few points for you.
Nice writeup. Maybe you could consider talking a bit about how to edit the HDR image once you’ve saved it. What software do you use for that? Also, is saving a HDR image in Blender as easy as just choosing the right image format, or are there other steps involved?
Thanks. Get out your spare hard drives folks, we’re gonna be chewing some bytes in the future. 3x32!!! wow. so like an uncompressed 640x480 image would take…<calculator cliking, pauses>…nah, that cant be right…<calc clicking agian> … 3.5 Megabytes…!!! and an HD 1080p takes…190Meg, which means a 60 second HD commercial, at US 30 fps, would need 333 Gig. Per layer which if your are compositing and have two inputs and one output, you need a FREAKING terabyte??? someone pinch me…
for some reason actual files aren’t as large as your calculation, (RLE compression, maybe?) and some times there is no need to use the full 32bit. Elephants Dream used 10bit EXR at 1080p HD resolution.
.hdr (radiance) is not 96bit (or 32bit per channel), not even nearly. It’s 32bit RGBE, that is 32bit in total per pixel. It supports RLE compression, but RLE will not work well on noisy images (real photographs, unbiased GI renderings etc.). It’s an ancient format, neither very precise nor very flexible.
There also doesn’t exist a 10bit EXR format, it’s either 16bit (“half”) or 32bit float or 32bit integer per channel. I don’t remember what exactly Elephants Dream used, but the amount of data was huge (or still is actually)…
OpenEXR does support several lossless compression algorithms.
Saving HDRIs from the latest versions of blender is indeed as simple as choosing the correct image format. Editing EXRs is not widely supported though, Adobe Photoshop got some tools over the last versions, apart from expensive specialized tools i only know rather rudimentary tools like HDR-shop 1.0, Cinepaint and a couple of command-line tools (pfstools+pfstmo for example).
hmm, thanks for the correction lynx3d, i always thought hdr/exr is 32bit (or at least 16bit) times 3 since that’s what photoshop report it as… I’ll do more research before i post then. btw what’s the ‘E’ in RGB’E’, i assume it’s some sort of luma channel?
Ah yes once you loaded it into photoshop or cinepaint, it gets converted to 32bit float per channel so you can edit it properly…
The “E” stands for exponent, you could think of it as some sort of “luma” information, in reality it’s just a shared exponent for R, G and B, so the 8bit channel information can be converted back to float values.
So unlike a “half” EXR image, where each component has its own exponent to describe the float value, the components of a Radiance image cannot differ very much unless set to zero.
Also, the “half”-type is much more sophisticated, and basically has all the features of “real” (32bit) floats, like support for negative values, infinity, NaN etc.