Merge to HDRI - why does the result look 'tonemapped'?

Hi,

I’ve been creating hdri images for a long time already, but one thing has always bothered me. I guess my main questions is this:
Why does a HDRI (merged from a set of jpg files) viewed on a computer screen at a given exposure, looks different to the original jpg file at that same exposure?

So for example you create a hdri from 9 exposures where number 7 is ‘well’ exposed. You create the hdri and then you set the base exposure for it to be as in image n°7. When you compare the hdri with the original n°7 image, it looks nothing like it. The hdri looks washed out, dark areas are lit up, and light areas are darker. It looks like it has tonemapping applied to it already.

I made a small test to compare different programs that can merge jpg files to hdri’s. Here is an image showing the results. I tried to show all files at very similar exposures. Also note the histograms.

Here’s an animated gif:
http://www.aversis.be/vrayrhino/hdri.gif

What surprises me is that the godfather of all hdri programs creates the most washed out image. It is very similar to photoshop’s result though. Then comes PTgui, I made a stitch directly from the bracketed sets of jpg’s and rendered it to hdri (so ptgui does the hdr merging). Photomatix produces the most saturated and contrasty result. But it is still different to the original LDR stitch.

My concern is, that when you would use a hdri as your rendered background, it will have this washed out look. I am in the process of creating very high res hdri files, and would like to get it right so that they are usable as rendered backgrounds too.

I know I can adjust the merged hdri in photoshop to give it more contrast, but I tried stuff like that in the past, and noticed it can have drastic results on lighting in the 3d scene.

Any thoughts would be greatly appreciated :slight_smile:

Would this have anything to do with srgb or adobe rgb profiles in jpg files? Is it just a gamma issue? Why do different programs produce such a different result?

Thanks,

wouter

Here is a render with the ptgui hdri as light/reflection source and as background image:

Here is a render with the ptgui hdri as light/reflection source and the jpg stitched version as background image:


Maybe the problem is that if you use the jpg files from your camera they already have a pretty strong color correction. The cameras are sharpening them and raise the contrast.

If you use a raw file from your camera and while developing it turn down all color corrections you will see that it originally looks pretty flat.

My general workflow is to shoot in raw. Develop those as flat 16 bit tif files and then merge those as hdri. This way I have the greatest possible information in my images.

After the stitch I do my adjustments on the HDRI panoramas.

Well i guess this has to do with the camera response curved used during merging to HDR. This can drastically change the results, and each
vendor cooks its own i guess.

I think it comes from that programs only use the well exposed portions of the images for merging and not the high contrast dark and bright parts so they somehow compensate the contrast that was introduced during creating the jpg images.