I downloaded the current version of hdrgen and ran it on the same images with same parameters than for an older version (around 6 years old). I could not find a command-line option to find out a version number, so I cannot say which versions I used…
The result I found is that the generated images differ significantly (e.g. in terms of calculated Ev for an fish-eye around 12%, if images are not scaled by a calibration factor which would then eliminate the difference of course), and the newer image looks slightly more “yellowish” than the old version. The header in the generated files are exactly the same. The file size of the hdr actually also differ (old version 27MB, new version 33MB).
So my question is, was there a change in default settings or in the code that cause this? Is the newer version “better” or let’s say more accurate than the old one?
If you don’t save and re-use the computed camera response function via the -r option, then random sampling can lead hdrgen to choose slightly different parts of the image for its calibration, which can lead to differences. It can even be the same source code, just a different random number generator from the library or operating system.
There have of course been small changes to the code over the years that can affect the output as well. I confess that I have not kept close track or bothered with versioning, which is lazy practice. The most significant change I made to this code in the past few years was done to improve noise characteristics, but I don’t think this would have much affect on value calibration. Regarding noise, the newer version of hdrgen is preferred. In terms of absolute accuracy, I cannot say.
Generally speaking, I recommend using the raw2hdr script whenever possible for consistent results, since it relies on the linear nature of raw encodings to avoid response function and color-mapping differences.