Message to Fujifilm: a Trivial Fix for Focus BKT Failure to Reach INF with GFX100 II/100S/100
Related: botched design, focus stacking, Fujifilm, Fujifilm GFX100 II, Fujifilm medium format, Fujifilm mirrorless, Glenn K, imcompetence crisis, medium format, please fix this, please fix this camera, please fix this Fujifilm
The engineers in the audience might appreciate this... the whole situation is ridiculous.
Dear Fujifilm engineers,
Your Focus BKT feature has been broken for 5+ years. Back in 2019, it was a complete joke with some lenses, so badly broken that one can only conclude that no field testing was ever done. Or if it were done, whoever did it was incompetent. And it was never tested again, seemingly.
Today, Focus BKT still does not work correctly, failing to achieve infinity focus virtually every time (to varying degrees), which I have observed repeatedly with every lens I have tried (and proven it in published articles moreover I have hundreds of proofs since it always happens).
I know how hard it is to get code written correctly since I’ve been a software engineer for 46 years, since I was 12 years old. Still, this situation is pathetic.
A plausible explanation
The behavior I see imagines one of the most time-honored software bugs that has ever existed: some engineer forgot that assigning a floating point number to an integer will round down by default. It’s possible that this is not the explanation and the answer is more complex, but all behavior that I observe is totally consistent with that idea.
The errant code would look something like this:
float fc = calcFrameCount(...);
int numFramesToShoot = (int)fc; <=== 9.1, 9.3, 9.7, 9.99, etc all become 9 frames
Such a coding error would explain perfectly why the amount of error (the degree of softness/defocus of the last frame) varies from one focus stack to another.
A 12-year-old engineer (me in my boyhood) might make this mistake, but a major corporation with professional engineers? Any code review by peers* should catch it as a novice mistake.
* If the software engineering best practice of code review by fellow engineers doing is even implemented?
Fixing the issue with a kludge
So, Fujifilm, I offer you my at-a-glance and guaranteed-to-work fix, at least for all the failure cases I have observed:
// mask the bug; shoot one more frame than the bad code calculates
numFramesToShoot = numFramesToShoot + 1;
There you have it. It is zero risk and code review should take all of 30 seconds. Please remit a check for US$10K once the fix is implemented. And send me a Fujifilm GFX100 II with fixed firmware as compensation for the enormous and ongoing hassle and workload this failure has inflicted upon not just me, but your entire user base.
Glenn K writes:
cc: Sony
DIGLLOYD: yep.
