This is the first (decent) result of a night-long time-lapse, including sunset and dawn.
2020 shots, from 19:30 to 06:45, with exposures ranging from 1/2000", 160 ISO to 20", 1500 ISO.
Here is the video: Night-long time-lapse 1
Better in MPEG1 format (abt 20MB): Night-long time-lapse 1
and below is the graph of the movie:
There are evident problems:
- flickers in brightness around frames 650 and 1350 (we are at the very end of the sensors' reliability) and 1200 (moving averages method is not good for this sort of smoothing)
- around frame 700, the exposure is kept constant for a while. (In the movie, you get a short period of pitch black sky). In that moment the "over-exposure protection" was somehow triggered. I have to study exactly why.
Did you ever figure out why the overexposure correction kicked in? Mine has done the same thing on a sunset. It likely would have before had I let it run until it was darker.
My thoughts are your values for h being < 10 may be unreasonable (or maybe I didn't have enough sky - maybe 2/3 of frame). My h values started at 60 (sun still bright) and were nearly a straight line, still being 55 when the protection kicked in long after the sun set. Then it eventually dropped to near 0. I thought I had added in Bruce Allen's overexposure compensation (L-25 & M+25) but it turns out I forgot to copy that version over so maybe my h values would be much lower.
On comparing the shots at the end with values of h near 10, they seem OK exposure-wise for my conditions (city lights). I kind of prefer the dark bluer sky though of 30 minutes earlier (and 70 minutes after the sun had set) when the correction kept the shutter constant at 4" and h had finally dropped to 10 versus the dark gray sky at the end when the shutter was 13". I know I could change the Tv limit and likely will for those time-lapses I do at home. The ambient light should be far less where I'll be this week so results should be better with what I had used.
I also note that the only black pic I've had occurred when T=0 so I have now added a check to set it to 1 if that happens again. Maybe a fluke, but maybe a bug in set_tv96_direct?
Peace
