Why isn't Deepstack correctly identifying cars?

As I have said many times, DeepStack, and especially night time in B/W, struggles with trying to detect images on the edges of the field of view.

You can try running it in color or if B/W then play with brightness and contrast so that there is more difference.
 
The really odd part of the previous .dat analysis is the it shows the BBQ (identified as chair) as an object in motion. Huh? But I'm not all that surprised, as I have seen similar oddities from DeepStack.
In daytime, the combined.pt works quite well, and with reasonable analysis times (even on CPU only, no NVIDIA card). The dark.pt is just a bit of a monster in terms of processing time for a CPU only setup, and with so-so results for B/W IR only scenes. I eagerly await news from @MikeLud1 on a new version of dark.pt, supposedly in the works. (Edit: he just updated his timeline for that this evening, it is looking like he hopes to release a new version by the end of this month).
 
Last edited:
Yes, the final "result" status shows as occupied, but the two colored dots next to some of the individual frame analyses/ identified objects are indicative of motion. Of course, sometimes all it takes for "motion" to be picked up by BI is a change in lighting, which is the likely explanation here.
 
Last edited: