Does CodeAi work on thermal images?

E

erkme73

BIT Beta Team
Nov 9, 2014
1,618
1,538
I've finally started using the AI again on a few cameras. For the daytime optical it seems to be pretty accurate - or accurate enough. However, on my thermal cameras AI is discarding nearly everything as "nothing found" despite very obvious captures of animals.

Is there a different model that I could use that would be more forgiving for B/W thermal images?

1758801367327.png1758801385690.png

It identifies coyotes as boats, birds or kites. Skunks as people (with 40% confidence), and the like...
 
It would be the same issue we experience with B/W images. CodeProject works best with color images.

Doesn't mean in the right situations is can't work, but I think there are enough differences that one would probably need to train a model for thermal images.
 
wittaj said:
It would be the same issue we experience with B/W images. CodeProject works best with color images.

Doesn't mean in the right situations is can't work, but I think there are enough differences that one would probably need to train a model for thermal images.
Click to expand...

So how does one train a model? If I go through and mark "nothing found" as "confirmed" in BI, does that "teach" the model? I wouldn't think so, since I can't correct it to tell it what it really way, right?
 
You must log in or register to reply here.
Top Bottom