Multiple "confidence" hits, which one wins?

kc8tmv

Getting the hang of it
Joined
May 27, 2017
Messages
177
Reaction score
82
Location
Cincinnati, Ohio
Question on the confidence level. How does DeepStack / BI decide what to burn into the image and the final "confidence level"?
Example below, "dark model" found People @ 75% and "objects" model found person @ 44%, yet only labeled image with "person:44%".
I would have thought that the HIGHEST confidence level object would "win" and be the one burned into the image. Not that this example would really make a difference, but it could depending on the objects. If it is 55% chance it is a dog and 95% change is it s bear, I want to know "BEAR"!

It would be nice it could have a new &MACRO of something like &SUMMARY (or something more clever maybe). Let &JSON give you everything (great for troubleshooting) and then &SUMMARY be JUST the object label that had the greatest confidence level (above your set threshold of course).

FYI, no idea where it found a chair going up my stairs in the example below!

[{"api":"objects","found":{"success":true,"predictions":
[{"confidence":0.4422658,"label":"person","y_min":1219,"x_min":33,"y_max":2302,"x_max":998},
{"confidence":0.5098691,"label":"chair","y_min":1525,"x_min":5,"y_max":2304,"x_max":1001}]}},
{"api":"faces","found":{"success":true,"predictions":[]}},
{"api":"dark","found":{"success":true,"predictions":[{"confidence":0.7568857,"label":"People","y_min":1196,"x_min":27,"y_max":2300,"x_max":978}]}}]
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,432
Reaction score
47,556
Location
USA
Mine shows both, although in many cases they are right on top of one another. Eventually I will tell it which model to use, but an offset would be nice to see them.
 
Top