You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Background: It would be more helpful if Antenna never shows low-confidence results (eg. "I am 0.01% certain this is a Spotted tussock moth"). Instead it's ideal if we keep rolling up to the next higher taxon rank until the confidence score is above a given threshold, until we can say "I am 60% certain this is in the family of Erebidae moths".
Find the method used in Fagner's evaluation code as a starting point.
If the calibrated softmax scores of a prediction are below a threshold, roll up all predictions for that image to the next taxon level, re-weight the scores and repeat until the top-1 prediction is above the threshold. Or use another method?
The inference API should return more in-depth data about each prediction.
Background: It would be more helpful if Antenna never shows low-confidence results (eg. "I am 0.01% certain this is a Spotted tussock moth"). Instead it's ideal if we keep rolling up to the next higher taxon rank until the confidence score is above a given threshold, until we can say "I am 60% certain this is in the family of Erebidae moths".
Find the method used in Fagner's evaluation code as a starting point.
If the calibrated softmax scores of a prediction are below a threshold, roll up all predictions for that image to the next taxon level, re-weight the scores and repeat until the top-1 prediction is above the threshold. Or use another method?
The inference API should return more in-depth data about each prediction.
Depends on #43
The text was updated successfully, but these errors were encountered: