Currently animal_presence assumes Positive means animal present, and negative means animal not present.
This is actually the opposite of what we are interested in for this test... and makes the Precision and Recall less useful.
We want to achieve a precision which is "percentage of images identified as blank which are indeed blank" and recall of "percentage of images that are blank that are correctly identified as such".
To do this, we should rename the test to emptiness_detection, and flip its definitions of positive and negative.
for the test validation doc, we will just use animal_count_0 instead. But we need to fix this for the demo.
Currently animal_presence assumes Positive means animal present, and negative means animal not present.
This is actually the opposite of what we are interested in for this test... and makes the Precision and Recall less useful. We want to achieve a precision which is "percentage of images identified as blank which are indeed blank" and recall of "percentage of images that are blank that are correctly identified as such".
To do this, we should rename the test to emptiness_detection, and flip its definitions of positive and negative.
for the test validation doc, we will just use animal_count_0 instead. But we need to fix this for the demo.