What do you mean box scores overfitting? The confidence of boxes? ... Because how do you know it's really the confidenece that overfitting and not the classification or dimensions? You need the other two as well.
@halsted what is the right format for submission,..cause some images have multiple bounding boxes...should we seperate them to diffent entries in the submission?
Yes - different rows. each row is a bounding box with a class and confidence. it's possible to have two classes in an image. an image can have no classes so the class is "NEG" (I put confidence = 1 and all 0's for dims like the samplesubmission). images can have any number of bounding boxes / rows. hope that helps. remember you need to have a row (at least NEG) for each IMage in the test.csv. good luck
there are several YOLO type models...
Thanks, what I meant was that the box scores are overfitting, even with yolov10
What do you mean box scores overfitting? The confidence of boxes? ... Because how do you know it's really the confidenece that overfitting and not the classification or dimensions? You need the other two as well.
@halsted what is the right format for submission,..cause some images have multiple bounding boxes...should we seperate them to diffent entries in the submission?
Yes - different rows. each row is a bounding box with a class and confidence. it's possible to have two classes in an image. an image can have no classes so the class is "NEG" (I put confidence = 1 and all 0's for dims like the samplesubmission). images can have any number of bounding boxes / rows. hope that helps. remember you need to have a row (at least NEG) for each IMage in the test.csv. good luck
and is there any threshold,..e.g for confidence?
Does this help: https://zindi.africa/competitions/lacuna-malaria-detection-challenge/discussions/22926
I didn't think confidence mattered (0 or 1 for NEG), but haven't tried both to test.