Closed jianlong-yuan closed 5 years ago
We annotated all the bounding-boxes. But because mask & key-point annotation is a quite time-consuming job, a few instances have no such annotation. (It doesn't affect the evaluation.) Our visualisation code shows instances with full annotations here.
We annotated all the bounding-boxes. But because mask & key-point annotation is a quite time-consuming job, a few instances have no such annotation. (It doesn't affect the evaluation.) Our visualisation code shows instances with full annotations here.
hello, I'm wondering how you've obtained evaluation with the face that some instances not annoted? like for example you've got predictions of two instances for the image above but we have only one gt instance.
what's the logic to match specific prediction instance with its according gt instance? Something like bounding box closeness or what?
The coco evaluation toolbox can tolerate case like this.
I can't remember precisely but the logic is something like for each gt, you find a prediction that has max iou with it, and consider it as a match.
Why only some people are marked, others in a picture are not marked?