Hello everyone. I work on TrackingNet dataset and encountered that there are lots of frames whose bounding box annotations are incorrect (either they don't cover the object on the whole or the coordinates are out of the image). I wonder, is there a solution?
Note that there are around 14.5M images and around 40% of all annotations are inconsistent according to my very rough estimate.
Hello everyone. I work on TrackingNet dataset and encountered that there are lots of frames whose bounding box annotations are incorrect (either they don't cover the object on the whole or the coordinates are out of the image). I wonder, is there a solution? Note that there are around 14.5M images and around 40% of all annotations are inconsistent according to my very rough estimate.