Hi, great work, I have a small detailed question, the MIN_SIZE_TEST is set to 640, so in fact the shortest edge of test images are larger or equal than 640, but then RandomCrop will randomly crop a 640 x 640 patch, while there is still a certain portion of test images are discarded, so as the annotations, does this influence the test mAP? Cause a portion of annotations are threw away and the model does not evaluate on these annotations.
Hi, great work, I have a small detailed question, the MIN_SIZE_TEST is set to 640, so in fact the shortest edge of test images are larger or equal than 640, but then RandomCrop will randomly crop a 640 x 640 patch, while there is still a certain portion of test images are discarded, so as the annotations, does this influence the test mAP? Cause a portion of annotations are threw away and the model does not evaluate on these annotations.