With the originally released code, the number of the predicted bounding boxes is determined by each image. Is there a method that we could make sure the number of bboxes is consistent for all images? Currently, I am looking at the roi_heads.py and rpn.py but I got no luck.
With the originally released code, the number of the predicted bounding boxes is determined by each image. Is there a method that we could make sure the number of bboxes is consistent for all images? Currently, I am looking at the
roi_heads.py
andrpn.py
but I got no luck.