-
Notifications
You must be signed in to change notification settings - Fork 50
Open
Description
Hello author, thank you for your hard work!
Drawing on your ideas, I am developing my own single-stage object detection network, but there are a lot of extra boxes in the visualization results.
Since my network structure is deep, I guess it may be caused by the disappearance of the gradient or the explosion of the gradient, but why do these two situations lead to the appearance of a large number of abnormal boxes, can you give any suggestions?

Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels