In the traditional rotated object detector, In order to ensure consistent classification and localization of objects in different scales, the output features are usually changed to 7*7 size during the rotated roi operation, but I want to extract the corresponding features according to the real size of the object. How can I handle this?
In the traditional rotated object detector, In order to ensure consistent classification and localization of objects in different scales, the output features are usually changed to 7*7 size during the rotated roi operation, but I want to extract the corresponding features according to the real size of the object. How can I handle this?