librbox.cpp

[dps42]

Hi. Thanks for sharing the code.

1. Is librbox.cpp that you used to create librbox.so exactly the same as the one in https://github.com/liulei01/DRBox/blob/master/examples/rbox/deploy/librbox.cpp.code ? Or was there any modification? Due to weird IT policy, I only have access to a GPU on a Windows machine. I checked the code on a slow Linux machine, and the librbox.so seemed to work. But when I tried to port it to Windows using Python's ctypes LoadLibrary() after creating a Windows DLL using Visual Studio, the code showed weird and different inference results. So I'd like to know whether the librbox.cpp is exactly the same.

2. Could you briefly explain how OverlapSum() works? In the code, I saw things like line cross point, inner point, etc. but I couldn't find much information from the paper.

3. I couldn't understand the statement in the paper: "In the test stage, the test images are divided
   into image chips with the size of 300×300 pixels and adjacent chips have an overlap area of 15% to ensure that the targets at the edge of each chip will not be missed." Where in the code can I find this process? Is this actually used when the input image is 300 x 300? Then how do we divide the 300x300 image into 300x300 image chips?

Thank you.