microsoft / RepPoints

Represent Visual Objects by Point Sets
MIT License
589 stars 75 forks source link

[code question] difference between pseudo bbox and final bbox #10

Open shwoo93 opened 5 years ago

shwoo93 commented 5 years ago

Hello,

I have a question about the code.

To make final boxes, the predicted points go through two steps, 1) offset_to_pts function and 2) points2bbox function. https://github.com/microsoft/RepPoints/blob/ee07b3fd1b00e2e1bbcc9116c8d56c2b67ce6536/src/reppoints_head/reppoints_head.py#L424 https://github.com/microsoft/RepPoints/blob/ee07b3fd1b00e2e1bbcc9116c8d56c2b67ce6536/src/reppoints_head/reppoints_head.py#L389

On the other hand, for the pseudo box generation, the predicted points are directly passed to the points2bbox function and the generated boxes are used for matching. https://github.com/microsoft/RepPoints/blob/ee07b3fd1b00e2e1bbcc9116c8d56c2b67ce6536/src/reppoints_head/reppoints_head.py#L456 https://github.com/microsoft/RepPoints/blob/ee07b3fd1b00e2e1bbcc9116c8d56c2b67ce6536/src/reppoints_head/reppoints_head.py#L459

Why do they differ? shouldn't we use the same procedure (i.e., offset_to_pts and then points2bbox)?