MohsenZand / ObjectBox

(ECCV 22 Oral) ObjectBox: From Centers to Boxes for Anchor-Free Object Detection
GNU General Public License v3.0
131 stars 16 forks source link

# det_layers #13

Closed 1358422654 closed 1 year ago

1358422654 commented 1 year ago

det_layers

det_layers:

For P3,P4 and P5, why are they the same size for objects of different sizes? Can I directly replace it with anchors below? Will it affect the accuracy and speed?

anchors:

MohsenZand commented 1 year ago

They are constant scale gains, as explained on page 7 of the paper. They cannot be replaced by anchors, since they are interpreted differently.

1358422654 commented 1 year ago

Thank you very much for your reply。