How to mask invalid reference point when doing deformable attention?

fundamentalvision / Deformable-DETR

Deformable DETR: Deformable Transformers for End-to-End Object Detection.

Apache License 2.0

3.15k stars 513 forks source link

How to mask invalid reference point when doing deformable attention? #147

Open kaixinbear opened 2 years ago

kaixinbear commented 2 years ago

Thanks for your great work ! I want to extend the deform detr for 3D detection task. Just like DETR3D vs DETR. But when I implement this, I find that projecting 3D reference point generated by object query may out of the image range ! This means that when I use deform attention API, I need to mask some invalid reference point. How could I achieve this ?