Question on hybrid ref pt 2d

[george0407]

        ref_2d_hw = self.ref_2d_hw.clone().expand(bs, -1, -1, -1)
        hybird_ref_2d = torch.cat([ref_2d_hw, ref_2d_hw], 0)

Firstly appreciate your work, and I wonder why you concatenate the ref pts of hw for 2 layers as shown above. Thanks a lot.

[huang-yh]

Hi, the code is built upon BEVFormer. BEVFormer has both temporal version and spatial-only version. BEVFormer concatenates the current ref_2d and the aligned ref_2d_shift from another timestamp in its temporal version, while concatenating two same ref_2ds for the spatial-only version. Since TPVFormer does not use temporal information, we simply follow the practice of BEVFormer.

[george0407]

Thanks for your answer