Thank you so much for sharing your interesting work.
I have a question regarding the NMS process for 3D bounding boxes. In line 325 of EPro-PnP-v2/EPro-PnP-Det_v2/epropnp_det/core/bbox_3d/misc.py, the keep_inds are specified by the following code when n (x3d.size(0)) <= 1:
keep_inds = bbox_3d.new_zeros(0, dtype=torch.int64)
However, bbox_3d.new_zeros(0, dtype=torch.int64) returns an empty tensor, resulting in an empty bbox_3d_out. I'm wondering if this is the expected result. It seems to me that this code always outputs nothing if there is only one object in an image.
Could you please advise me on this part?