BEVFormer code: num_heads and num_points are swapped

[vniclas]

Hi,
thank you for the nice work and for sharing your code!

I believe that your implementation of BEVFormer has a small bug:

simple_bev/nets/bevformernet.py

Line 296 in be46f0e

self.deformable_attention = MSDeformAttn(d_model=dim, n_levels=1, n_heads=4, n_points=8)

It looks like the values for the parameters n_heads and n_points have been swapped compared to the normal initialization

simple_bev/nets/ops/modules/ms_deform_attn.py

Line 31 in be46f0e

def __init__(self, d_model=256, n_levels=4, n_heads=8, n_points=4):

See also the original implementation of BEVFormer:
def __init__(self, embed_dims=256, num_heads=8, num_levels=4, num_points=4,
https://github.com/fundamentalvision/BEVFormer/blob/20923e66aa26a906ba8d21477c238567fa6285e9/projects/mmdet3d_plugin/bevformer/modules/decoder.py#L160-L164

as well as the Deformable DETR paper:

M = 8 and K = 4 are set for deformable attentions by default.
K number of sampled keys in each feature level for each attention head
M number of attention heads

I am not sure how much of a difference it is going to make but just to warn other people.

[aharley]

Wow thanks. Good catch.