## key point

trying to attach attention module to CNN. but instead of blindly attaching it which would compute a 3D attention map which is computationally expensive, this work proposes to compute spatial attention and channel attention separately which achieves a similar effect with much less parameters.

(more…)