In [4] we propose group-convolution layers that are equivariant (the ouput transforms accordingly) with respect to rotations and translations of the input.
確定! 回上一頁