Layerz’
    MLP with depth-wise convolution, from [PVTv2: Improved Baselines with Pyramid Vision
    Transformer](https://arxiv.org/abs/2106.13797).
    Ú