cvpr2022的 mobileformer中用到了mlp多层感知机，就来学习一下

其实就是3个全连接层，前面两个加了bn，最后一层没有加bn。

import time

import torch
from torch import nn

class MLP(nn.Module):
    '''widths [in_channel, ..., out_channel], with ReLU within'''

    def __init__(self, widths, bn=True, p=0.5):
        super(MLP, self).__init__()
        self.widths = widths
        self.bn = bn
        self.p = p
        self.layers = []
        for n in range(len(self.widths) - 2):
            layer_ = nn.Sequential(nn.Linear(self.widths[n], self.widths[n + 1]).cuda(), nn.Dropout(p=self.p).cuda(), nn.ReLU6(inplace=True).cuda(), )
            self.layers.append(layer_)
        self.layers.append(nn.Sequential(nn.Linear(self.widths[-2], self.widths[-1]), nn.Dropout(p=self.p)))
        self.mlp = nn.Sequential(*self.layers).cuda()
        if self.bn:
            self.mlp