File size: 5,615 Bytes
2cd560a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 |
# --------------------------------------------------------
# High Resolution Transformer
# Copyright (c) 2021 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Rao Fu, RainbowSecret
# --------------------------------------------------------
import torch
import torch.nn as nn
class Mlp(nn.Module):
def __init__(
self,
in_features,
hidden_features=None,
out_features=None,
act_layer=nn.GELU,
drop=0.0,
):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Linear(in_features, hidden_features)
self.act = act_layer()
self.fc2 = nn.Linear(hidden_features, out_features)
self.drop = nn.Dropout(drop)
def forward(self, x, H, W):
x = self.fc1(x)
x = self.act(x)
x = self.drop(x)
x = self.fc2(x)
x = self.drop(x)
return x
class MlpDW(nn.Module):
def __init__(
self,
in_features,
hidden_features=None,
out_features=None,
act_layer=nn.GELU,
dw_act_layer=nn.GELU,
drop=0.0,
):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Conv2d(in_features, hidden_features, kernel_size=1)
self.act1 = act_layer()
self.dw3x3 = nn.Conv2d(
hidden_features,
hidden_features,
kernel_size=3,
stride=1,
groups=hidden_features,
padding=1,
)
self.act2 = dw_act_layer()
self.fc2 = nn.Conv2d(hidden_features, out_features, kernel_size=1)
self.drop = nn.Dropout(drop)
def forward(self, x, H, W):
B, N, C = x.shape
if N == (H * W + 1):
cls_tokens = x[:, 0, :]
x_ = x[:, 1:, :].permute(0, 2, 1).contiguous().reshape(B, C, H, W)
else:
x_ = x.permute(0, 2, 1).contiguous().reshape(B, C, H, W)
x_ = self.fc1(x_)
x_ = self.act1(x_)
x_ = self.dw3x3(x_)
x_ = self.act2(x_)
x_ = self.drop(x_)
x_ = self.fc2(x_)
x_ = self.drop(x_)
x_ = x_.reshape(B, C, -1).permute(0, 2, 1).contiguous()
if N == (H * W + 1):
x = torch.cat((cls_tokens.unsqueeze(1), x_), dim=1)
else:
x = x_
return x
class MlpDWBN(nn.Module):
def __init__(
self,
in_features,
hidden_features=None,
out_features=None,
act_layer=nn.GELU,
dw_act_layer=nn.GELU,
drop=0.0,
):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Conv2d(in_features, hidden_features, kernel_size=1)
self.act1 = act_layer()
self.norm1 = nn.BatchNorm2d(hidden_features)
self.dw3x3 = nn.Conv2d(
hidden_features,
hidden_features,
kernel_size=3,
stride=1,
groups=hidden_features,
padding=1,
)
self.act2 = dw_act_layer()
self.norm2 = nn.BatchNorm2d(hidden_features)
self.fc2 = nn.Conv2d(hidden_features, out_features, kernel_size=1)
self.act3 = act_layer()
self.norm3 = nn.BatchNorm2d(out_features)
self.drop = nn.Dropout(drop)
def forward(self, x, H, W):
B, N, C = x.shape
if N == (H * W + 1):
cls_tokens = x[:, 0, :]
x_ = x[:, 1:, :].permute(0, 2, 1).contiguous().reshape(B, C, H, W)
else:
x_ = x.permute(0, 2, 1).contiguous().reshape(B, C, H, W)
x_ = self.fc1(x_)
x_ = self.norm1(x_)
x_ = self.act1(x_)
x_ = self.dw3x3(x_)
x_ = self.norm2(x_)
x_ = self.act2(x_)
x_ = self.drop(x_)
x_ = self.fc2(x_)
x_ = self.norm3(x_)
x_ = self.act3(x_)
x_ = self.drop(x_)
x_ = x_.reshape(B, C, -1).permute(0, 2, 1).contiguous()
if N == (H * W + 1):
x = torch.cat((cls_tokens.unsqueeze(1), x_), dim=1)
else:
x = x_
return x
class MlpDWBN2D(nn.Module):
def __init__(
self,
in_features,
hidden_features=None,
out_features=None,
act_layer=nn.GELU,
dw_act_layer=nn.GELU,
drop=0.0,
):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Conv2d(in_features, hidden_features, kernel_size=1)
self.act1 = act_layer()
self.norm1 = nn.BatchNorm2d(hidden_features)
self.dw3x3 = nn.Conv2d(
hidden_features,
hidden_features,
kernel_size=3,
stride=1,
groups=hidden_features,
padding=1,
)
self.act2 = dw_act_layer()
self.norm2 = nn.BatchNorm2d(hidden_features)
self.fc2 = nn.Conv2d(hidden_features, out_features, kernel_size=1)
self.act3 = act_layer()
self.norm3 = nn.BatchNorm2d(out_features)
self.drop = nn.Dropout(drop)
def forward(self, x):
x = self.fc1(x)
x = self.norm1(x)
x = self.act1(x)
x = self.dw3x3(x)
x = self.norm2(x)
x = self.act2(x)
x = self.drop(x)
x = self.fc2(x)
x = self.norm3(x)
x = self.act3(x)
x = self.drop(x)
return x
|