(self,
dim,
num_heads,
window_size=7,
shift_size=0,
mlp_ratio=4.,
qkv_bias=True,
qk_scale=None,
drop=0.,
attn_drop=0.,
drop_path=0.,
act_layer=nn.GELU,
norm_layer=nn.LayerNorm)
| 186 | norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm |
| 187 | """ |
| 188 | def __init__(self, |
| 189 | dim, |
| 190 | num_heads, |
| 191 | window_size=7, |
| 192 | shift_size=0, |
| 193 | mlp_ratio=4., |
| 194 | qkv_bias=True, |
| 195 | qk_scale=None, |
| 196 | drop=0., |
| 197 | attn_drop=0., |
| 198 | drop_path=0., |
| 199 | act_layer=nn.GELU, |
| 200 | norm_layer=nn.LayerNorm): |
| 201 | super().__init__() |
| 202 | self.dim = dim |
| 203 | self.num_heads = num_heads |
| 204 | self.window_size = window_size |
| 205 | self.shift_size = shift_size |
| 206 | self.mlp_ratio = mlp_ratio |
| 207 | assert 0 <= self.shift_size < self.window_size, 'shift_size must in 0-window_size' |
| 208 | |
| 209 | self.norm1 = norm_layer(dim) |
| 210 | self.attn = WindowAttention(dim, |
| 211 | window_size=to_2tuple(self.window_size), |
| 212 | num_heads=num_heads, |
| 213 | qkv_bias=qkv_bias, |
| 214 | qk_scale=qk_scale, |
| 215 | attn_drop=attn_drop, |
| 216 | proj_drop=drop) |
| 217 | |
| 218 | self.drop_path = DropPath( |
| 219 | drop_path) if drop_path > 0. else nn.Identity() |
| 220 | self.norm2 = norm_layer(dim) |
| 221 | mlp_hidden_dim = int(dim * mlp_ratio) |
| 222 | self.mlp = Mlp(in_features=dim, |
| 223 | hidden_features=mlp_hidden_dim, |
| 224 | act_layer=act_layer, |
| 225 | drop=drop) |
| 226 | |
| 227 | self.H = None |
| 228 | self.W = None |
| 229 | |
| 230 | def forward(self, x, mask_matrix): |
| 231 | """Forward function. |
nothing calls this directly
no test coverage detected