分类： 论文

code:https://github.com/microsoft/Swin-Transformer

Swin MLP 代码来自 Swin Transformer 的官方实现。Swin Transformer 作者们在已有模型的基础上实现了 Swin MLP 模型，证明了 Window-based attention 对于 MLP 模型的有效性。

把张量 (B, H, W, C) 分成 window (B×H/M×W/M, M, M, C)，其中M是 window_size。这一步相当于得到 B×H/M×W/M 个大小为 (M, M, C) 的 window。

def window_partition(x, window_size):
    """
    Args:
        x: (B, H, W, C)
        window_size (int): window size

    Returns:
        windows: (num_windows*B, window_size, window_size, C)
    """
    B, H, W, C = x.shape
    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
    return windows

把 window (B×H/M×W/M, M, M, C) 变回张量 (B, H, W, C)。

def window_reverse(windows, window_size, H, W):
    """
    Args:
        windows: (num_windows*B, window_size, window_size, C)
        window_size (int): Window size
        H (int): Height of image
        W (int): Width of image

    Returns:
        x: (B, H, W, C)
    """
    B = int(windows.shape[0] / (H * W / window_size / window_size))
    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
    return x

一个 Swin MLP Block

class SwinMLPBlock(nn.Module):
    r""" Swin MLP Block.

    Args:
        dim (int): Number of input channels.
        input_resolution (tuple[int]): Input resolution.
        num_heads (int): Number of attention heads.
        window_size (int): Window size.
        shift_size (int): Shift size for SW-MSA.
        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
        drop (float, optional): Dropout rate. Default: 0.0
        drop_path (float, optional): Stochastic depth rate. Default: 0.0
        act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
    """

    def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0,
                 mlp_ratio=4., drop=0., drop_path=0.,
                 act_layer=nn.GELU, norm_layer=nn.LayerNorm):
        super().__init__()
        self.dim = dim
        self.input_resolution = input_resolution
        self.num_heads = num_heads
        self.window_size = window_size
        self.shift_size = shift_size
        self.mlp_ratio = mlp_ratio
        if min(self.input_resolution) <= self.window_size:
            # if window size is larger than input resolution, we don't partition windows
            self.shift_size = 0
            self.window_size = min(self.input_resolution)
        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"

        self.padding = [self.window_size - self.shift_size, self.shift_size,
                        self.window_size - self.shift_size, self.shift_size]  # P_l,P_r,P_t,P_b

        self.norm1 = norm_layer(dim)
        # use group convolution to implement multi-head MLP
        self.spatial_mlp = nn.Conv1d(self.num_heads * self.window_size ** 2,
                                     self.num_heads * self.window_size ** 2,
                                     kernel_size=1,
                                     groups=self.num_heads)

        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

    def forward(self, x):
        H, W = self.input_resolution
        B, L, C = x.shape
        assert L == H * W, "input feature has wrong size"

        shortcut = x
        x = self.norm1(x)
        x = x.view(B, H, W, C)

        # shift
        if self.shift_size > 0:
            P_l, P_r, P_t, P_b = self.padding
            shifted_x = F.pad(x, [0, 0, P_l, P_r, P_t, P_b], "constant", 0)
        else:
            shifted_x = x
        _, _H, _W, _ = shifted_x.shape

        # partition windows
        x_windows = window_partition(shifted_x, self.window_size)  # nW*B, window_size, window_size, C
        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)  # nW*B, window_size*window_size, C

        # Window/Shifted-Window Spatial MLP
        x_windows_heads = x_windows.view(-1, self.window_size * self.window_size, self.num_heads, C // self.num_heads)
        x_windows_heads = x_windows_heads.transpose(1, 2)  # nW*B, nH, window_size*window_size, C//nH
        x_windows_heads = x_windows_heads.reshape(-1, self.num_heads * self.window_size * self.window_size,
                                                  C // self.num_heads)
        spatial_mlp_windows = self.spatial_mlp(x_windows_heads)  # nW*B, nH*window_size*window_size, C//nH
        spatial_mlp_windows = spatial_mlp_windows.view(-1, self.num_heads, self.window_size * self.window_size,
                                                       C // self.num_heads).transpose(1, 2)
        spatial_mlp_windows = spatial_mlp_windows.reshape(-1, self.window_size * self.window_size, C)

        # merge windows
        spatial_mlp_windows = spatial_mlp_windows.reshape(-1, self.window_size, self.window_size, C)
        shifted_x = window_reverse(spatial_mlp_windows, self.window_size, _H, _W)  # B H' W' C

        # reverse shift
        if self.shift_size > 0:
            P_l, P_r, P_t, P_b = self.padding
            x = shifted_x[:, P_t:-P_b, P_l:-P_r, :].contiguous()
        else:
            x = shifted_x
        x = x.view(B, H * W, C)

        # FFN
        x = shortcut + self.drop_path(x)
        x = x + self.drop_path(self.mlp(self.norm2(x)))

        return x

    def extra_repr(self) -> str:
        return f"dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, " \
               f"window_size={self.window_size}, shift_size={self.shift_size}, mlp_ratio={self.mlp_ratio}"

注意 F.pad(x, [0, 0, P_l, P_r, P_t, P_b], “constant”, 0) 的对象是 x，维度是 (B, H, W, C)。
padding相当于是第3维 (C 这一维) 不填充，第2维 (W 这一维) 左右分别填充 P_l, P_r，第1维 (H 这一维) 左右分别填充 P_t, P_b。
x_windows = window_partition(shifted_x, self.window_size) # nW*B, window_size, window_size, C：
这句代码把 shifted_x 分成 nW*B 个 windows，其中每个 window 的维度是 (window_size, window_size, C)。

# reverse shift
if self.shift_size > 0:
P_l, P_r, P_t, P_b = self.padding
x = shifted_x[:, P_t:-P_b, P_l:-P_r, :].contiguous()
else:
x = shifted_x
这里是如果进行了 shift 操作，则最后取得结果也应该是没有 padding 的部分，正好是 shifted_x[:, P_t:-P_b, P_l:-P_r, :]。

一个 Swin MLP Block 的 FLOPs，注意 WSA 的计算量是：

FLOPs (WSA) = (window_size * window_size)^2 * dim * number_window

def flops(self):
        flops = 0
        H, W = self.input_resolution
        # norm1
        flops += self.dim * H * W

        # Window/Shifted-Window Spatial MLP
        if self.shift_size > 0:
            nW = (H / self.window_size + 1) * (W / self.window_size + 1)
        else:
            nW = H * W / self.window_size / self.window_size
        flops += nW * self.dim * (self.window_size * self.window_size) * (self.window_size * self.window_size)
        # mlp
        flops += 2 * H * W * self.dim * self.dim * self.mlp_ratio
        # norm2
        flops += self.dim * H * W
        return flops

每个 stage 之间的 PatchMerging连接，把 resolution 变为一半，dim 变为2倍。

class PatchMerging(nn.Module):
    r""" Patch Merging Layer.

    Args:
        input_resolution (tuple[int]): Resolution of input feature.
        dim (int): Number of input channels.
        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
    """

    def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
        super().__init__()
        self.input_resolution = input_resolution
        self.dim = dim
        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
        self.norm = norm_layer(4 * dim)

    def forward(self, x):
        """
        x: B, H*W, C
        """
        H, W = self.input_resolution
        B, L, C = x.shape
        assert L == H * W, "input feature has wrong size"
        assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even."

        x = x.view(B, H, W, C)

        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C

        x = self.norm(x)
        x = self.reduction(x)

        return x

    def flops(self):
        H, W = self.input_resolution
        # norm
        flops = H * W * self.dim
        # reduction
        flops += (H // 2) * (W // 2) * 4 * self.dim * 2 * self.dim
        return flops

• Patch Merging 操作把相邻的 2×2 个 tokens 给合并到一起，得到的 token 的维度是4C。
  Patch Merging 操作再通过一次线性变换把维度降为2C。

一个 Swin MLP Layer

class BasicLayer(nn.Module):
    """ A basic Swin MLP layer for one stage.

    Args:
        dim (int): Number of input channels.
        input_resolution (tuple[int]): Input resolution.
        depth (int): Number of blocks.
        num_heads (int): Number of attention heads.
        window_size (int): Local window size.
        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
        drop (float, optional): Dropout rate. Default: 0.0
        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
    """

    def __init__(self, dim, input_resolution, depth, num_heads, window_size,
                 mlp_ratio=4., drop=0., drop_path=0.,
                 norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False):

        super().__init__()
        self.dim = dim
        self.input_resolution = input_resolution
        self.depth = depth
        self.use_checkpoint = use_checkpoint

        # build blocks
        self.blocks = nn.ModuleList([
            SwinMLPBlock(dim=dim, input_resolution=input_resolution,
                         num_heads=num_heads, window_size=window_size,
                         shift_size=0 if (i % 2 == 0) else window_size // 2,
                         mlp_ratio=mlp_ratio,
                         drop=drop,
                         drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
                         norm_layer=norm_layer)
            for i in range(depth)])

        # patch merging layer
        if downsample is not None:
            self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer)
        else:
            self.downsample = None

    def forward(self, x):
        for blk in self.blocks:
            if self.use_checkpoint:
                x = checkpoint.checkpoint(blk, x)
            else:
                x = blk(x)
        if self.downsample is not None:
            x = self.downsample(x)
        return x

    def extra_repr(self) -> str:
        return f"dim={self.dim}, input_resolution={self.input_resolution}, depth={self.depth}"

    def flops(self):
        flops = 0
        for blk in self.blocks:
            flops += blk.flops()
        if self.downsample is not None:
            flops += self.downsample.flops()
        return flops

• 包含 depth 个 Swin MLP Block。
  注意计算 FLOPs 的方式：每个 blk 和 downsample 都自带 flops() 方法，可以直接来调用。

PatchEmbedded 操作

class PatchEmbed(nn.Module):
    r""" Image to Patch Embedding

    Args:
        img_size (int): Image size.  Default: 224.
        patch_size (int): Patch token size. Default: 4.
        in_chans (int): Number of input image channels. Default: 3.
        embed_dim (int): Number of linear projection output channels. Default: 96.
        norm_layer (nn.Module, optional): Normalization layer. Default: None
    """

    def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)
        patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]
        self.img_size = img_size
        self.patch_size = patch_size
        self.patches_resolution = patches_resolution
        self.num_patches = patches_resolution[0] * patches_resolution[1]

        self.in_chans = in_chans
        self.embed_dim = embed_dim

        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
        if norm_layer is not None:
            self.norm = norm_layer(embed_dim)
        else:
            self.norm = None

    def forward(self, x):
        B, C, H, W = x.shape
        # FIXME look at relaxing size constraints
        assert H == self.img_size[0] and W == self.img_size[1], \
            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
        x = self.proj(x).flatten(2).transpose(1, 2)  # B Ph*Pw C
        if self.norm is not None:
            x = self.norm(x)
        return x

    def flops(self):
        Ho, Wo = self.patches_resolution
        flops = Ho * Wo * self.embed_dim * self.in_chans * (self.patch_size[0] * self.patch_size[1])
        if self.norm is not None:
            flops += Ho * Wo * self.embed_dim
        return flops

• 和 ViT 的 Patch Embedded 操作一样，本质上是一个 K=patch size，s=patch size 的 nn.Conv2d 操作，注意卷积 FLOPs 的计算公式即可。

SwinMLP 整体模型架构

class SwinMLP(nn.Module):
    r""" Swin MLP

    Args:
        img_size (int | tuple(int)): Input image size. Default 224
        patch_size (int | tuple(int)): Patch size. Default: 4
        in_chans (int): Number of input image channels. Default: 3
        num_classes (int): Number of classes for classification head. Default: 1000
        embed_dim (int): Patch embedding dimension. Default: 96
        depths (tuple(int)): Depth of each Swin MLP layer.
        num_heads (tuple(int)): Number of attention heads in different layers.
        window_size (int): Window size. Default: 7
        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4
        drop_rate (float): Dropout rate. Default: 0
        drop_path_rate (float): Stochastic depth rate. Default: 0.1
        norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.
        ape (bool): If True, add absolute position embedding to the patch embedding. Default: False
        patch_norm (bool): If True, add normalization after patch embedding. Default: True
        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False
    """

    def __init__(self, img_size=224, patch_size=4, in_chans=3, num_classes=1000,
                 embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24],
                 window_size=7, mlp_ratio=4., drop_rate=0., drop_path_rate=0.1,
                 norm_layer=nn.LayerNorm, ape=False, patch_norm=True,
                 use_checkpoint=False, **kwargs):
        super().__init__()

        self.num_classes = num_classes
        self.num_layers = len(depths)
        self.embed_dim = embed_dim
        self.ape = ape
        self.patch_norm = patch_norm
        self.num_features = int(embed_dim * 2 ** (self.num_layers - 1))
        self.mlp_ratio = mlp_ratio

        # split image into non-overlapping patches
        self.patch_embed = PatchEmbed(
            img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim,
            norm_layer=norm_layer if self.patch_norm else None)
        num_patches = self.patch_embed.num_patches
        patches_resolution = self.patch_embed.patches_resolution
        self.patches_resolution = patches_resolution

        # absolute position embedding
        if self.ape:
            self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim))
            trunc_normal_(self.absolute_pos_embed, std=.02)

        self.pos_drop = nn.Dropout(p=drop_rate)

        # stochastic depth
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule

        # build layers
        self.layers = nn.ModuleList()
        for i_layer in range(self.num_layers):
            layer = BasicLayer(dim=int(embed_dim * 2 ** i_layer),
                               input_resolution=(patches_resolution[0] // (2 ** i_layer),
                                                 patches_resolution[1] // (2 ** i_layer)),
                               depth=depths[i_layer],
                               num_heads=num_heads[i_layer],
                               window_size=window_size,
                               mlp_ratio=self.mlp_ratio,
                               drop=drop_rate,
                               drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])],
                               norm_layer=norm_layer,
                               downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
                               use_checkpoint=use_checkpoint)
            self.layers.append(layer)

        self.norm = norm_layer(self.num_features)
        self.avgpool = nn.AdaptiveAvgPool1d(1)
        self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, (nn.Linear, nn.Conv1d)):
            trunc_normal_(m.weight, std=.02)
            if m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)

    @torch.jit.ignore
    def no_weight_decay(self):
        return {'absolute_pos_embed'}

    @torch.jit.ignore
    def no_weight_decay_keywords(self):
        return {'relative_position_bias_table'}

    def forward_features(self, x):
        x = self.patch_embed(x)
        if self.ape:
            x = x + self.absolute_pos_embed
        x = self.pos_drop(x)

        for layer in self.layers:
            x = layer(x)

        x = self.norm(x)  # B L C
        x = self.avgpool(x.transpose(1, 2))  # B C 1
        x = torch.flatten(x, 1)
        return x

    def forward(self, x):
        x = self.forward_features(x)
        x = self.head(x)
        return x

    def flops(self):
        flops = 0
        flops += self.patch_embed.flops()
        for i, layer in enumerate(self.layers):
            flops += layer.flops()
        # adaptive average pool
        flops += self.num_features * self.patches_resolution[0] * self.patches_resolution[1] // (2 ** self.num_layers)
        # head
        flops += self.num_features * self.num_classes
        return flops

• 由4个 Stage 组成，每个 Stage 由 BasicLayer 实现。
  传入的 depths 代表每个 Stage 的层数，比如 Swin-T 就是：[2, 2, 6, 2]。