[Segmentation] 05. swin 코드 리뷰

이전에 리뷰한 Swin Transformer 논문을 코드와 함께 input/output 사이즈를 중심으로 다시 리뷰해 봅시다.

Official Github -> Swin-Transformer/models/swin_transformer.py

Swin Architecture

SwinTransformer

class SwinTransformer(nn.Module):
    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., qkv_bias=True, qk_scale=None,
                 drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1,
                 norm_layer=nn.LayerNorm, ape=False, patch_norm=True,
                 use_checkpoint=False, fused_window_process=False, **kwargs):
        super().__init__()

        # ...

        # 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)
        
        # ...

        # 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,
                               qkv_bias=qkv_bias, qk_scale=qk_scale,
                               drop=drop_rate, attn_drop=attn_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,
                               fused_window_process=fused_window_process)
            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 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

    # ...

코드를 보면 전체적인 흐름은 다음과 같습니다:

PatchEmbed -> BasicLayer -> norm & pooling -> Linear

이중에서 PatchEmbed 모듈은 맨 앞의 Patch Partition과 Linear Embedding block역할이고, BasicLayer 모듈은 stage 1~4의 기능을 하는 것으로 보입니다.

요 두 모듈을 중점적으로 리뷰해 봅시다.

PatchEmbed

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

    # ...

input 이미지를 patch_size로 나누고, 각 패치를 linear embedding해줍니다. 디폴트 값은 image.shape = (3,224,224), patch_size = 4, embedding_dim = 96입니다.

처음 input으로 들어오는 x는 B개의 이미지 입니다. (x.shape = (B, 3,244,244)). 위의 코드에서는 convolution layer을 이용해 patch partition과 linear embedding을 한번에 진행합니다. kernel_size = patch_size = 4, output_dim = embedding_dim로 설정해 각 convolution 연산마다 4x4x3 크기의 subimage를 입력으로 받아 크기 96의 1d embedding vector을 출력으로 얻습니다. 이때, stride = patch_size = 4로 설정하여 각각의 receptive field가 겹치지 않도록 합니다.
-> shape = (B, 96, 224/4, 224/4) = (B, 96, 56, 56)

이후에 flatten(2).transpose(1,2)를 통해 output의 shape은 (B, 56x56, 96)이 됩니다.

flatten(n): n번째 dim부터 모두 flatten
transpose(n, m): n번째 dim과 m번째 dim 순서 변경

BasicLayer

class BasicLayer(nn.Module):
    """ A basic Swin Transformer 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.
        #... 
    """

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

        super().__init__()
        # ... 

        # build blocks
        self.blocks = nn.ModuleList([
            SwinTransformerBlock(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,
                                 qkv_bias=qkv_bias, qk_scale=qk_scale,
                                 drop=drop, attn_drop=attn_drop,
                                 drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
                                 norm_layer=norm_layer,
                                 fused_window_process=fused_window_process)
            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

    # ... 

depth개의 SwinTransformerBlock과 downsample로 구성되어 있습니다. downsample은 각 stage의 끝에 있는 patch merging 과정입니다.

SwinTransformerBlock

class SwinTransformerBlock(nn.Module):
    r""" Swin Transformer Block.

    Args:
        dim (int): Number of input channels.
        input_resolution (tuple[int]): Input resulotion.
        num_heads (int): Number of attention heads.
        window_size (int): Window size.
        shift_size (int): Shift size for SW-MSA.
    """

    def __init__(self, dim, input_resolution, 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,
                 fused_window_process=False):
        super().__init__()

        # ...

        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"

        self.norm1 = norm_layer(dim)
        self.attn = WindowAttention(
            dim, window_size=to_2tuple(self.window_size), num_heads=num_heads,
            qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)

        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)

        if self.shift_size > 0:
            # calculate attention mask for SW-MSA
            H, W = self.input_resolution
            img_mask = torch.zeros((1, H, W, 1))  # 1 H W 1
            h_slices = (slice(0, -self.window_size),
                        slice(-self.window_size, -self.shift_size),
                        slice(-self.shift_size, None))
            w_slices = (slice(0, -self.window_size),
                        slice(-self.window_size, -self.shift_size),
                        slice(-self.shift_size, None))
            cnt = 0
            for h in h_slices:
                for w in w_slices:
                    img_mask[:, h, w, :] = cnt
                    cnt += 1

            mask_windows = window_partition(img_mask, self.window_size)  # nW, window_size, window_size, 1
            mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
            attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
            attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
        else:
            attn_mask = None

        self.register_buffer("attn_mask", attn_mask)
        self.fused_window_process = fused_window_process

    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)

        # cyclic shift
        if self.shift_size > 0:
            if not self.fused_window_process:
                shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
                # partition windows
                x_windows = window_partition(shifted_x, self.window_size)  # nW*B, window_size, window_size, C
            else:
                x_windows = WindowProcess.apply(x, B, H, W, C, -self.shift_size, self.window_size)
        else:
            shifted_x = x
            # 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

        # W-MSA/SW-MSA
        attn_windows = self.attn(x_windows, mask=self.attn_mask)  # nW*B, window_size*window_size, C

        # merge windows
        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)

        # reverse cyclic shift
        if self.shift_size > 0:
            if not self.fused_window_process:
                shifted_x = window_reverse(attn_windows, self.window_size, H, W)  # B H' W' C
                x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
            else:
                x = WindowProcessReverse.apply(attn_windows, B, H, W, C, self.shift_size, self.window_size)
        else:
            shifted_x = window_reverse(attn_windows, self.window_size, H, W)  # B H' W' C
            x = shifted_x
        x = x.view(B, H * W, C)
        x = shortcut + self.drop_path(x)

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

        return x

    # ...

현재 SwinTransformerBlock 모듈의 input shape은 (B, 56x56, 96)이다. (stage 1) normLayer을 거친 뒤에 view 함수를 통해 (B, 56, 56, 96)으로 변한다. 이후 window_partition() 함수를 거치게 된다:

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

window_size로 패치들 window별로 나누어 준다. 단계별로 보면 다음과 같다:

• window_partition의 input shape은 (B, 56, 56, 96)이다.
• view() -> (B, 56/7, 7, 56/7, 7, 96) = (B, 8, 7, 8, 7, 96)
• permute() -> (B, 8, 8, 7, 7, 96)
• view() -> (Bx8x8, 7, 7, 96)

따라서, window_partition() 함수는 Bx8x8개의 window마다 7x7개의 patch가 있고, 각 patch당 길이 96의 embedding vector을 같은 window끼리 partition 해주는 역할을 한다.

다시 SwinTransformerBlock코드로 돌아가서, output은 view()함수를 통해 shape이 (Bx8x8, 7x7, 96)으로 변경되고, 이후에 각 window별로 self-attention 연산이 진행된다.

self-attention연산은 input과 output이 같은 shape을 갖는다. output의 shape은 window_reverse() 함수를 거쳐 다시 (B, 56x56, 96)이 된다.

window_partition() -> (Bx8x8, 7, 7, 96) -> (Bx8x8, 7x7, 96) -> self-attention -> (Bx8x8, 7x7, 96) -> (Bx8x8, 7, 7, 96) -> window_reverse() -> (B, 56x56, 96)

Shifted-window (SW-MSA)의 경우, torch.roll() 함수를 통해 window를 구성해주고, 다시 self-attention 이후에 반대방향으로 torch.roll()함수를 이용해 원상복귀 시킨다.

Patch Merging

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
    
    # ...

input의 shape은 (B, H, W, C) 이고, 이 input을 4개로 쪼개 주는데, 아래의 그림과 같이 H, W 부분을 한칸씩 띄어서 4개로 나누어 준다. 그 후, embedding vector 차원 기준으로 4개를 concat한다. (B, H/2, W/2, 4C). 마지막으로 x2 downsampling을 통해 block의 최종 output shape은 (B, H/2, W/2, 2C) 가 된다.

(B, 56x56, 96) -> (B, 56, 56, 96) -> (B, 56/2, 56/2, 96x4) -> (B, 56/2, 56/2, 96x2)

---

Reference

https://byeongjo-kim.tistory.com/36

---