--- # System prepended metadata title: Swin Transformer tags: [Transformers, Machine Learning] --- # Swin Transformer 源自: [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/pdf/2103.14030) ![image](https://hackmd.io/_uploads/rJlicK-ZWg.png) ## 原理 ### Hierarchical Architecture >Swin Transformer, which constructs hierarchical feature maps and has linear computational complexity to image size. Swin Transformer 則建構了分層特徵圖 (hierarchical feature maps)，使其能夠像傳統的卷積神經網絡（CNNs，如 VGG 和 ResNet）一樣，產生具有不同解析度的特徵表示 ![image](https://hackmd.io/_uploads/HyRNoKZ-Zl.png) ### Patch Partition and Embedding > It first splits an input RGB image into non-overlapping patches by a patch splitting module, like ViT. Each patch is treated as a “token” and its feature is set as a concatenation of the raw pixel RGB values. - 模型首先透過圖像塊分割模組 (patch splitting module) 將輸入的 RGB 圖像分割成不重疊的圖像塊 (non-overlapping patches)。 - 在實作中，每個 4×4 的圖像塊被視為一個「token」，其特徵是原始像素 RGB 值的串聯，維度為 4×4×3=48。 - 接著，應用線性嵌入層 (linear embedding layer) 將此原始 feature 投影到任意維度 C。 ![image](https://hackmd.io/_uploads/ryeghtbWbl.png) ### Patch Merging - 隨著網路深入，模型透過圖像塊合併層 (patch merging layers) 來減少 tokens 的數量，從而建立分層表示。 >To produce a hierarchical representation, the number of tokens is reduced by patch merging layers as the network gets deeper. - 第一個合併層會將相鄰的 2×2 個圖像塊的特徵進行串聯，形成 4C 維度的 feature。 >The first patch merging layer concatenates the features of each group of 2 × 2 neighboring patches, and applies a linear layer on the 4C-dimensional concatenated features. - 隨後，應用一個線性層，使 tokens 數量減少 2×2=4 倍（解析度降低一半），但輸出維度增加到 2C >This reduces the number of tokens by a multiple of 2×2 = 4 (2× downsampling of resolution), and the out-put dimension is set to 2C. reference: [Swin Transformer解读](https://datawhalechina.github.io/thorough-pytorch/%E7%AC%AC%E5%8D%81%E7%AB%A0/Swin-Transformer%E8%A7%A3%E8%AF%BB.html) ![image](https://hackmd.io/_uploads/BkM-85-bbg.png) ![image](https://hackmd.io/_uploads/ry3BI9-Z-l.png) ### Shifted Window based Self-Attention (SW-MSA) ![image](https://hackmd.io/_uploads/r1-SJsb-bg.png) >To intro-duce cross-window connections while maintaining the effi-cient computation of non-overlapping windows, we propose a shifted window partitioning approach which alternates be-tween two partitioning configurations in consecutive Swin Transformer blocks. 因為在僅在固定窗口內計算 attention 會導致窗口之間缺乏連接，限制了模型的建模能力 ![image](https://ask.qcloudimg.com/http-save/yehe-7220647/3e9687789154a6138a9fbe5b87910468.gif) 所以在各自 window 中算完 MSA(Multi-Head Self-Attention) 後，再用 Shifted Window MSA，增加模型 cross-window connection的能力 reference: [使用动图深入解释微软的Swin Transformer](https://cloud.tencent.com/developer/article/2015888) ![image](https://ask.qcloudimg.com/http-save/yehe-7220647/1f201eee7e0d61403ea7714c306bffe5.gif) ## Source Code reference: https://github.com/microsoft/Swin-Transformer/blob/main/models/swin_transformer.py ### `MLP` ```python class Mlp(nn.Module): def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=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): x = self.fc1(x) x = self.act(x) x = self.drop(x) x = self.fc2(x) x = self.drop(x) return x ``` 1. 為什麼 `fc` 不能用只用一個? **MLP 數學本質** $$y=W_2 \cdot \sigma(W_1 \cdot + b_1) + b_2$$ 基本上 $W_1$ 和 $W_2$ 是兩個不同矩陣，中間還有個 $\sigma$ 是激活函數 (activation function) 如果共用一個 `fc` $$y=W\cdot \sigma(W\cdot x)$$ 表達能力下降，可能造成 underfitting，且兩次線性轉換的目的不同 ``` x = fc1(x) # in_features → hidden_features x = GELU(x) x = fc2(x) # hidden_features → out_features ``` 第一次線性轉換: 展開特徵，改變座標系，feature projection 激活函數: 打破線性限制 ![image](https://hackmd.io/_uploads/BJv4DJ-mbx.png) 第二次線性轉換: 篩選出已經被非線性處理過的高維特徵，feature mixing，對齊模型需要的輸出格式 ### `window partition` ```python 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 ``` 比較不同維度操作的差別 | 操作 | 做什麼| 改變資料順序？| 需要 contiguous？ | 常見用途 | | -| -| -| - | - | | `view`| 改 shape| ❌ 否| ✅ 必須| 最快，純 reshape | | `reshape` | 改 shape| ❌ 否*| ❌（必要時會 copy）| 安全版 view | | `permute` | 換維度順序 | ❌（只是換 index） | ❌| NCHW ↔ NHWC | | `transpose` | 交換兩個維度 | ❌ | ❌| 矩陣轉置 | 所以，如果要避免出錯，可以先用 `reshape`，否則一定要改成 `contiguous()` ```python x = x.permute(0, 2, 1).contiguous().view(...) # 或 x = x.permute(0, 2, 1).reshape(...) ``` 所以 `window_partition` 不能直接 ```python x = x.view(B, H/window_size, W/window_size,window_size, window_size, C ) ``` 因為 `view` 不能不能改變資料在記憶體中的排列順序，只能重新解讀線性連續的一維記憶體 ### `window_reverse` ```python 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 ``` 從 partition 後的 window，轉回原本的維度 `(B,H,W,C)` ### `WindowAttention` ```python class WindowAttention(nn.Module): r""" Window based multi-head self attention (W-MSA) module with relative position bias. It supports both of shifted and non-shifted window. Args: dim (int): Number of input channels. window_size (tuple[int]): The height and width of the window. num_heads (int): Number of attention heads. qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0 proj_drop (float, optional): Dropout ratio of output. Default: 0.0 """ def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.): super().__init__() self.dim = dim self.window_size = window_size # Wh, Ww self.num_heads = num_heads head_dim = dim // num_heads self.scale = qk_scale or head_dim ** -0.5 # define a parameter table of relative position bias self.relative_position_bias_table = nn.Parameter( torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)) # 2*Wh-1 * 2*Ww-1, nH # get pair-wise relative position index for each token inside the window coords_h = torch.arange(self.window_size[0]) coords_w = torch.arange(self.window_size[1]) coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2 relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0 relative_coords[:, :, 1] += self.window_size[1] - 1 relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1 relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww self.register_buffer("relative_position_index", relative_position_index) self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias) self.attn_drop = nn.Dropout(attn_drop) self.proj = nn.Linear(dim, dim) self.proj_drop = nn.Dropout(proj_drop) trunc_normal_(self.relative_position_bias_table, std=.02) self.softmax = nn.Softmax(dim=-1) def forward(self, x, mask=None): """ Args: x: input features with shape of (num_windows*B, N, C) mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None """ B_, N, C = x.shape qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple) q = q * self.scale attn = (q @ k.transpose(-2, -1)) relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view( self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww attn = attn + relative_position_bias.unsqueeze(0) if mask is not None: nW = mask.shape[0] attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0) attn = attn.view(-1, self.num_heads, N, N) attn = self.softmax(attn) else: attn = self.softmax(attn) attn = self.attn_drop(attn) x = (attn @ v).transpose(1, 2).reshape(B_, N, C) x = self.proj(x) x = self.proj_drop(x) return x def extra_repr(self) -> str: return f'dim={self.dim}, window_size={self.window_size}, num_heads={self.num_heads}' def flops(self, N): # calculate flops for 1 window with token length of N flops = 0 # qkv = self.qkv(x) flops += N * self.dim * 3 * self.dim # attn = (q @ k.transpose(-2, -1)) flops += self.num_heads * N * (self.dim // self.num_heads) * N # x = (attn @ v) flops += self.num_heads * N * N * (self.dim // self.num_heads) # x = self.proj(x) flops += N * self.dim * self.dim return flops ``` 就是論文中的 $W-MSA$ ![image](https://ask.qcloudimg.com/http-save/yehe-7220647/3e9687789154a6138a9fbe5b87910468.gif) #### `relative_position` 可以參考: https://datawhalechina.github.io/thorough-pytorch/%E7%AC%AC%E5%8D%81%E7%AB%A0/Swin-Transformer%E8%A7%A3%E8%AF%BB.html#id3 透過 relative position bias 可以讓模型學習 **距離感知** 的權重 - 為什麼不用絕對位置? 會對每個 token 的固定位置產生 bias - 保留相對空間信息: 讓模型知道哪個 patch 在上下左右的相對距離有多遠 - 平移不變: 這樣對於之後要做 shifted windows(SW-MSA) 很重要 - 提升注意力經度: 模型能更合理分配注意力給空間上相近或相關的 patch #### `QKV` 1. `q`, `k`, `v` : `(B_, nH, N, d)` `B_`: $\cfrac{H \times W}{window_size}$ `nH`: num of head `N` : $H \times W$ `d` : $\cfrac{C}{n\_H}$ 2. `q = q * self.scale` `(B_, nH, N, d)` $\to$ `(B_, nH, N, d)` $$scale = \cfrac{1}{\sqrt{d}}$$ 3. `attn = (q @ k.transpose(-2, -1))` `k.transpose(-2, -1)` : `(B_, nH, N, d)` $\to$ `(B_, hH, d, N)` `attn` : `(B_, nH, N, d)` $\otimes$ `(B_, nH, d, N)` $=$ `(B_, nH, N, N)` 4. attn = attn + relative_position_bias.unsqueeze(0) `relative_position_bias` : $(N \times N, nH)$ reshape $\to$ `(N, N, nH)` permute $\to$ `(nH, N, N)` unsqueeze $\to$ `(1, nH, N, N)` 所以 attn : `(B_, nH, N, N)` 5. `x = (attn @ v).transpose(1, 2).reshape(B_, N, C)` `(attn @ v)`: `(B_, nH, N, N)` $\otimes$ `(B_, nH, N, d)` $\to$ `(B_, nH, N, d)` transpose $\to$ `(B_, N, nH, d)` reshape $\to$ `(B_, N, C)` ### `SwinTransformerBlock` ```python 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. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set. drop (float, optional): Dropout rate. Default: 0.0 attn_drop (float, optional): Attention 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 fused_window_process (bool, optional): If True, use one kernel to fused window shift & window partition for acceleration, similar for the reversed part. Default: False """ 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__() 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.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 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}" def flops(self): flops = 0 H, W = self.input_resolution # norm1 flops += self.dim * H * W # W-MSA/SW-MSA nW = H * W / self.window_size / self.window_size flops += nW * self.attn.flops(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 ``` ![image](https://hackmd.io/_uploads/HyK2Kp7QZx.png) 輸入: `x` $\to$ `(B, L, C)` \# $L = H \times W$ $\to$ `(B, H, W, C)` #### Cyclic Shift ```python shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)) ``` `torch.roll` : 對張量沿指定維度平移（循環移動）元素 1 維平移 ``` x = torch.tensor([1, 2, 3, 4, 5]) y = torch.roll(x, shifts=2) print(y) # tensor([4, 5, 1, 2, 3]) ``` 2 維平移 ``` x = torch.tensor([[1, 2, 3], [4, 5, 6]]) y = torch.roll(x, shifts=1, dims=0) print(y) # tensor([[4, 5, 6], # [1, 2, 3]]) ``` 在 SwinTransformer 是這樣運作的 ![image](https://hackmd.io/_uploads/r1-SJsb-bg.png) 在維度 1, 2: `H`(Height), `W`(Width) 進行平移 ```python dims=(1, 2) ``` 平移範圍 ```python shifts=(-self.shift_size, -self.shift_size) ``` 也就是，如果 `shift_size` 是 1 ```python 原始矩陣 x: [[1, 2, 3], [4, 5, 6], [7, 8, 9]] torch.roll(x, shifts=(-1, -1), dims=(0,1)) => [[5, 6, 4], [8, 9, 7], [2, 3, 1]] ``` - 高度方向平移 `-shift_size`，寬度方向也平移 `-shift_size` - 負號表示向上和向左平移 向上平移 ```python [[4, 5, 6], [7, 8, 9], [1, 2, 3]] ``` 向左平移 ```python [[5, 6, 4], [8, 9, 7], [2, 3, 1]] ``` - 循環移動：溢出的元素從另一端回來 #### Partition windows 把整個 image 切成多個 patch `(B, H, W, C)` $\to$ `(B * H/window_size * W/window_size, window_size, window_size, C )` ![image](https://hackmd.io/_uploads/B1domp7m-l.png) 這樣看來是 shift 完才 partition 的 #### Attention ```python attn_windows = self.attn(x_windows, mask=self.attn_mask) ``` 用 [WindowAttention](https://hackmd.io/@clh/BkQ0dg2Nkl/https%3A%2F%2Fhackmd.io%2F%40clh%2FBypDcKbZ-g#WindowAttention) 來計算每個 window 的 attention 現在維度: `(B * W/window_size * H/window_size, window_size * window_size, C)` #### Merge Windows $\to$ Reverse Cyclic Shift 進行 merge ```python attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) ``` 其實只是轉換維度 `(B * W/window_size * H/window_size, window_size * window_size, C)` $\to$ `(nW, window_size, window_size, C)` `nW` : `B * W/window_size * H/window_size` 最後 reverse 回去 ```pytyhon shifted_x = window_reverse(attn_windows, self.window_size, H, W) ``` $\to$ `(B, H, W, C)` cyclic shift 也要平移回去 ```python x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2)) ``` #### Residual Connection ![image](https://hackmd.io/_uploads/Hkcl9677be.png) ```python shortcut = x ... x = x.view(B, H * W, C) x = shortcut + self.drop_path(x) ``` 原本輸入的 `x` 維度: `(B, H*W, C)` 再加上 `MSA` 的 attention 加上 Residual Connection 的好處是 - 梯度流暢：殘差連接可以幫助梯度直接傳回，避免深層網路訓練困難。 - 保留原始訊息：原始輸入訊息不會完全被改變，注意力只負責補充特徵。 - 提升模型穩定性：避免注意力輸出過度影響網路表現。 #### Block ![image](https://hackmd.io/_uploads/SyM9z0QXbx.png) 最後經過 MLP ```python x = x + self.drop_path(self.mlp(self.norm2(x))) ``` 維度: `(B, H*W, C)` ### `PatchMerging` ![image](https://hackmd.io/_uploads/ryH6SpQ7-l.png) ```python 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 ``` 它的作用類似於卷積神經網絡（CNN）中的 Pooling（池化） 層，目的是降低解析度（下採樣）並增加通道數，從而讓模型能夠捕捉到更大範圍的特徵（增大感受野）。 x: `(B, H*W, C)` view $\to$ `(B, H, W, C)` 這樣做是為了在降低計算量（減少 Token 數量）的同時，極大化保留影像資訊並增加特徵維度。 $\to$ `(B, H/2, W/2, 4C)` view $\to$ `(B, H/2 * W/2, 4C)` ### `BasicLayer` ```python lass 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. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set. drop (float, optional): Dropout rate. Default: 0.0 attn_drop (float, optional): Attention 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. fused_window_process (bool, optional): If True, use one kernel to fused window shift & window partition for acceleration, similar for the reversed part. Default: False """ 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__() self.dim = dim self.input_resolution = input_resolution self.depth = depth self.use_checkpoint = use_checkpoint # 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 ``` 指這每一塊 ![image](https://hackmd.io/_uploads/BJNcBRmQWg.png) ### `PatchEmbed` ```python 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 ``` 把 image 進行 patch embedding: 將一張連續的影像（像素）轉換成一個個離散的序列（Tokens），這樣 Transformer 才能處理。 假設輸入影像是 $224 \times 224 \times 3$，patch_size=4，embed_dim=96： |步驟|程式碼操作|Shape|說明| |-|-|-|-| |輸入|`x`|$(B, 3, 224, 224)$|原始影像 $(C, H, W)$| |卷積投影|`self.proj(x)`|$(B, 96, 56, 56)$|$224/4 = 56$。影像縮小，深度變厚| |展平|`.flatten(2)`|$(B, 96, 3136)$|將 $56 \times 56$ 的空間維度拉直成 $3136$ 個點| |轉置|`.transpose(1, 2)`|$(B, 3136, 96)$|最終格式：$(B, L, C)$，符合 Transformer 要求| ### `SwinTransformer`