0%

ViT:视觉Transformer开创者(论文+代码详解)

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Alexey Dosovitskiy (Google, Brain Team)

论文地址: ViT, arxiv

代码地址: timm-vision_transformer.py

动机:

  • 是否CNN对于图像特征提取来说不是必须的?
  • NLP领域中,Transformer相关的Encoder和Decoder这一套框架是否可以代替CNN来完成图像领域的任务?

创新点:

  • 将NLP领域的Transformer架构成功迁移到了CV领域,并且取得了和CNN相当的结果
  • 不降低图像分辨率,将图像分成小的patch,加入位置编码后加入到网络中
  • 实验表明,在超大数据集上训练,在常规数据集上Finetune后可以和CNN的SOTA结果相比较,所用计算资源也大大减少

实验参数:

结果汇总:

网络架构:

预处理(Embedding):
  1. Patch Embedding:将3维图像分patch,并且压缩到2维

    • 操作分解:

      1. 图像原始尺寸:\(B \times C \times H \times W\)

      2. 按照patch块大小 \(P\) 切块,把patch内的像素放到channel维度 \((C*P^2) \times (H/P) \times (W/P)\)
      3. 将channel维度线性映射到embedding dimension \(D\),即变为 \(D \times (H/P) \times (W/P)\)
      4. 将后两维flatten成一个维度 \(N\),得到 \(D \times N\) 的张量
      5. \(D\)\(N\) 所在的维度transpose,得到 \(N \times D\) 的张量,因为此时 \(D\) 中存的是patch的信息,这里就变成了feature map,而 \(N\) 则是patch的数量,相当于原始的channel
      6. 对处理过后的张量做LN操作,即对每个feature map做归一化,LN共 \(2B \times N\) 个参数
      7. 处理后的张量大小:\(B \times N \times D\)

    • 代码解读:./timm/models/layers/patch_embed.py

      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
      class PatchEmbed(nn.Module):
      """ 2D Image to Patch Embedding
      """
      def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, norm_layer=None, flatten=True):
      super().__init__()
      img_size = to_2tuple(img_size)
      patch_size = to_2tuple(patch_size)
      self.img_size = img_size
      self.patch_size = patch_size
      self.grid_size = (img_size[0] // patch_size[0], img_size[1] // patch_size[1])
      self.num_patches = self.grid_size[0] * self.grid_size[1]
      self.flatten = flatten

      self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size) # 用卷积实现上述2、3步骤
      self.norm = norm_layer(embed_dim) if norm_layer else nn.Identity()

      def forward(self, x):
      B, C, H, W = x.shape
      _assert(H == self.img_size[0], f"Input image height ({H}) doesn't match model ({self.img_size[0]}).")
      _assert(W == self.img_size[1], f"Input image width ({W}) doesn't match model ({self.img_size[1]}).")
      x = self.proj(x)
      if self.flatten:
      x = x.flatten(2).transpose(1, 2) # 先把宽高维度flatten为N,然后将其与channel维度transpose # BCHW -> BNC 这里后者的C其实为embed_dim D
      x = self.norm(x) # 这里还有一个标准化层
      return x
  2. Learnable Class Embedding

    • 操作分解:(如果存在蒸馏,则同理加入蒸馏token)

      1. 新建大小为 \(1 \times 1 \times D\) 维度的可训练的变量
      2. .expand\(B \times 1 \times D\) 维度
      3. 把patch embedding的结果与class embedding的结果在第一维上concat torch.cat()
    • 代码解读:./timm/models/vision_transformer.py

      1
      2
      3
      4
      5
      6
      7
      8
      9
      10
      11
      12
      13
      14
      15
      16
      17
      18
      19
      20
      21
      22
      class VisionTransformer(nn.Module):
      def __init__(...):
      # 只截取了相关的部分
      self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim)) # 类别token 对应第1步
      self.dist_token = nn.Parameter(torch.zeros(1, 1, embed_dim)) if distilled else None # 蒸馏token
      self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + self.num_tokens, embed_dim)) # PE Position Embedding
      self.pos_drop = nn.Dropout(p=drop_rate)

      def forward_features(self, x):
      x = self.patch_embed(x) # patch embedding
      cls_token = self.cls_token.expand(x.shape[0], -1, -1) # 对应第2步
      if self.dist_token is None:
      x = torch.cat((cls_token, x), dim=1) # 对应第3步
      else:
      x = torch.cat((cls_token, self.dist_token.expand(x.shape[0], -1, -1), x), dim=1) # 如果蒸馏,则concat分类token、蒸馏token和patch token
      x = self.pos_drop(x + self.pos_embed) # PE
      x = self.blocks(x)
      x = self.norm(x)
      if self.dist_token is None:
      return self.pre_logits(x[:, 0])
      else:
      return x[:, 0], x[:, 1]
  3. Position Embedding

    • 操作分解:
      1. 新建大小为 \(1 \times (N+1) \times D\) 维度的可训练的变量,其中 \(x\) 为额外加的token数量,通常为1或者2
      2. 将新建的变量+到embedding后的tensor(常量)上,不同batch共享同样的PE变量
    • 代码解读:见上方Class Embedding
Transformer Encoder *L 重复L次
  1. 总览: \[ { z_0 = [x_{class};\ x_p^1E;\ x_p^2E;\ ···\ ;\ x_p^NE]+\ E_{pos},\ \ \ E \in R^{P^2C \times D},\ E_{pos} \in R^{(N+1) \times D} \ \ \ (1)\\\\\\ z'_l = MSA(LN(z_{l-1}))\ + \ z_{l-1}, \ \ \ l=1...L \ \ \ (2)\\\\ z_l = MLP(LN(z'_{l}))\ + \ z'_{l}, \ \ \ l=1...L \ \ \ (3)\\\\ y = LN(z^{0}_{L}) \ \ \ (4) } \]

    • 操作分解:
  2. \(z_0\) 为embedding后的结果,\(x_{class}\) 为分类token,\(x_p^i \in R^{1\times P^2C}\) 为图像分完patch后的向量,\(E\) 为线性变换矩阵,\(E_{pos}\) 为位置编码
    1. 共有\(L\)个transformer结构,每个结构包含两个部分
    2. 先做LN再做MSA,接一个shotcut
      1. 先做LN再做MLP,接一个shotcut
    3. 将最后一个transformer的输出中的第0个patch取出来做LN后得到 \(y\) 用于分类
  • 代码解读:

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
class Block(nn.Module):

def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, drop=0., attn_drop=0.,
drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm):
super().__init__()
self.norm1 = norm_layer(dim)
self.attn = Attention(dim, num_heads=num_heads, qkv_bias=qkv_bias, attn_drop=attn_drop, proj_drop=drop) # 这里对应Attention模块,其中Multi-head的数量就是 num_heads也就是注意力模块的数量
# NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
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) # 对应MLP模块,其中hidden_features就是指中间隐藏层降到的维数

def forward(self, x):
x = x + self.drop_path(self.attn(self.norm1(x))) # 对应上方的公式(2)
x = x + self.drop_path(self.mlp(self.norm2(x))) # 对应上方的公式(3)
return x

class VisionTransformer(nn.Module):
def __init__(...):
# 只截取了相关代码
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)] # stochastic depth decay rule
self.blocks = nn.Sequential(*[
Block(dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, drop=drop_rate,attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer, act_layer=act_layer)
for i in range(depth)]) # 用nn.Sequential(*[Block() for i in range(depth)])来完成transfermor结构重复depth次,depth=L
self.norm = norm_layer(embed_dim)

def forward_features(self, x):
# 只截取部分代码
x = self.blocks(x)
x = self.norm(x)
if self.dist_token is None:
return self.pre_logits(x[:, 0])
else:
return x[:, 0], x[:, 1]

  1. MSA(Multi-head Self Attention) 想了解一下head数量对于模型的影响!

    • Self Attention 机制\(Attention(x) \ = \ Softmax(Q \times K^T/\sqrt{d}) \times V\)
      • 为什么除以根号d?:防止 \(Q\times K^T\) 的数值过大,过 \(Softmax\) 时梯度消失
      • \(Softmax(Q \times K^T/\sqrt{d})\) 整个相当于一个注意力矩阵,乘以原始图像 \(V\)
    • 操作分解:

      1. 将输入的 \(B\times N\times D\) 做线性变换为原来的三倍 \(B\times N\times 3*D\)
      2. reshape\(B\times N\times 3 \times H \times (D/H)\),其中 \(H\) 为Multi-head的数量,相当于把每个patch在分成H分,每一份都做不同的attention
      3. permute重新排列为 \(3\times B\times H \times N \times (D/H)\),然后unbind(0)得到 \(q,k,v\) 三个张量,大小为\(B\times H \times N \times (D/H)\)
      4. 利用公式 \(Attention(x) \ = \ Softmax(q \times k^T/\sqrt{D/H}) \times v\) 计算注意力,中间会添加dropout防止过拟合
      5. transpose(1,2).reshape(B, N, C)后将维度变换回了 \(B\times N\times D\)
      6. 之后加一个全连接层,再接一个dropout层
    • 代码解读:

      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
      class Attention(nn.Module):
      def __init__(self, dim, num_heads=8, qkv_bias=False, attn_drop=0., proj_drop=0.):
      super().__init__()
      assert dim % num_heads == 0, 'dim should be divisible by num_heads'
      self.num_heads = num_heads
      head_dim = dim // num_heads
      self.scale = head_dim ** -0.5

      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)

      def forward(self, x):
      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.unbind(0) # make torchscript happy (cannot use tensor as tuple)

      attn = (q @ k.transpose(-2, -1)) * self.scale
      attn = attn.softmax(dim=-1)
      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
  2. MLP(Multi Layer Processor)

    • 操作分解:其实就是一个简单的降维再升维的操作,用全连接层实现

      1. nn.Linear(in_dim, out_dim)全连接层将 \(B\times N\times D\) 变成 \(B\times N\times H\),其中H为隐藏层维度
      2. nn.Gelu激活函数层,增加非线性
      3. nn.Dropout(p)随机丢弃一些点,不做正向传播和反向梯度更新
      4. nn.Linear(in_dim, out_dim)全连接层将 \(B\times N\times H\) 变回 \(B\times N\times D\),其中H为隐藏层维度
      5. nn.Gelu激活函数层,增加非线性
      6. nn.Dropout(p)随机丢弃一些点,不做正向传播和反向梯度更新
    • 代码解读:

      1
      2
      3
      4
      5
      6
      7
      8
      9
      10
      11
      12
      13
      14
      15
      16
      17
      18
      19
      20
      21
      22
      class Mlp(nn.Module):
      """ MLP as used in Vision Transformer, MLP-Mixer and related networks
      """
      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
      drop_probs = to_2tuple(drop)

      self.fc1 = nn.Linear(in_features, hidden_features)
      self.act = act_layer()
      self.drop1 = nn.Dropout(drop_probs[0])
      self.fc2 = nn.Linear(hidden_features, out_features)
      self.drop2 = nn.Dropout(drop_probs[1])

      def forward(self, x):
      x = self.fc1(x)
      x = self.act(x)
      x = self.drop1(x)
      x = self.fc2(x)
      x = self.drop2(x)
      return x
后处理(Classification):

  1. 分类:

    • 操作分解:

      1. 对于Transformer的输出中class token对应的位置进行操作,先过一个MLP层,然后再过一个分类头后计算cross entropy loss
      2. 这里预训练和微调时不太一样,微调时会把MLP层删掉,只留下分类头。这个操作应该是借鉴了自监督,因为自监督在预训练是没有类别信息,所以MLP后跟的是一个contrastive loss,用于计算输出与原始图像的loss
    • 代码解读:

      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
      class VisionTransformer(nn.Module):       
      def __init__(...):
      # 只截取了相关代码
      # Representation layer
      if representation_size and not distilled:
      self.num_features = representation_size
      self.pre_logits = nn.Sequential(OrderedDict([ # 这里是预训练时的MLP层
      ('fc', nn.Linear(embed_dim, representation_size)),
      ('act', nn.Tanh())
      ]))
      else:
      self.pre_logits = nn.Identity()

      def forward_features(self, x):
      # 只截取部分代码
      if self.dist_token is None:
      return self.pre_logits(x[:, 0]) # 选取class token对应的区域进行分类
      else:
      return x[:, 0], x[:, 1]

      def forward(self, x):
      x = self.forward_features(x)
      if self.head_dist is not None:
      x, x_dist = self.head(x[0]), self.head_dist(x[1]) # x must be a tuple
      if self.training and not torch.jit.is_scripting():
      # during inference, return the average of both classifier predictions
      return x, x_dist
      else:
      return (x + x_dist) / 2
      else:
      x = self.head(x) # ViT不蒸馏直接分类
      return x
欲戴皇冠,必承其重,加油!