深入浅出一文图解Vision in Transformer（ViT）

ViT 快速理解 Vision in Transformer

---

全文翻译：

Vision in Transformer

---

站在巨人的肩膀上看世界

参考文献

• 知乎-深入理解 ViT(2)：https://www.zhihu.com/zvideo/1455593125721272320
• ViT论文逐段精读【论文精读】：作者-跟李沐学AI，主讲bryanyzhu
• VIT (Vision Transformer) 模型论文+代码(源码)从零详细解读，看不懂来打我：作者-NLP从入门到放弃
• Transformer中Self-Attention以及Multi-Head Attention详解：作者-霹雳吧啦Wz
• 11.1 Vision Transformer(vit)网络详解：作者-霹雳吧啦Wz
• 代码11.2 使用pytorch搭建Vision Transformer(vit)模型：作者-霹雳吧啦Wz
• 博客-详解Transformer中Self-Attention以及Multi-Head Attention：太阳花的小绿豆
• Transformer 论文详细解读（文中插图有助于直观了解注意力机制-让我芜湖灌顶！）：北方一瓶川
• 原文：https://arxiv.org/abs/2010.11929

2022年02月14日14:31:34学习完上述Vision Transformer教程，站在巨人的肩膀上可以让我们更快的前进。完成学习后对内容进行一个总结，以及对ViT模型进行一个简单介绍，让才接触的同学快速了解ViT模型。

---

Vision Transformer

我们以原文中的一张图，来快速了解ViT这个模型，我们把整体结构分成五个部分：

1. 将输入的图像进行patch的划分
2. Linear Projection of Flatted Patches，将patch拉平并进行线性映射
3. 生成CLS token特殊字符*，生成Position Embedding，Patch+Position Embedding相加作为inputs token
4. Transformer Encoder编码，特征提取
5. MLP Head进行分类输出结果

---

---

第一部分：图像划分Patch

名称	说明
输入图像维度	224×224×3
Patch块大小	16×16
token个数	196
token维度	768

将图像分成16×16的patch(小方块)，每个patch块可以看做是一个token(词向量)，共有（224/16=14）14×14=196个token，每个token的维度为16×16×3=768。

patch块大小是16×16，每个patch块是RGB三通道的彩色图像）

---

第二部分：Linear Projection of Flatted Patches

一个patch块它的维度是16×16×3=768，我们把它flatten拉平成行向量它的长度就为768，一共有14×14=196个patch，所以输入的维度是[196, 768]，我们经过一个Linear Projection（映射）到指定的维度，比如1024或2048，我们用全连接层来实现，但映射的维度我们任然选择为768，那么此时的输出是[196, 768]。
整个过程我们把它称作patch embedding；输出的结果维度是[196, 768]

但是有大佬就不服啦，反正都是线性映射，为啥我要这么复杂，把它flatten后在用全连接层呢？聪明的大佬就想到，我直接使用卷积来实现，用16×16大小的卷积核，步长stride=16，维度设为768，输入[3, 224, 224]->[768, 14, 14]，然后交换并合并一下维度不就得到结果了吗？[196, 768]

---

第三部分：Patch+Position Embedding

patch embedding的维度为[196, 768]

1. 首先生成一个cls token它的维度为[1, 768]，然后拼接到输入的path embedding，得到的维度为[197, 768]

2. 对197个patch都生成一个位置信息，它的维度同patch维度为[197, 768]

3. Patch+Position Embedding，直接相加为新的的token作为encoder的输入

---

疑惑解答

• cls token和位置信息编码是如何来的呢？随机生成的可学习参数，可以全零初始化，也可以0-1随机初始化

        cls_token = nn.Parameter(torch.zeros(1, 768))
        pos_embedding = nn.Parameter(torch.zeros(197, 768))

• cls token的作用是为了同NLP领域的Transformer保持一致，最后直接提取cls token作为网络提取得到的特征，作为类别预测的输出，放入MLP进行分类

• 不使用cls token也是可以的，对196个维度为768的patch进行全局均值池化，得到结果维度为[196，1]后作为类别预测的输出，放入MPL进行分类

• 通过学习率的调整可以让cls token和对patch全局均值池化两种方法获得相近似的结果，但为尽可能接近原始Transformer模型，选择使用额外的cls标记

---

• 位置信息编码是为了给patch加上相对位置，不然在后面的特征提取中丢掉位置信息可就不好了

• 位置信息编码被用于增加patch embedding的原始位置信息

• 位置信息编码有可学习参数类型，有通过公式计算的方法，可以是一维、二维；但不使用性能会差点

• 没有位置编码性能差距会很大，但选择不同位置编码几乎没有差异，原因是Transformer是直接在patch上操作而不是基于像素级，因此，空间信息编码方式差异没那么重要

---

---

第四部分：Transformer Encoder

整个Transformer Encoder结构如下，详细代码实现和Multi-Head Attention的理解请看开头的视频讲解。

名称	说明
输入token（Embedded Patches）	维度为[197, 768]
编码后的输出	[197, 768]
L×	重复EncoderL次
Norm	Layer Norm归一化
Multi-Head Attention	多头注意力机制
+	跳跃连接
MLP	Linear全连接层

Transformer Encoder网络结构如下：

Encoder输入的维度为[197, 768]，输出的维度为[197, 768]，可以把中间过程简单的理解成为特征提取的过程
其中的Multi-Head Attention多头注意力机制，看完开头 太阳花的小绿豆的博客后，做一个简单的认识：

---

Self-Attention:

$$\text{ Attention }(Q,K,V)=\mathrm{softmax}\left(\frac{Q{K}^T}{\sqrt{d_k}}\right)V$$

---

Multi-Head Attention:

$$\begin{array}{rl}\mathrm{MultiHead}(Q,K,V)& =\text{ Concat }\left({\text{ head }}_1,\dots ,{\text{ head }}_{\mathrm{h}}\right)W^O\\ {\text{ where head }}_{\mathrm{i}}& =\text{ Attention }\left(Q{W}_i^Q,K{W}_i^K,V{W}_i^V\right)\end{array}$$

---

注意力机制细节

---

第五步：MLP Head

分类头的意思就是：特征提取工作已经全部完成，现在你要做分类就加上对应的操作。（同CNN特征提取层后的Linear层+Softmax进行分类预测）。

整个Encoder的输出为[197, 768]我们仅仅保留最前面的CLS token作为全连接的输入[1, 768]，然后接上全连接层及分类数n_class，使用交叉熵损失函数计算损失，反向传播更新网络的权重和参数。

原文分类头的细节

• 分类头在预训练时由含有一个隐含层的MLP实现，微调时由简单的线性层来实现

• 在将ViT模型迁移到另一个数据集时，我们移除整个head（两个全连接层-隐含层和输出层），并将其替换为一个简单的、全零初始化的线性层，输出的类别数为目标数据集的类别数

• 我们发现这样初始化比简单地重新初始化最后一层要更加具有鲁棒性

---

2022年02月15日15:26:37，over。

---

新增相关原文插图说明：

• 模型参数量和层结构：
  

• 计算资源与结果
  

• CLASS TOKEN和全局均值池化：
  

• 不同位置和编码方式：
  

• 位置信息编码的相似性图，表示的是每一个图像块本身，与其它所有块的相似度（距离近相似度高，与自己的相似度为1，颜色为黄色）
  

• 不同超参数模型训练的位置编码
  

• 不同数据集进行迁移学习后微调的实验结果：
  

• 自注意力机制示意图
  

• 原文细节请看全文翻译：点击跳转到原文翻译

---

源码可直接运行

• 代码来自于哔哩哔哩博主：霹雳吧啦Wz
• 提供源码的视频链接（见置顶评论）：11.2 使用pytorch搭建Vision Transformer(vit)模型

ViT Model

"""
original code from rwightman:
https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
"""
import os
from collections import OrderedDict
from functools import partial

import torch
import torch.nn as nn

os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"


def drop_path(x, drop_prob: float = 0., training: bool = False):
    """
    Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
    This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
    changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
    'survival rate' as the argument.
    """
    if drop_prob == 0. or not training:
        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
    random_tensor.floor_()  # binarize
    output = x.div(keep_prob) * random_tensor
    return output


class DropPath(nn.Module):
    """
    Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks).
    """

    def __init__(self, drop_prob=None):
        super(DropPath, self).__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        return drop_path(x, self.drop_prob, self.training)


class PatchEmbed(nn.Module):
    """
    2D Image to Patch Embedding
    """

    def __init__(self, img_size=224, patch_size=16, in_c=3, embed_dim=768, norm_layer=None):
        super().__init__()
        img_size = (img_size, img_size)
        patch_size = (patch_size, 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.proj = nn.Conv2d(in_c, embed_dim, kernel_size=patch_size, stride=patch_size)
        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] and W == self.img_size[1], \
            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."

        # flatten: [B, C, H, W] -> [B, C, HW]
        # transpose: [B, C, HW] -> [B, HW, C]
        x = self.proj(x).flatten(2).transpose(1, 2)
        x = self.norm(x)
        return x


class Attention(nn.Module):
    def __init__(self,
                 dim,  # 输入token的dim
                 num_heads=8,
                 qkv_bias=False,
                 qk_scale=None,
                 attn_drop_ratio=0.,
                 proj_drop_ratio=0.):
        super(Attention, self).__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5
        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop_ratio)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop_ratio)

    def forward(self, x):
        # [batch_size, num_patches + 1, total_embed_dim]
        B, N, C = x.shape

        # qkv(): -> [batch_size, num_patches + 1, 3 * total_embed_dim]
        # reshape: -> [batch_size, num_patches + 1, 3, num_heads, embed_dim_per_head]
        # permute: -> [3, batch_size, num_heads, num_patches + 1, embed_dim_per_head]
        print("输入维度:", self.qkv(x).shape)
        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        # [batch_size, num_heads, num_patches + 1, embed_dim_per_head]
        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)
        print("q的维度:", q.shape)

        # transpose: -> [batch_size, num_heads, embed_dim_per_head, num_patches + 1]
        # @: multiply -> [batch_size, num_heads, num_patches + 1, num_patches + 1]
        attn = (q @ k.transpose(-2, -1)) * self.scale
        print("缩放尺度:", self.scale)
        print("k的维度:", k.shape)
        print("k转置的维度:", k.transpose(-2, -1).shape)
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)
        print("注意力权重的维度: ", attn.shape)

        # @: multiply -> [batch_size, num_heads, num_patches + 1, embed_dim_per_head]
        # transpose: -> [batch_size, num_patches + 1, num_heads, embed_dim_per_head]
        # reshape: -> [batch_size, num_patches + 1, total_embed_dim]
        print("v的权重: ", v.shape)
        print("注意力权重*v的维度: ", (attn @ v).shape)
        print("注意力结果的维度: ", (attn @ v).transpose(1, 2).shape)
        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        print("映射后encode输出维度: ", x.shape)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


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
        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


class Block(nn.Module):
    def __init__(self,
                 dim,
                 num_heads,
                 mlp_ratio=4.,
                 qkv_bias=False,
                 qk_scale=None,
                 drop_ratio=0.,
                 attn_drop_ratio=0.,
                 drop_path_ratio=0.,
                 act_layer=nn.GELU,
                 norm_layer=nn.LayerNorm):
        super(Block, self).__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,
                              attn_drop_ratio=attn_drop_ratio, proj_drop_ratio=drop_ratio)
        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
        self.drop_path = DropPath(drop_path_ratio) if drop_path_ratio > 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_ratio)

    def forward(self, x):
        x = x + self.drop_path(self.attn(self.norm1(x)))
        x = x + self.drop_path(self.mlp(self.norm2(x)))
        return x


class VisionTransformer(nn.Module):
    def __init__(self, img_size=224, patch_size=16, in_c=3, num_classes=1000,
                 embed_dim=768, depth=12, num_heads=12, mlp_ratio=4.0, qkv_bias=True,
                 qk_scale=None, representation_size=None, distilled=False, drop_ratio=0.,
                 attn_drop_ratio=0., drop_path_ratio=0., embed_layer=PatchEmbed, norm_layer=None,
                 act_layer=None):
        """
        Args:
            img_size (int, tuple): input image size
            patch_size (int, tuple): patch size
            in_c (int): number of input channels
            num_classes (int): number of classes for classification head
            embed_dim (int): embedding dimension
            depth (int): depth of transformer
            num_heads (int): number of attention heads
            mlp_ratio (int): ratio of mlp hidden dim to embedding dim
            qkv_bias (bool): enable bias for qkv if True
            qk_scale (float): override default qk scale of head_dim ** -0.5 if set
            representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
            distilled (bool): model includes a distillation token and head as in DeiT models
            drop_ratio (float): dropout rate
            attn_drop_ratio (float): attention dropout rate
            drop_path_ratio (float): stochastic depth rate
            embed_layer (nn.Module): patch embedding layer
            norm_layer: (nn.Module): normalization layer
        """
        super(VisionTransformer, self).__init__()
        self.num_classes = num_classes
        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
        self.num_tokens = 2 if distilled else 1
        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
        act_layer = act_layer or nn.GELU

        self.patch_embed = embed_layer(img_size=img_size, patch_size=patch_size, in_c=in_c, embed_dim=embed_dim)
        num_patches = self.patch_embed.num_patches

        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
        self.dist_token = nn.Parameter(torch.zeros(1, 1, embed_dim)) if distilled else None
        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + self.num_tokens, embed_dim))
        self.pos_drop = nn.Dropout(p=drop_ratio)

        dpr = [x.item() for x in torch.linspace(0, drop_path_ratio, 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, qk_scale=qk_scale,
                  drop_ratio=drop_ratio, attn_drop_ratio=attn_drop_ratio, drop_path_ratio=dpr[i],
                  norm_layer=norm_layer, act_layer=act_layer)
            for i in range(depth)
        ])
        self.norm = norm_layer(embed_dim)

        # Representation layer
        if representation_size and not distilled:
            self.has_logits = True
            self.num_features = representation_size
            self.pre_logits = nn.Sequential(OrderedDict([
                ("fc", nn.Linear(embed_dim, representation_size)),
                ("act", nn.Tanh())
            ]))
        else:
            self.has_logits = False
            self.pre_logits = nn.Identity()

        # Classifier head(s)
        self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()
        self.head_dist = None
        if distilled:
            self.head_dist = nn.Linear(self.embed_dim, self.num_classes) if num_classes > 0 else nn.Identity()

        # Weight init
        nn.init.trunc_normal_(self.pos_embed, std=0.02)
        if self.dist_token is not None:
            nn.init.trunc_normal_(self.dist_token, std=0.02)

        nn.init.trunc_normal_(self.cls_token, std=0.02)
        self.apply(_init_vit_weights)

    def forward_features(self, x):
        # [B, C, H, W] -> [B, num_patches, embed_dim]
        x = self.patch_embed(x)  # [B, 196, 768]
        # [1, 1, 768] -> [B, 1, 768]
        cls_token = self.cls_token.expand(x.shape[0], -1, -1)
        if self.dist_token is None:
            x = torch.cat((cls_token, x), dim=1)  # [B, 197, 768]
        else:
            x = torch.cat((cls_token, self.dist_token.expand(x.shape[0], -1, -1), x), dim=1)

        x = self.pos_drop(x + self.pos_embed)
        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]

    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])
            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)
        return x


def _init_vit_weights(m):
    """
    ViT weight initialization
    :param m: module
    """
    if isinstance(m, nn.Linear):
        nn.init.trunc_normal_(m.weight, std=.01)
        if m.bias is not None:
            nn.init.zeros_(m.bias)
    elif isinstance(m, nn.Conv2d):
        nn.init.kaiming_normal_(m.weight, mode="fan_out")
        if m.bias is not None:
            nn.init.zeros_(m.bias)
    elif isinstance(m, nn.LayerNorm):
        nn.init.zeros_(m.bias)
        nn.init.ones_(m.weight)


def vit_base_patch16_224(num_classes: int = 1000):
    """
    ViT-Base model (ViT-B/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    weights ported from official Google JAX impl:
    链接: https://pan.baidu.com/s/1zqb08naP0RPqqfSXfkB2EA  密码: eu9f
    """
    model = VisionTransformer(img_size=224,
                              patch_size=16,
                              embed_dim=768,
                              depth=12,
                              num_heads=3,
                              representation_size=None,
                              num_classes=num_classes)
    return model


def vit_base_patch16_224_in21k(num_classes: int = 21843, has_logits: bool = True):
    """
    ViT-Base model (ViT-B/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    weights ported from official Google JAX impl:
    https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_base_patch16_224_in21k-e5005f0a.pth
    """
    model = VisionTransformer(img_size=224,
                              patch_size=16,
                              embed_dim=768,
                              depth=12,
                              num_heads=12,
                              representation_size=768 if has_logits else None,
                              num_classes=num_classes)
    return model


def vit_base_patch32_224(num_classes: int = 1000):
    """
    ViT-Base model (ViT-B/32) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    weights ported from official Google JAX impl:
    链接: https://pan.baidu.com/s/1hCv0U8pQomwAtHBYc4hmZg  密码: s5hl
    """
    model = VisionTransformer(img_size=224,
                              patch_size=32,
                              embed_dim=768,
                              depth=12,
                              num_heads=12,
                              representation_size=None,
                              num_classes=num_classes)
    return model


def vit_base_patch32_224_in21k(num_classes: int = 21843, has_logits: bool = True):
    """
    ViT-Base model (ViT-B/32) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    weights ported from official Google JAX impl:
    https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_base_patch32_224_in21k-8db57226.pth
    """
    model = VisionTransformer(img_size=224,
                              patch_size=32,
                              embed_dim=768,
                              depth=12,
                              num_heads=12,
                              representation_size=768 if has_logits else None,
                              num_classes=num_classes)
    return model


def vit_large_patch16_224(num_classes: int = 1000):
    """
    ViT-Large model (ViT-L/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    weights ported from official Google JAX impl:
    链接: https://pan.baidu.com/s/1cxBgZJJ6qUWPSBNcE4TdRQ  密码: qqt8
    """
    model = VisionTransformer(img_size=224,
                              patch_size=16,
                              embed_dim=1024,
                              depth=24,
                              num_heads=16,
                              representation_size=None,
                              num_classes=num_classes)
    return model


def vit_large_patch16_224_in21k(num_classes: int = 21843, has_logits: bool = True):
    """
    ViT-Large model (ViT-L/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    weights ported from official Google JAX impl:
    https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_large_patch16_224_in21k-606da67d.pth
    """
    model = VisionTransformer(img_size=224,
                              patch_size=16,
                              embed_dim=1024,
                              depth=24,
                              num_heads=16,
                              representation_size=1024 if has_logits else None,
                              num_classes=num_classes)
    return model


def vit_large_patch32_224_in21k(num_classes: int = 21843, has_logits: bool = True):
    """
    ViT-Large model (ViT-L/32) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    weights ported from official Google JAX impl:
    https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_large_patch32_224_in21k-9046d2e7.pth
    """
    model = VisionTransformer(img_size=224,
                              patch_size=32,
                              embed_dim=1024,
                              depth=24,
                              num_heads=16,
                              representation_size=1024 if has_logits else None,
                              num_classes=num_classes)
    return model


def vit_huge_patch14_224_in21k(num_classes: int = 21843, has_logits: bool = True):
    """
    ViT-Huge model (ViT-H/14) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    NOTE: converted weights not currently available, too large for github release hosting.
    """
    model = VisionTransformer(img_size=224,
                              patch_size=14,
                              embed_dim=1280,
                              depth=32,
                              num_heads=16,
                              representation_size=1280 if has_logits else None,
                              num_classes=num_classes)
    return model


if __name__ == '__main__':
    model = vit_base_patch16_224(num_classes=9)
    inputs = torch.rand(1, 3, 224, 224)
    outputs = model(inputs)
    print(model)
    print(outputs.shape)

---

多头自注意力计算

import torch
import torch.nn as nn


class Attention(nn.Module):
    def __init__(self,
                 dim,  # 输入token的dim
                 num_heads=8,
                 qkv_bias=False,
                 qk_scale=None,
                 attn_drop_ratio=0.,
                 proj_drop_ratio=0.):
        super(Attention, self).__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5
        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop_ratio)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop_ratio)

    def forward(self, x):
        # [batch_size, num_patches + 1, total_embed_dim]
        B, N, C = x.shape

        # qkv(): -> [batch_size, num_patches + 1, 3 * total_embed_dim]
        # reshape: -> [batch_size, num_patches + 1, 3, num_heads, embed_dim_per_head]
        # permute: -> [3, batch_size, num_heads, num_patches + 1, embed_dim_per_head]
        print("输入维度:", self.qkv(x).shape)
        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        # [batch_size, num_heads, num_patches + 1, embed_dim_per_head]
        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)
        print("q的维度:", q.shape)

        # transpose: -> [batch_size, num_heads, embed_dim_per_head, num_patches + 1]
        # @: multiply -> [batch_size, num_heads, num_patches + 1, num_patches + 1]
        attn = (q @ k.transpose(-2, -1)) * self.scale
        print("缩放尺度:", self.scale)
        print("k的维度:", k.shape)
        print("k转置的维度:", k.transpose(-2, -1).shape)
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)
        print("注意力权重的维度: ", attn.shape)

        # @: multiply -> [batch_size, num_heads, num_patches + 1, embed_dim_per_head]
        # transpose: -> [batch_size, num_patches + 1, num_heads, embed_dim_per_head]
        # reshape: -> [batch_size, num_patches + 1, total_embed_dim]
        print("v的权重: ", v.shape)
        print("注意力权重*v的维度: ", (attn @ v).shape)
        print("注意力结果的维度: ", (attn @ v).transpose(1, 2).shape)
        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        print("映射后encode输出维度: ", x.shape)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


if __name__ == '__main__':
    x = torch.rand(10, 16, 32, 46)
    B, C, H, W = x.shape

    x = x.view(B, C, H * W).permute(0, 2, 1)
    print(x.shape)
    model = Attention(dim=C, num_heads=4)
    x = model(x)

    x = x.permute(0, 2, 1).view(B, C, H, W)
    print(x.shape)

---

更新记录

1. 2022年04月14日16:08:09 -> 新增原文插图、全文翻译

2. 在原博客的基础上补充原文相关的细节描述

3. 准备新增注意力机制的矩阵的维度变换，及从理论和代码层次上理解Attention机制

4. 准备把大神噼里啪啦的代码贴在文中，以及自己组会的PPT上传

5. 新增ViT源码可直接运行

6. 2022年04月23日17:56:39，组会讲完了，尽力了；不过，没把人家讲明白是自己实力不足的表现，以后加油吧。