【图像分类】【深度学习】ViT算法Pytorch代码讲解-Toy模板网

这篇具有很好参考价值的文章主要介绍了【图像分类】【深度学习】ViT算法Pytorch代码讲解。希望对大家有所帮助。如果存在错误或未考虑完全的地方，请大家不吝赐教，您也可以点击"举报违法"按钮提交疑问。

【图像分类】【深度学习】【Pytorch版本】VisionTransformer模型算法详解

前言

ViT是由谷歌公司的Dosovitskiy, Alexey等人在《 An Image Is Worth 16x16 Words: Transformers For Image Recognition At Scale【ICLR2021】》【论文地址】一文中提出的模型，提出了一种基于transformer结构的模型，摒弃传统的CNN结构，直接将Transformer应用到图像块序列上一样可以达到非常好的性能。

ViT(Vision Transformer)讲解

论文在模型设计方面尽可能和原始Transformer保持一致。这种有意设计简单化的优点在于对NLP的Transformer有良好的扩展性，几乎可以实现“开箱即用”。
【图像分类】【深度学习】ViT算法Pytorch代码讲解
如上面的动态图所示，ViT在分类任务中的分为以下几个基本步骤：

图片切分为若干图片序列块
所有序列块经过线性映射后展平为一维向量

以上俩步是 patch embedding过程，成功将一个视觉问题转化为了一个seq2seq问题。

positional embedding：嵌入类别编码和位置编码(用于保留位置信息)
Transformer Encoder：主体编码器部分
MLP

patch embedding

输入图片大小为224x224，将图片切分为固定大小16x16的patch，则每张图像会生成224x224/16x16=196个patch，输出序列为196；196个patch(196x3)通过线性投射层(3x768)之后的维度为196x768，即一共有196个token，每个token的维度是768。

源码其实就是卷积核大小16x16(巧妙切分成固定大小16x16的patch)，卷积核通道3×768的卷积操作

通过patch embedding将一个视觉问题转化为了一个seq2seq问题，patch embedding示意图如下图所示：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

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__()
        # 图片分辨率 h w
        img_size = (img_size, img_size)
        # 卷积核大小
        patch_size = (patch_size, patch_size)
        self.img_size = img_size
        self.patch_size = patch_size
        # 分别计算w、h方向上的patch的个数
        self.grid_size = (img_size[0] // patch_size[0], img_size[1] // patch_size[1])
        # 一张图片的pacth个数
        self.num_patches = self.grid_size[0] * self.grid_size[1]
        # 卷积的步长巧妙地实现图片切分操作,而后与patch大小一致的卷积核完成线性映射
        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()      # nn.Identity()恒等函数 f(x)=x

    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]
        # 一维展平
        x = self.proj(x).flatten(2).transpose(1, 2)
        x = self.norm(x)
        return x

demo在动态流程图中的对应部分：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

positional embedding

经过patch embedding后输出维度是196x768，concat一个特殊字符cls(1x768)，维度变为197x768。ViT同样需要加入位置编码，位置编码可以理解为一张map，map的行数与输入序列个数相同(197)，每一行代表一个向量，向量的维度和输入序列embedding的维度相同（768）。位置编码的操作是sum，所以维度依旧保持197x768。

个人理解cls和pos的作用就是ax+b中的b，是可训练的归纳偏置，因为Vision Transformer相比CNN具有更少的针对图像的归纳偏置。position embeddings在初始化时不携带任何2维位置信息，图像块之间的所有空间关系都是从头学习得到的。

positional embedding结构如下图所示：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

class PositionEmbs(nn.Module):
    def __init__(self, num_patches, emb_dim, dropout_rate=0.1):
        super(PositionEmbs, self).__init__()
        self.pos_embedding = nn.Parameter(torch.zeros(1, num_patches + 1, emb_dim))
        if dropout_rate > 0:
            self.dropout = nn.Dropout(dropout_rate)
        else:
            self.dropout = None
    def forward(self, x):
        out = x + self.pos_embedding

        if self.dropout:
            out = self.dropout(out)
        return out

在class VisionTransformer的__init__函数中

# class token
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
# positional embedding
self.pos_embedding = PositionEmbs(num_patches, embed_dim, drop_ratio)

在class VisionTransformer的forward_features函数中

cls_token = self.cls_token.expand(x.shape[0], -1, -1)
# self.dist_token暂时可以忽略
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_embedding(x)

demo在动态流程图中的对应部分：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

Transformer Encoder

Transformer EncoderL个重复堆叠 Encoder Block组成。

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__()
        # 类别个数
        self.num_classes = num_classes
        # embed_dim默认tansformer的base 768
        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
        # 源码distilled是为了其他任务,分类暂时不考虑
        self.num_tokens = 2 if distilled else 1
        # LayerNorm:对每单个batch进行的归一化
        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
        # act_layer默认tansformer的GELU
        act_layer = act_layer or nn.GELU
        # embed_layer默认是patch embedding,在其他应用中应该会有其他选择
        # patch embedding过程
        self.patch_embed = embed_layer(img_size=img_size, patch_size=patch_size, in_c=in_c, embed_dim=embed_dim)
        # patche个数
        num_patches = self.patch_embed.num_patches
        # class token
        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
        # positional embedding过程
        self.pos_embedding = PositionEmbs(num_patches, embed_dim, drop_ratio)

        # self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + self.num_tokens, embed_dim))
        # self.pos_drop = nn.Dropout(p=drop_ratio)

        # depth是Block的个数
        # 不同block层数 drop_ratio的概率不同,越深度越高
        dpr = [x.item() for x in torch.linspace(0, drop_path_ratio, depth)]  # stochastic depth decay rule
        # blocks搭建
        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
        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)
        # self.dist_token暂时可以忽略
        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_embedding(x)
        # 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

demo在动态流程图中的对应部分：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

Encoder Block

组成 Transformer Encoder的单位，从低到高由LayerNorm (针对NLP领域提出的对每个token进行Norm处理)、Multi-Head Attention、DropPath(可直接换成Dropout)和 MLP等构成。
Block结构如下图所示：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

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)
        # Multi-head attention模块
        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)
        #  MLP模块
        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

Multi-head attention

读者先阅读大佬的博文【Multi-head attention流程与原理参考】。这里博主结合代码也做了自己的总结供大家参考。

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]
        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)
        # 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
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)
        # @: 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]
        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x

对于输入到Multi-head attention模块的特征 F(N×E) ，通过attention模块的nn.Linear进一步提取特征获得输出特征 v(value) 。为了考虑 N 个特征之间存在的亲疏和位置关系对于 v 的影响，所以需要一个额外 attn(attention) 或者理解为权重 w(weight) 对 v 进行加权操作，这引出了计算 w 所需的 q(query) 与 k(key) ， q 与 k 同样是特征 F 通过nn.Linear获取。
博主将 attn习惯了称作权重 weight了，有点误导人！！！，Attention结构如下图所示：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

下图简要展示了q、k和v的作用，便于理解原始特征v和原始特征v经过w加权计算后的特征V。

( ${{\rm{q}}_{\rm{n}}}$ , ${{\rm{k}}_{\rm{n}}}$ , ${{\rm{v}}_{\rm{n}}}$ )是特征n的经过attention模块后输出的三个特征。 ${{\rm{q}}_{\rm{1}}}$ 与 ${{\rm{k}}_{\rm{3}}}$ 矩阵计算得到权重 ${{\rm{w}}_{\rm{13}}}$ ，以此类推。以原始特征 ${{\rm{v}}_{\rm{1}}}$ 计算得到加权特征 ${{\rm{V}}_{\rm{1}}}$ 的过程为例， ${{\rm{v}}_{\rm{1}}}$ 分别于 ${{\rm{w}}_{\rm{1x}}}$ 、 ${{\rm{w}}_{\rm{2x}}}$ 和 ${{\rm{w}}_{\rm{3x}}}$ 矩阵运算得到 ${{\rm{V}}_{\rm{11}}}$ 、 ${{\rm{V}}_{\rm{12}}}$ 和 ${{\rm{V}}_{\rm{13}}}$ ，即最终的 ${{\rm{V}}_{\rm{1}}}$ 。
因此可以看到任何 ${{\rm{V}}_{\rm{i}}}$ 都考虑了 F 中 N 个特征之间相互的影响。

Multi-head attention的流程如下图所示(不考虑batchsize)：

首先将每个token的qkv特征维度embed_dim均拆分到每个head的上：
每个head分别通过q和k计算得到权重w，权重w和v得到输出output，合并所有head的output得到最终的output：

为了便于理解Multi-head self-attention的作用，可以类比卷积网络，单一的attention就类似于普通卷积，Multi-head self-attention则类似深度可分卷积。

demo在动态流程图中的对应部分：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

不考虑全连接的偏置，下图简单举例了3个Linear(3,2)和1个nn.Linear(3,6)再切片没有本质区别。

因此，无论q、k、v是特征F分别与3个Linear(dim,dim)操作获得，还是和一个nn.Linear(dim,3×dim)操作再切片获得，没有任何区别。

MLP

MLP结构如下图所示(不考虑batchsize)：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

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

demo在动态流程图中的对应部分：
【图像分类】【深度学习】ViT算法Pytorch代码讲解

完整代码

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

import torch
import torch.nn as nn

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)  # (shape[0],1,1,1)
    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)   # rand范围在[0~1]之间, +keep_prob在[keep_prob~keep_prob+1]之间
    random_tensor.floor_()  # 只保留0或者1
    output = x.div(keep_prob) * random_tensor   # x.div(keep_prob)个人理解是为了强化保留部分的x
    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__()
        # 图片分辨率 h w
        img_size = (img_size, img_size)
        # 卷积核大小
        patch_size = (patch_size, patch_size)
        self.img_size = img_size
        self.patch_size = patch_size
        # 分别计算w、h方向上的patch的个数
        self.grid_size = (img_size[0] // patch_size[0], img_size[1] // patch_size[1])
        # 一张图片的pacth个数
        self.num_patches = self.grid_size[0] * self.grid_size[1]
        # 卷积的步长巧妙地实现图片切分操作,而后与patch大小一致的卷积核完成线性映射
        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()      # nn.Identity()恒等函数 f(x)=x

    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]
        # 一维展平
        x = self.proj(x).flatten(2).transpose(1, 2)
        x = self.norm(x)
        return x

class PositionEmbs(nn.Module):
    def __init__(self, num_patches, emb_dim, dropout_rate=0.1):
        super(PositionEmbs, self).__init__()
        self.pos_embedding = nn.Parameter(torch.zeros(1, num_patches + 1, emb_dim))
        if dropout_rate > 0:
            self.dropout = nn.Dropout(dropout_rate)
        else:
            self.dropout = None
        nn.init.trunc_normal_(self.pos_embedding, std=0.02)
    def forward(self, x):
        out = x + self.pos_embedding

        if self.dropout:
            out = self.dropout(out)
        return out


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]
        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)
        # 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
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)
        # @: 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]
        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        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)
        # Multi-head attention模块
        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)
        #  MLP模块
        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__()
        # 类别个数
        self.num_classes = num_classes
        # embed_dim默认tansformer的base 768
        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
        # 源码distilled是为了其他任务,分类暂时不考虑
        self.num_tokens = 2 if distilled else 1
        # LayerNorm:对每单个batch进行的归一化
        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
        # act_layer默认tansformer的GELU
        act_layer = act_layer or nn.GELU
        # embed_layer默认是patch embedding,在其他应用中应该会有其他选择
        # patch embedding过程
        self.patch_embed = embed_layer(img_size=img_size, patch_size=patch_size, in_c=in_c, embed_dim=embed_dim)
        # patche个数
        num_patches = self.patch_embed.num_patches
        # class token
        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
        # positional embedding过程
        self.pos_embedding = PositionEmbs(num_patches, embed_dim, drop_ratio)

        # self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + self.num_tokens, embed_dim))
        # self.pos_drop = nn.Dropout(p=drop_ratio)

        # depth是Block的个数
        # 不同block层数 drop_ratio的概率不同,越深度越高
        dpr = [x.item() for x in torch.linspace(0, drop_path_ratio, depth)]  # stochastic depth decay rule
        # blocks搭建
        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
        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)
        # self.dist_token暂时可以忽略
        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_embedding(x)
        # 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=12,
                              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