ViT的极简pytorch实现及其即插即用

先放一张ViT的网络图

可以看到是把图像分割成小块,像NLP的句子那样按顺序进入transformer,经过MLP后,输出类别。每个小块是16x16,进入Linear Projection of Flattened Patches, 在每个的开头加上cls token和位置信息,也就是position embedding。

去掉数据读取部分,直接上一个极简的ViT代码:

python 复制代码
import torch
from torch import nn

from einops import rearrange, repeat
from einops.layers.torch import Rearrange

# helpers

def pair(t):
    return t if isinstance(t, tuple) else (t, t)

# classes

class PreNorm(nn.Module):
    def __init__(self, dim, fn):
        super().__init__()
        self.norm = nn.LayerNorm(dim)
        self.fn = fn
    def forward(self, x, **kwargs):
        return self.fn(self.norm(x), **kwargs)

class FeedForward(nn.Module):
    def __init__(self, dim, hidden_dim, dropout = 0.):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(dim, hidden_dim),
            nn.GELU(),
            nn.Dropout(dropout),
            nn.Linear(hidden_dim, dim),
            nn.Dropout(dropout)
        )
    def forward(self, x):
        return self.net(x)

class Attention(nn.Module):
    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0.):
        super().__init__()
        inner_dim = dim_head *  heads
        project_out = not (heads == 1 and dim_head == dim)

        self.heads = heads
        self.scale = dim_head ** -0.5

        self.attend = nn.Softmax(dim = -1)
        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)

        self.to_out = nn.Sequential(
            nn.Linear(inner_dim, dim),
            nn.Dropout(dropout)
        ) if project_out else nn.Identity()

    def forward(self, x):
        qkv = self.to_qkv(x).chunk(3, dim = -1)## 对tensor张量分块 x :1 197 1024   qkv 最后是一个元祖,tuple,长度是3,每个元素形状:1 197 1024
        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)

        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale

        attn = self.attend(dots)

        out = torch.matmul(attn, v)
        out = rearrange(out, 'b h n d -> b n (h d)')
        return self.to_out(out)

class Transformer(nn.Module):
    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0.):
        super().__init__()
        self.layers = nn.ModuleList([])
        for _ in range(depth):
            self.layers.append(nn.ModuleList([
                PreNorm(dim, Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout)),
                PreNorm(dim, FeedForward(dim, mlp_dim, dropout = dropout))
            ]))
    def forward(self, x):
        for attn, ff in self.layers:
            x = attn(x) + x
            x = ff(x) + x
        return x

class ViT(nn.Module):
    def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
        super().__init__()
        image_height, image_width = pair(image_size)   # 224*224
        patch_height, patch_width = pair(patch_size)   # 16 * 16

        assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'

        num_patches = (image_height // patch_height) * (image_width // patch_width)
        patch_dim = channels * patch_height * patch_width
        assert pool in {'cls', 'mean'}, 'pool type must be either cls (cls token) or mean (mean pooling)'

        self.to_patch_embedding = nn.Sequential(
            # (b,3,224,224) -> (b,196,768)    14*14=196  16*16*3=768
            Rearrange('b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = patch_height, p2 = patch_width),
            nn.Linear(patch_dim, dim),    # (b,196,1024)
        )

        self.pos_embedding = nn.Parameter(torch.randn(1, num_patches + 1, dim))
        self.cls_token = nn.Parameter(torch.randn(1, 1, dim))
        self.dropout = nn.Dropout(emb_dropout)

        self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, dropout)

        self.pool = pool
        self.to_latent = nn.Identity()

        self.mlp_head = nn.Sequential(
            nn.LayerNorm(dim),
            nn.Linear(dim, num_classes)
        )

    def forward(self, img):
        x = self.to_patch_embedding(img)        # img 1 3 224 224  输出形状x : 1 196 1024
        b, n, _ = x.shape                       # 1 196

        cls_tokens = repeat(self.cls_token, '() n d -> b n d', b = b)    # (1,1,1024)
        x = torch.cat((cls_tokens, x), dim=1)   # (1,197,1024)
        x += self.pos_embedding[:, :(n + 1)]    # (1,197,1024)
        x = self.dropout(x)                     # (1,197,1024)

        x = self.transformer(x)                 # (1,197,1024)

        x = x.mean(dim = 1) if self.pool == 'mean' else x[:, 0]     # (1,1024)

        x = self.to_latent(x)      # (1,1024)
        return self.mlp_head(x)    # (1,1000)


if __name__ == '__main__':
    v = ViT(
        image_size = 224,
        patch_size = 16,
        num_classes = 1000,
        dim = 1024,
        depth = 6,
        heads = 16,
        mlp_dim = 2048,
        dropout = 0.1,
        emb_dropout = 0.1
    )

    img = torch.randn(1, 3, 224, 224)

    preds = v(img)        # (1, 1000)

    print(preds.shape)

去掉cls和最后的全连接分类头,变成即插即用的模块:

python 复制代码
import torch
from torch import nn

from einops import rearrange
from einops.layers.torch import Rearrange

# helpers

def pair(t):
    return t if isinstance(t, tuple) else (t, t)

# classes

class PreNorm(nn.Module):
    def __init__(self, dim, fn):
        super().__init__()
        self.norm = nn.LayerNorm(dim)
        self.fn = fn
    def forward(self, x, **kwargs):
        return self.fn(self.norm(x), **kwargs)

class FeedForward(nn.Module):
    def __init__(self, dim, hidden_dim, dropout = 0.):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(dim, hidden_dim),
            nn.GELU(),
            nn.Dropout(dropout),
            nn.Linear(hidden_dim, dim),
            nn.Dropout(dropout)
        )
    def forward(self, x):
        return self.net(x)

class Attention(nn.Module):
    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0.):
        super().__init__()
        inner_dim = dim_head *  heads
        project_out = not (heads == 1 and dim_head == dim)

        self.heads = heads
        self.scale = dim_head ** -0.5

        self.attend = nn.Softmax(dim = -1)
        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)

        self.to_out = nn.Sequential(
            nn.Linear(inner_dim, dim),
            nn.Dropout(dropout)
        ) if project_out else nn.Identity()

    def forward(self, x):
        qkv = self.to_qkv(x).chunk(3, dim = -1)## 对tensor张量分块 x :1 197 1024   qkv 最后是一个元祖,tuple,长度是3,每个元素形状:1 197 1024
        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)

        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale

        attn = self.attend(dots)

        out = torch.matmul(attn, v)
        out = rearrange(out, 'b h n d -> b n (h d)')
        return self.to_out(out)

class Transformer(nn.Module):
    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0.):
        super().__init__()
        self.layers = nn.ModuleList([])
        for _ in range(depth):
            self.layers.append(nn.ModuleList([
                PreNorm(dim, Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout)),
                PreNorm(dim, FeedForward(dim, mlp_dim, dropout = dropout))
            ]))
    def forward(self, x):
        for attn, ff in self.layers:
            x = attn(x) + x
            x = ff(x) + x
        return x

class ViT(nn.Module):
    def __init__(self, *, image_size, patch_size, dim = 1024, depth = 3, heads = 16, mlp_dim = 2048, dim_head = 64, dropout = 0.1, emb_dropout = 0.1):
        super().__init__()
        channels, image_height, image_width = image_size   # 256,64,80
        patch_height, patch_width = pair(patch_size)       # 4*4

        assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'

        num_patches = (image_height // patch_height) * (image_width // patch_width)     # 16*20
        patch_dim = 64 * patch_height * patch_width    # 64*8*10

        self.conv1 = nn.Conv2d(256, 64, 1)

        self.to_patch_embedding = nn.Sequential(
            # (b,64,64,80) -> (b,320,1024)    16*20=320  4*4*64=1024
            Rearrange('b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = patch_height, p2 = patch_width),
            nn.Linear(patch_dim, dim),    # (b,320,1024)
        )

        self.to_img = nn.Sequential(
            # b c (h p1) (w p2) -> (b,64,64,80)      16*20=320  4*4*64=1024
            Rearrange('b (h w) (p1 p2 c) -> b c (h p1) (w p2)', \
                      p1 = patch_height, p2 = patch_width, h = image_height // patch_height, w = image_width // patch_width),
            nn.Conv2d(64, 256, 1),      # (b,64,64,80) -> (b,256,64,80)
        )
        # 位置编码
        self.pos_embedding = nn.Parameter(torch.randn(1, num_patches, dim))
        self.dropout = nn.Dropout(emb_dropout)

        self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, dropout)

    def forward(self, img):
        x = self.conv1(img)                     # img 1 256 64 80 -> 1 64 64 80
        x = self.to_patch_embedding(x)          # 1 320 1024
        b, n, _ = x.shape                       # 1 320

        x += self.pos_embedding[:, :(n + 1)]    # (1,320,1024)
        x = self.dropout(x)                     # (1,320,1024)

        x = self.transformer(x)                 # (1,320,1024)

        x = self.to_img(x)

        return x                                # (1 256 64 80)


if __name__ == '__main__':

    v = ViT(image_size = (256,64,80), patch_size = 4)

    img = torch.randn(1, 256, 64, 80)

    preds = v(img)         # (1, 256, 64, 80)

    print(preds.shape)
相关推荐
f6m274dww2cmd27 分钟前
拆一个类目第一的20个曝光词:43% 的搜索量来自第二页
人工智能
2601_9679801138 分钟前
Python中heapq模块的各种用法实例
数据结构·python·算法·应用实例·heapq模块
大胖爱学习_MJ39 分钟前
超越向量检索:用 Spring Boot + Neo4j 手搓 GraphRAG 知识图谱记忆
java·后端·python
RobinDevNotes1 小时前
24GB Mac 做数字人:声音放本地,画面租云端
人工智能·python·macos·开发工具
user4465117917912 小时前
Agent 也能「过安检」:AgentScope 权限系统设计拆解
人工智能
烟小栀2 小时前
Windows下的深度学习环境配置参考及配置教程
人工智能
tellmewhoisi2 小时前
深度学习0:数学知识:最大似然推导过程
人工智能·深度学习·概率论
oil欧哟2 小时前
从产品经理视角拆解 DeepSeek Super-agent:双层路由与异步调度架构
人工智能
张张123y2 小时前
个人助手型 Agent 的六个工程细节:Hermes 与 OpenClaw 的审批、委派、上下文与调度
网络·人工智能·开源·后端开发·ai agent
学无止境小袁同学3 小时前
LlamaIndex Workflows 学习
python·llm·tensorflow·llamaindex·agent开发·多agent·workflows