论文介绍
题目：Aggregating Global Features into Local Vision Transformer
论文地址：https://arxiv.org/pdf/2201.12903
QQ深度学习交流群：994264161
扫描下方二维码，加入深度学习论文指南星球！
加入即可获得，模块缝合、制作、写作技巧，学会“结构”创新、“创新点”创新，从三区到顶会，小论文大论文，毕业一站式服务
创新点
提出了多分辨率重叠注意力模块（MOA）：论文提出了一种简单且高效的模块，称为多分辨率重叠注意力（MOA），用于在局部窗口Transformer的每个阶段之后聚合全局特征。MOA模块通过使用稍大且重叠的关键值（key-value）补丁实现邻域像素信息的传递，从而增强局部Transformer的全局信息交换能力。
改进了局部Transformer的设计：在局部Transformer中引入了全局信息交换机制，克服了传统局部Transformer无法有效建立长距离依赖的问题。MOA模块仅在每个阶段后添加，而非在每个Transformer层中嵌入，因此增加的计算成本和参数数量有限。
优化架构设计：通过大量实验，论文探索并发现了局部Transformer中关键架构组件（如窗口大小、重叠区域比例、全局注意力维度缩减比例等）的最佳配置，显著提升了模型的性能。
取得了优越的性能：在CIFAR-10、CIFAR-100和ImageNet-1K数据集上的实验结果表明，提出的MOA Transformer在参数较少的情况下性能优于许多现有的视觉Transformer。与主流Transformer（如Swin和Twin Transformer）相比，MOA Transformer实现了更高的分类准确率，同时减少了参数数量和计算复杂度。
模块化设计的通用性：MOA模块具有良好的通用性，可以方便地嵌入到其他局部Transformer中，增强其全局信息处理能力。
方法
整体架构
这篇论文提出的模型是一个基于局部Transformer的多阶段架构，每个阶段包含补丁合并层、局部窗口Transformer块和多分辨率重叠注意力模块（MOA）。模型首先将输入图像分割为固定大小的补丁，通过嵌入层投影到隐藏维度。局部Transformer块通过窗口注意力捕获局部信息，而MOA模块在阶段末尾引入全局信息和邻域像素交互。随着阶段的推进，模型逐步降低特征图的分辨率并增加特征维度，最终通过平均池化和线性层完成分类任务。这种设计有效结合了局部和全局信息，提升了分类精度，同时保持较低的计算复杂度和参数。
1. 输入处理
模型将输入的RGB图像分割成固定大小的补丁（patches），每个补丁被看作一个token。
在实验中：
对于CIFAR数据集，patch大小为
4 × 4 4 \times 4 对于ImageNet数据集，patch大小为
14 × 14 14 \times 14
2. 第一阶段：Patch嵌入层
Patch嵌入层：
使用一个线性嵌入层，将补丁的特征投影到特定维度（hidden dimension），例如96、192或384。
3. 多阶段设计
模型分为多个阶段，每个阶段包含以下组件：
（1）Patch合并层（Patch Merging Layer）
该层通过将
2 × 2 2 \times 2 例如：如果输入维度是
（2）局部Transformer块（Local Transformer Block）
每个阶段包含多个局部Transformer块：
局部窗口注意力：基于固定窗口（如
4 × 4 4 \times 4 14 × 14 14 \times 14 两层MLP：每个局部Transformer块还包括两层带有GELU激活函数的MLP。
层归一化（Layer Norm）：在多头注意力模块和MLP前应用。
残差连接（Residual Connection）：在每个模块后加入。
（3）多分辨率重叠注意力模块（MOA Module）
MOA模块在每个阶段的末尾添加：
在输入特征映射中加入一个
1 × 1 1 \times 1 对生成的查询、关键值和值向量应用标准多头自注意力机制。
实现全局信息的传播和邻域窗口间的像素信息交换。
查询（Query）补丁大小与局部窗口大小一致，且不重叠。
关键值（Key-Value）补丁大小稍大且存在重叠（如使用16×16的Key）。
设计特点：
作用：
轻量化设计：
4. 最后阶段
在最后阶段的输出后，加入以下组件：
平均池化层（Average Pooling Layer）：对输出特征进行空间维度上的池化。
线性分类层（Linear Layer）：将池化后的特征用于生成分类结果。
即插即用模块作用
MOA 作为一个即插即用模块：
局部Transformer模型的改进：
MOA模块适合用于局部窗口注意力机制（如Swin Transformer）的模型，弥补其全局信息缺乏的问题。
例如，在图像分类、目标检测、分割等视觉任务中，局部窗口Transformer因其计算效率较高而广泛使用，但其对长距离依赖的建模能力较弱，MOA模块可以在这些模型中被嵌入以增强全局特征的建模能力。
需要全局信息传播的视觉任务：
MOA特别适用于需要在全局范围内交换信息的任务，例如图像分类（ImageNet-1K、CIFAR-10/100）、场景解析或其他视觉任务。
通过在每个阶段末尾引入全局注意力，MOA模块能在保持高效计算的同时增强模型对全局上下文信息的捕获能力。
参数和计算资源有限的应用场景：
MOA模块设计轻量化，仅在每个阶段嵌入一次，计算成本和参数增加非常有限，因此适合在资源受限的设备或实时应用中部署。
消融实验结果
内容：展示了在CIFAR-100和ImageNet数据集上，使用不同窗口大小对模型准确率和参数数量的影响。
结论：
较小的窗口（如
4 × 4 4 \times 4 14 × 14 14 \times 14 窗口大小影响模型的序列长度，从而直接影响计算成本和准确率之间的权衡。
内容：分析了MOA模块中关键值（Key-Value）补丁的重叠比例对CIFAR-100数据集准确率和参数数量的影响。
结论：
随着重叠比例的减少（例如从66%到17%），模型性能逐渐提升。
较小的重叠比例（17%）即可实现高准确率，同时减少了参数数量和计算复杂度，表明少量邻域信息的交换即可显著提升性能。
内容：探讨了MOA模块中全局注意力维度的降维比例
R R 结论：
降维比例
R = 32 R = 32 过高或过低的降维比例会导致性能下降，说明合理的维度缩减对性能优化至关重要。
即插即用模块
import torch
import torch.nn as nn
import torch.utils.checkpoint as checkpoint
from timm.models.layers import DropPath, to_2tuple, trunc_normal_
from einops.layers.torch import Rearrange, Reduce
# 论文地址：https://arxiv.org/pdf/2201.12903
# 论文：Aggregating Global Features into Local Vision Transformer
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
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
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
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.query_size = self.window_size
self.key_size = self.window_size[0] * 2
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):
"""
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)
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}'
return f'dim={self.dim}, 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
class GlobalAttention(nn.Module):
r""" MOA - multi-head self attention (W-MSA) module with relative position bias.
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, input_resolution,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.query_size = self.window_size[0]
self.key_size = self.window_size[0] + 2
h,w = input_resolution
self.seq_len = h//self.query_size
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = qk_scale or head_dim ** -0.5
self.reduction = 32
self.pre_conv = nn.Conv2d(dim, int(dim//self.reduction), 1)
# define a parameter table of relative position bias
self.relative_position_bias_table = nn.Parameter(
torch.zeros((2 * self.seq_len - 1) * (2 * self.seq_len - 1), num_heads)) # 2*Wh-1 * 2*Ww-1, nH
#print(self.relative_position_bias_table.shape)
# get pair-wise relative position index for each token inside the window
coords_h = torch.arange(self.seq_len)
coords_w = torch.arange(self.seq_len)
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.seq_len - 1 # shift to start from 0
relative_coords[:, :, 1] += self.seq_len - 1
relative_coords[:, :, 0] *= 2 * self.seq_len - 1
relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww
self.register_buffer("relative_position_index", relative_position_index)
self.queryembedding = Rearrange('b c (h p1) (w p2) -> b (p1 p2 c) h w', p1 = self.query_size, p2 = self. query_size)
self.keyembedding = nn.Unfold(kernel_size=(self.key_size, self.key_size), stride = 14, padding=1)
self.query_dim = int(dim//self.reduction) * self.query_size * self.query_size
self.key_dim = int(dim//self.reduction) * self.key_size * self.key_size
self.q = nn.Linear(self.query_dim, self.dim,bias=qkv_bias)
self.kv = nn.Linear(self.key_dim, 2*self.dim,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, H, W):
"""
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, H, W, C = x.shape
B,_, C = x.shape
x = x.reshape(-1, C, H, W)
x = self.pre_conv(x)
query = self.queryembedding(x).view(B,-1,self.query_dim)
query = self.q(query)
B,N,C = query.size()
q = query.reshape(B,N,self.num_heads, C//self.num_heads).permute(0,2,1,3)
key = self.keyembedding(x).view(B,-1,self.key_dim)
kv = self.kv(key).reshape(B,N,2,self.num_heads,C//self.num_heads).permute(2,0,3,1,4)
k = kv[0]
v = kv[1]
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.seq_len * self.seq_len, self.seq_len * self.seq_len, -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)
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
class LocalTransformerBlock(nn.Module):
r""" Local 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
"""
def __init__(self, dim, input_resolution, num_heads, window_size=7,
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):
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.num_heads = num_heads
self.window_size = window_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.window_size = min(self.input_resolution)
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)
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)
x_windows = window_partition(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
attn_windows = self.attn(x_windows) # nW*B, window_size*window_size, C
attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
x = window_reverse(attn_windows, self.window_size, H, W) # B H' W' C
x = x.view(B, H * W, C)
x = shortcut + self.drop_path(x)
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}, 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
class PatchMerging(nn.Module):
""" 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
def extra_repr(self) -> str:
return f"input_resolution={self.input_resolution}, dim={self.dim}"
def flops(self):
H, W = self.input_resolution
flops = H * W * self.dim
flops += (H // 2) * (W // 2) * 4 * self.dim * 2 * self.dim
return flops
class 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.
"""
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, drop_path_global=0., use_checkpoint=False):
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.depth = depth
self.use_checkpoint = use_checkpoint
self.window_size = window_size
self.drop_path_gl = DropPath(drop_path_global) if drop_path_global > 0. else nn.Identity()
# build blocks
self.blocks = nn.ModuleList([
LocalTransformerBlock(dim=dim, input_resolution=input_resolution,
num_heads=num_heads, window_size=window_size,
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)
for i in range(depth)])
# patch merging layer
if downsample is not None:
if min(self.input_resolution) >= self.window_size:
self.glb_attn = GlobalAttention(dim, to_2tuple(window_size), self.input_resolution, num_heads = num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
self.post_conv = nn.Conv2d(dim, dim, 3, padding=1)
self.norm1 = norm_layer(dim)
self.norm2 = norm_layer(dim)
else:
self.post_conv = None
self.glb_attn = None
self.norm1 = None
self.norm2 = 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:
if min(self.input_resolution) >= self.window_size:
shortcut = x
x = self.norm1(x)
H, W = self.input_resolution
B,_,C = x.size()
no_window = int(H*W/self.window_size**2)
local_attn = x.view(B,no_window,self.window_size, self.window_size,C)
glb_attn = self.glb_attn(x, H, W)
glb_attn = glb_attn.view(B,no_window,1,1,C)
x = torch.add(local_attn, glb_attn).view(B,C,H,W)
x = shortcut.view(B,C,H,W) + self.drop_path_gl(x)
x = self.norm2(x.view(B,H*W,C))
post_conv = self.drop_path_gl(self.post_conv(x.view(B,C,H,W))).view(B, H*W, C)
x = x + post_conv
x = self.downsample(x)
return x
def extra_repr(self) -> str:
return f"dim={self.dim}, input_resolution={self.input_resolution}, depth={self.depth}"
def flops(self):
flops = 0
for blk in self.blocks:
flops += blk.flops()
if self.downsample is not None:
flops += self.downsample.flops()
return flops
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
def flops(self):
Ho, Wo = self.patches_resolution
flops = Ho * Wo * self.embed_dim * self.in_chans * (self.patch_size[0] * self.patch_size[1])
if self.norm is not None:
flops += Ho * Wo * self.embed_dim
return flops
class MOATransformer(nn.Module):
r""" Swin Transformer
A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows` -
https://arxiv.org/pdf/2103.14030
Args:
img_size (int | tuple(int)): Input image size. Default 224
patch_size (int | tuple(int)): Patch size. Default: 4
in_chans (int): Number of input image channels. Default: 3
num_classes (int): Number of classes for classification head. Default: 1000
embed_dim (int): Patch embedding dimension. Default: 96
depths (tuple(int)): Depth of each Swin Transformer layer.
num_heads (tuple(int)): Number of attention heads in different layers.
window_size (int): Window size. Default: 7
mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4
qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True
qk_scale (float): Override default qk scale of head_dim ** -0.5 if set. Default: None
drop_rate (float): Dropout rate. Default: 0
attn_drop_rate (float): Attention dropout rate. Default: 0
drop_path_rate (float): Stochastic depth rate. Default: 0.1
norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.
ape (bool): If True, add absolute position embedding to the patch embedding. Default: False
patch_norm (bool): If True, add normalization after patch embedding. Default: True
use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False
"""
def __init__(self, img_size=224, patch_size=4, in_chans=3, num_classes=1000,
embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24],
window_size=7, mlp_ratio=4., qkv_bias=True, qk_scale=None,
drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1,
norm_layer=nn.LayerNorm, ape=False, patch_norm=True,
use_checkpoint=False, **kwargs):
super().__init__()
self.num_classes = num_classes
self.num_layers = len(depths)
self.embed_dim = embed_dim
self.ape = ape
self.patch_norm = patch_norm
self.num_features = int(embed_dim * 2 ** (self.num_layers - 1))
self.mlp_ratio = mlp_ratio
# split image into non-overlapping patches
self.patch_embed = PatchEmbed(
img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim,
norm_layer=norm_layer if self.patch_norm else None)
num_patches = self.patch_embed.num_patches
patches_resolution = self.patch_embed.patches_resolution
self.patches_resolution = patches_resolution
# absolute position embedding
if self.ape:
self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim))
trunc_normal_(self.absolute_pos_embed, std=.02)
self.pos_drop = nn.Dropout(p=drop_rate)
# stochastic depth
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule
dpr_global = [x.item() for x in torch.linspace(0, 0.2, len(depths)-1)]
# build layers
self.layers = nn.ModuleList()
for i_layer in range(self.num_layers):
layer = BasicLayer(dim=int(embed_dim * 2 ** i_layer),
input_resolution=(patches_resolution[0] // (2 ** i_layer),
patches_resolution[1] // (2 ** i_layer)),
depth=depths[i_layer],
num_heads=num_heads[i_layer],
window_size=window_size,
mlp_ratio=self.mlp_ratio,
qkv_bias=qkv_bias, qk_scale=qk_scale,
drop=drop_rate, attn_drop=attn_drop_rate,
drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])],
norm_layer=norm_layer,
downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
drop_path_global = (dpr_global[i_layer]) if (i_layer < self.num_layers -1) else 0,
use_checkpoint=use_checkpoint)
self.layers.append(layer)
self.norm = norm_layer(self.num_features)
self.avgpool = nn.AdaptiveAvgPool1d(1)
self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
trunc_normal_(m.weight, std=.02)
if isinstance(m, nn.Linear) and m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.LayerNorm):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
@torch.jit.ignore
def no_weight_decay(self):
return {'absolute_pos_embed'}
@torch.jit.ignore
def no_weight_decay_keywords(self):
return {'relative_position_bias_table'}
def forward_features(self, x):
x = self.patch_embed(x)
if self.ape:
x = x + self.absolute_pos_embed
x = self.pos_drop(x)
for layer in self.layers:
x = layer(x)
x = self.norm(x) # B L C
x = self.avgpool(x.transpose(1, 2)) # B C 1
x = torch.flatten(x, 1)
return x
def forward(self, x):
x = self.forward_features(x)
x = self.head(x)
return x
def flops(self):
flops = 0
flops += self.patch_embed.flops()
for i, layer in enumerate(self.layers):
flops += layer.flops()
flops += self.num_features * self.patches_resolution[0] * self.patches_resolution[1] // (2 ** self.num_layers)
flops += self.num_features * self.num_classes
return flops
# --------------------------------------------------------
# Adopted from Swin Transformer
# Modified by Krushi Patel
# print(sum(p.numel() for p in model.parameters() if p.requires_grad), 'parameters')
# --------------------------------------------------------
if __name__ == '__main__':
input=torch.randn(1,3,224,224)
model = MOATransformer(
img_size=224,
patch_size=4,
in_chans=3,
num_classes=1000,
embed_dim=96,
depths=[2, 2, 6],
num_heads=[3, 6, 12],
window_size=14,
mlp_ratio=4.,
qkv_bias=True,
qk_scale=None,
drop_rate=0.0,
drop_path_rate=0.1,
ape=False,
patch_norm=True,
use_checkpoint=False
)
output=model(input) print(output.shape)
便捷下载方式
浏览打开网址:https://github.com/ai-dawang/PlugNPlay-Modules
更多分析可见原文
