论文介绍
题目：D-Net: Dynamic Large Kernel with Dynamic Feature Fusion for Volumetric Medical Image Segmentation
论文地址：https://arxiv.org/abs/2403.10674
QQ深度学习交流群：719278780
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
提出了空间和频域融合网络（SFFNet）框架：该框架采用两阶段网络设计，第一阶段使用空间方法提取特征，获取具有丰富空间细节和语义信息的特征；第二阶段将这些特征映射到空间和频域中，以充分利用不同域的信息。
引入了小波变换特征分解器（WTFD）：在频域映射中，作者设计了WTFD结构，利用Haar小波变换将特征分解为低频和高频分量，并将其与空间特征融合，增强了模型对灰度变化显著区域的处理能力。
设计了多尺度双重表示对齐滤波器（MDAF）：为弥合频域和空间特征之间的语义差距，促进不同表示域特征的有效组合，作者提出了MDAF结构，利用多尺度卷积和双交叉注意机制，实现特征的对齐和选择，提升了分割性能。
方法
整体结构
编码器-解码器框架：SFFNet使用了经典的编码器-解码器结构，其中编码器负责提取空间特征，解码器负责逐步恢复图像的空间分辨率，从而获得精确的分割结果。
小波变换特征分解器（WTFD）：在编码器部分，模型将提取的空间特征通过小波变换特征分解器进行分解。WTFD利用Haar小波将特征分解为低频和高频成分，从而更好地保留图像中的灰度变化和纹理信息。这样一来，模型能够捕获更多细节信息，有助于提升分割精度。
多尺度双重表示对齐滤波器（MDAF）：在解码器部分，为了有效融合空间特征和频域特征，模型引入了MDAF模块。MDAF通过多尺度卷积和双交叉注意机制，将空间和频域信息对齐并融合。MDAF的设计使得不同尺度和不同域的特征能够被有效结合，从而提升模型在复杂场景中的分割性能。
频域与空间域的融合：模型通过两阶段网络的设计，将空间和频域信息融合。第一阶段模型先提取空间特征，在第二阶段引入频域特征，通过小波变换和多尺度滤波，增强了对细节信息的关注和对复杂背景的适应能力。
即插即用模块作用
FMS 作为一个即插即用模块，主要适用于：
遥感图像分割：FMS在处理遥感图像时，能够有效提升对复杂地物和纹理细节的捕捉能力。通过频域特征提取和多尺度选择，它能增强模型在复杂场景中的分割性能。
复杂背景的目标检测：在背景复杂或目标与背景对比度较低的场景中，FMS模块可以帮助模型识别并区分细微的目标信息。通过融合多尺度和频域特征，FMS提高了模型在细节识别方面的准确性。
多尺度信息融合：FMS利用频域和空间域特征的结合，可以有效融合不同尺度的信息，从而更好地捕捉图像中的全局和局部细节，增强模型的鲁棒性和泛化性。
消融实验结果
该表通过逐步移除模型的关键组件，评估每个模块对整体性能的影响。实验结果表明，移除小波变换特征分解器（WTFD）或多尺度双重表示对齐滤波器（MDAF）都会导致模型性能下降，这验证了这些模块在提升模型性能中的重要作用。
即插即用模块
import torch
import torch.nn as nn
import torch.nn.functional as F
from pytorch_wavelets import DWTForward
from einops import rearrange
from timm.models.layers import DropPath, trunc_normal_
# 论文地址：https://arxiv.org/pdf/2405.01992
# 论文：SFFNet: A Wavelet-Based Spatial and Frequency Domain Fusion Network for Remote Sensing Segmentation, arxiv2405
class Bconv(nn.Module):
def __init__(self, ch_in, ch_out, k, s):
'''
:param ch_in: 输入通道数
:param ch_out: 输出通道数
:param k: 卷积核尺寸
:param s: 步长
:return:
'''
super(Bconv, self).__init__()
self.conv = nn.Conv2d(ch_in, ch_out, k, s, padding=k // 2)
self.bn = nn.BatchNorm2d(ch_out)
self.act = nn.SiLU()
def forward(self, x):
'''
:param x: 输入
:return:
'''
return self.act(self.bn(self.conv(x)))
class SppCSPC(nn.Module):
def __init__(self, ch_in, ch_out):
'''
:param ch_in: 输入通道
:param ch_out: 输出通道
'''
super(SppCSPC, self).__init__()
# 分支一
self.conv1 = nn.Sequential(
Bconv(ch_in, ch_out, 1, 1),
Bconv(ch_out, ch_out, 3, 1),
Bconv(ch_out, ch_out, 1, 1)
)
# 分支二（SPP）
self.mp1 = nn.MaxPool2d(5, 1, 5 // 2) # 卷积核为5的池化
self.mp2 = nn.MaxPool2d(9, 1, 9 // 2) # 卷积核为9的池化
self.mp3 = nn.MaxPool2d(13, 1, 13 // 2) # 卷积核为13的池化
# concat之后的卷积
self.conv1_2 = nn.Sequential(
Bconv(4 * ch_out, ch_out, 1, 1),
Bconv(ch_out, ch_out, 3, 1)
)
# 分支三
self.conv3 = Bconv(ch_in, ch_out, 1, 1)
# 此模块最后一层卷积
self.conv4 = Bconv(2 * ch_out, ch_out, 1, 1)
def forward(self, x):
# 分支一输出
output1 = self.conv1(x)
# 分支二池化层的各个输出
mp_output1 = self.mp1(output1)
mp_output2 = self.mp2(output1)
mp_output3 = self.mp3(output1)
# 合并以上并进行卷积
result1 = self.conv1_2(torch.cat((output1, mp_output1, mp_output2, mp_output3), dim=1))
# 分支三
result2 = self.conv3(x)
return self.conv4(torch.cat((result1, result2), dim=1))
class ConvBNReLU(nn.Sequential):
def __init__(self, in_channels, out_channels, kernel_size=3, dilation=1, stride=1, norm_layer=nn.BatchNorm2d, bias=False):
super(ConvBNReLU, self).__init__(
nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, bias=bias,
dilation=dilation, stride=stride, padding=((stride - 1) + dilation * (kernel_size - 1)) // 2),
norm_layer(out_channels),
nn.ReLU6()
)
class ConvBN(nn.Sequential):
def __init__(self, in_channels, out_channels, kernel_size=3, dilation=1, stride=1, norm_layer=nn.BatchNorm2d, bias=False):
super(ConvBN, self).__init__(
nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, bias=bias,
dilation=dilation, stride=stride, padding=((stride - 1) + dilation * (kernel_size - 1)) // 2),
norm_layer(out_channels)
)
class Conv(nn.Sequential):
def __init__(self, in_channels, out_channels, kernel_size=3, dilation=1, stride=1, bias=False):
super(Conv, self).__init__(
nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, bias=bias,
dilation=dilation, stride=stride, padding=((stride - 1) + dilation * (kernel_size - 1)) // 2)
)
class SeparableConvBNReLU(nn.Sequential):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1, dilation=1,
norm_layer=nn.BatchNorm2d):
super(SeparableConvBNReLU, self).__init__(
nn.Conv2d(in_channels, in_channels, kernel_size, stride=stride, dilation=dilation,
padding=((stride - 1) + dilation * (kernel_size - 1)) // 2,
groups=in_channels, bias=False),
norm_layer(out_channels),
nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False),
nn.ReLU6()
)
class SeparableConvBN(nn.Sequential):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1, dilation=1,
norm_layer=nn.BatchNorm2d):
super(SeparableConvBN, self).__init__(
nn.Conv2d(in_channels, in_channels, kernel_size, stride=stride, dilation=dilation,
padding=((stride - 1) + dilation * (kernel_size - 1)) // 2,
groups=in_channels, bias=False),
norm_layer(out_channels),
nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False)
)
class SeparableConv(nn.Sequential):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1, dilation=1):
super(SeparableConv, self).__init__(
nn.Conv2d(in_channels, in_channels, kernel_size, stride=stride, dilation=dilation,
padding=((stride - 1) + dilation * (kernel_size - 1)) // 2,
groups=in_channels, bias=False),
nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False)
)
class Mlp(nn.Module):
def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.ReLU6, drop=0.):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Conv2d(in_features, hidden_features, 1, 1, 0, bias=True)
self.act = act_layer()
self.fc2 = nn.Conv2d(hidden_features, out_features, 1, 1, 0, bias=True)
self.drop = nn.Dropout(drop, inplace=True)
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 GlobalAttention(nn.Module):
def __init__(self,
dim=256,
num_heads=16,
qkv_bias=False,
window_size=8,
relative_pos_embedding=True
):
super().__init__()
self.num_heads = num_heads
head_dim = dim // self.num_heads
self.scale = head_dim ** -0.5
self.ws = window_size
self.qkv = Conv(dim, 3*dim, kernel_size=1, bias=qkv_bias)
self.proj = SeparableConvBN(dim, dim, kernel_size=window_size)
self.attn_x = nn.Conv2d(dim,dim,kernel_size=(window_size, 1), stride=1, padding=(window_size//2 - 1, 0))
self.attn_y = nn.Conv2d(dim,dim,kernel_size=(1, window_size), stride=1, padding=(0, window_size//2 - 1))
self.relative_pos_embedding = relative_pos_embedding
if self.relative_pos_embedding:
# define a parameter table of relative position bias
self.relative_position_bias_table = nn.Parameter(
torch.zeros((2 * window_size - 1) * (2 * window_size - 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.ws)
coords_w = torch.arange(self.ws)
coords = torch.stack(torch.meshgrid([coords_h, coords_w], indexing='ij')) # 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.ws - 1 # shift to start from 0
relative_coords[:, :, 1] += self.ws - 1
relative_coords[:, :, 0] *= 2 * self.ws - 1
relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww
self.register_buffer("relative_position_index", relative_position_index)
trunc_normal_(self.relative_position_bias_table, std=.02)
def pad(self, x, ps):
_, _, H, W = x.size()
if W % ps != 0:
x = F.pad(x, (0, ps - W % ps), mode='reflect')
if H % ps != 0:
x = F.pad(x, (0, 0, 0, ps - H % ps), mode='reflect')
return x
def pad_out(self, x):
x = F.pad(x, pad=(0, 1, 0, 1), mode='reflect')
return x
def forward(self, x):
B, C, H, W = x.shape
x = self.pad(x, self.ws)
B, C, Hp, Wp = x.shape
qkv = self.qkv(x)
q, k, v = rearrange(qkv, 'b (qkv h d) (hh ws1) (ww ws2) -> qkv (b hh ww) h (ws1 ws2) d', h=self.num_heads,
d=C//self.num_heads, hh=Hp//self.ws, ww=Wp//self.ws, qkv=3, ws1=self.ws, ws2=self.ws)
dots = (q @ k.transpose(-2, -1)) * self.scale
if self.relative_pos_embedding:
relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
self.ws * self.ws, self.ws * self.ws, -1) # Wh*Ww,Wh*Ww,nH
relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww
dots += relative_position_bias.unsqueeze(0)
attn = dots.softmax(dim=-1)
attn = attn @ v
attn = rearrange(attn, '(b hh ww) h (ws1 ws2) d -> b (h d) (hh ws1) (ww ws2)', h=self.num_heads,
d=C//self.num_heads, hh=Hp//self.ws, ww=Wp//self.ws, ws1=self.ws, ws2=self.ws)
attn = attn[:, :, :H, :W]
out = self.attn_x(F.pad(attn, pad=(0, 0, 0, 1), mode='reflect')) + \
self.attn_y(F.pad(attn, pad=(0, 1, 0, 0), mode='reflect'))
out = self.pad_out(out)
out = self.proj(out)
# print(out.size())
out = out[:, :, :H, :W]
return out
class LocalAttention(nn.Module):
def __init__(self,
dim=256,
window_size=8,
):
super().__init__()
self.local = SppCSPC(dim,dim)
# self.bam = BAM(gate_channel=dim)
self.proj = SeparableConvBN(dim, dim, kernel_size=window_size)
def pad(self, x, ps):
_, _, H, W = x.size()
if W % ps != 0:
x = F.pad(x, (0, ps - W % ps), mode='reflect')
if H % ps != 0:
x = F.pad(x, (0, 0, 0, ps - H % ps), mode='reflect')
return x
def pad_out(self, x):
x = F.pad(x, pad=(0, 1, 0, 1), mode='reflect')
return x
def forward(self, x):
B, C, H, W = x.shape
local = self.local(x)
out = self.pad_out(local)
out = self.proj(out)
out = out[:, :, :H, :W]
return out
class LocalBlock(nn.Module):
expansion = 1
def __init__(self, dim=256, num_heads=16, mlp_ratio=4., qkv_bias=False, drop=0., attn_drop=0.,
drop_path=0., act_layer=nn.ReLU6, norm_layer=nn.BatchNorm2d, window_size=8,C=0,H=0,W=0):
super().__init__()
self.norm1 = norm_layer(dim)
self.attn =LocalAttention(dim,window_size=window_size)
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, out_features=dim, act_layer=act_layer, drop=drop)
self.norm2 = norm_layer(dim)
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 multilocalBlock(nn.Module):
expansion = 1
def __init__(self,dim=256,outdim=256, num_heads=16, mlp_ratio=4., qkv_bias=False, drop=0., attn_drop=0.,
drop_path=0., act_layer=nn.ReLU6, norm_layer=nn.BatchNorm2d, window_size=8,C=0,H=0,W=0):
super().__init__()
self.down = Conv(dim,outdim,kernel_size=3,stride=2,dilation=1,bias=False)
self.norm1 = norm_layer(outdim)
self.attn =LocalAttention(outdim,window_size=window_size)
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
# mlp_hidden_dim = int(dim * mlp_ratio)
# self.mlp = Mlp(in_features=outdim, hidden_features=mlp_hidden_dim, out_features=outdim, act_layer=act_layer, drop=drop)
self.norm2 = norm_layer(outdim)
def forward(self, x):
x = self.down(x)
x = x + self.drop_path(self.attn(self.norm1(x)))
x = self.drop_path(self.norm2(x))
return x
class GlobalBlock(nn.Module):
expansion = 1
def __init__(self, dim=256, num_heads=16, mlp_ratio=4., qkv_bias=False, drop=0., attn_drop=0.,
drop_path=0., act_layer=nn.ReLU6, norm_layer=nn.BatchNorm2d, window_size=8,C=0,H=0,W=0):
super().__init__()
self.norm1 = norm_layer(dim)
self.attn = GlobalAttention(dim, num_heads=num_heads, qkv_bias=qkv_bias, window_size=window_size)
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, out_features=dim, act_layer=act_layer, drop=drop)
self.norm2 = norm_layer(dim)
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 GlBlock(nn.Module):
expansion = 1
def __init__(self, dim=256,outdim = 256, num_heads=16, mlp_ratio=4., qkv_bias=False, drop=0., attn_drop=0.,
drop_path=0., act_layer=nn.ReLU6, norm_layer=nn.BatchNorm2d, window_size=8,C=0,H=0,W=0):
super().__init__()
self.norm1 = norm_layer(dim)
self.attn = GlobalAttention(dim, num_heads=num_heads, qkv_bias=qkv_bias, window_size=window_size)
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
# mlp_hidden_dim = int(dim * mlp_ratio)
# self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, out_features=dim, act_layer=act_layer, drop=drop)
self.norm2 = norm_layer(dim)
self.down = Conv(dim, outdim, kernel_size=3, stride=2, dilation=1, bias=False)
def forward(self, x):
x = self.down(x)
x = x + self.drop_path(self.attn(self.norm1(x)))
x = self.norm2(x)
return x
# feature mapping stage(FMS)
class FMS(nn.Module):
def __init__(self, in_ch, out_ch,num_heads=8, window_size=8):
super(FMS, self).__init__()
self.wt = DWTForward(J=1, mode='zero', wave='haar')
self.glb = GlBlock(dim=in_ch,outdim=in_ch,num_heads=num_heads, window_size=window_size)
self.localb=multilocalBlock(dim=in_ch,outdim=in_ch,num_heads=8, window_size=window_size)
self.conv_bn_relu = nn.Sequential(
nn.Conv2d(in_ch*3, in_ch, kernel_size=1, stride=1),
nn.BatchNorm2d(in_ch),
nn.ReLU(inplace=True),
)
self.outconv_bn_relu_L = nn.Sequential(
nn.Conv2d(in_ch, out_ch, kernel_size=1, stride=1),
nn.BatchNorm2d(out_ch),
nn.ReLU(inplace=True),
)
self.outconv_bn_relu_H = nn.Sequential(
nn.Conv2d(in_ch, out_ch, kernel_size=1, stride=1),
nn.BatchNorm2d(out_ch),
nn.ReLU(inplace=True),
)
self.outconv_bn_relu_glb = nn.Sequential(
nn.Conv2d(in_ch, out_ch, kernel_size=1, stride=1),
nn.BatchNorm2d(out_ch),
nn.ReLU(inplace=True),
)
self.outconv_bn_relu_local = nn.Sequential(
nn.Conv2d(in_ch, out_ch, kernel_size=1, stride=1),
nn.BatchNorm2d(out_ch),
nn.ReLU(inplace=True),
)
def forward(self, x,imagename=None):
yL, yH = self.wt(x)
y_HL = yH[0][:,:,0,::]
y_LH = yH[0][:,:,1,::]
y_HH = yH[0][:,:,2,::]
yH = torch.cat([y_HL, y_LH, y_HH], dim=1)
yH = self.conv_bn_relu(yH)
yL = self.outconv_bn_relu_L(yL)
yH = self.outconv_bn_relu_H(yH)
glb = self.outconv_bn_relu_glb(self.glb(x))
local = self.outconv_bn_relu_local(self.localb(x))
return yL,yH,glb,local
if __name__ == '__main__':
block = FMS(in_ch=64, out_ch=128)
input = torch.randn(1, 64, 256, 256)
# 前向传播
yL, yH, glb, local = block(input)
# 打印输入和输出的形状
print(input.size())
print(yL.size())
print(yH.size())
print(glb.size()) print(local.size())
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
