RT-DETR改进策略【注意力机制篇】| CVPRW-2024 分层互补注意力混合层 H-RAMi 针对低质量图像的特征提取模块
一、本文介绍
本文记录的是
利用
H-RAMi
模块优化
RT-DETR
的目标检测网络模型
。
H-RAMi
结合了对来自分层编码器阶段的
多尺度注意力
的处理能力和对
语义信息
的利用能力，
有效地补偿了因下采样特征导致的像素级信息损失
。本文将其应用到
RT-DETR
中，并进行
二次创新
，使网络能够在处理具有
复杂结构或丰富语义信息的图像
时，提升对
不同尺度和不同内容的图像区域的恢复能力
。
二、H-RAMi 介绍
2.1 设计出发点
-
许多证据表明层次化网络对图像恢复（IR）任务通常不太有效，因为IR的目标是逐个预测像素值（密集预测），而缩小特征图会丢失重要的像素级信息。然而，
层次化结构有降低时间复杂度以及学习语义级和像素级特征表示的优点
。为了弥补缺点并利用优点，设计了
H - RAMi层。
2.2 原理
-
H - RAMi层通过对来自分层编码器阶段的注意力进行处理， 补偿因下采样特征导致的像素级信息损失 ，并利用语义级信息。它将不同层次阶段的多尺度注意力进行混合，重新考虑在给定输入特征图中应关注的位置和程度。
2.3 结构
-
如图c所示，
H - RAMi接收来自分层阶段1、2、3、4中最后D - RAMiT块在**层归一化（LN）**之前由MobiVari合并的注意力。它首先将混合的二维注意力（输入）的分辨率上采样到 H × W H×W H × W ，然后将它们连接并由MobiVari混合。
2.4 优势
-
提高图像恢复精度
：从图可以看出，阶段4的输出（b）在细粒度区域产生相对不清晰的边缘，这是由于像素级信息不如非层次化网络丰富。而
H - RAMi通过利用 像素级 和 语义级 信息，在（c）处重建了关注区域并产生更清晰的边界，使得重新关注的特征图（d）包含更明显的边界，从而 提高图像恢复精度 。
-
高效利用资源
：
H - RAMi在提高模型性能的同时， 所需的额外操作和参数很少 ，分别最多只占总成本的3.01%和2.25%。
论文： https://arxiv.org/pdf/2305.11474
源码： https://github.com/rami0205/RAMiT
三、HRAMi的实现代码
HRAMi
及其改进的实现代码如下：
import torch.nn as nn
import torch
import torch.nn.functional as F
class MobiVari1(nn.Module): # MobileNet v1 Variants
def __init__(self, dim, kernel_size, stride, act=nn.LeakyReLU, out_dim=None):
super(MobiVari1, self).__init__()
self.dim = dim
self.kernel_size = kernel_size
self.out_dim = out_dim or dim
self.dw_conv = nn.Conv2d(dim, dim, kernel_size, stride, kernel_size // 2, groups=dim)
self.pw_conv = nn.Conv2d(dim, self.out_dim, 1, 1, 0)
self.act = act()
def forward(self, x):
out = self.act(self.pw_conv(self.act(self.dw_conv(x)) + x))
return out + x if self.dim == self.out_dim else out
def flops(self, resolutions):
H, W = resolutions
flops = H * W * self.kernel_size * self.kernel_size * self.dim + H * W * 1 * 1 * self.dim * self.out_dim # self.dw_conv + self.pw_conv
return flops
class MobiVari2(MobiVari1): # MobileNet v2 Variants
def __init__(self, dim, kernel_size, stride, act=nn.LeakyReLU, out_dim=None, exp_factor=1.2, expand_groups=4):
super(MobiVari2, self).__init__(dim, kernel_size, stride, act, out_dim)
self.expand_groups = expand_groups
expand_dim = int(dim * exp_factor)
expand_dim = expand_dim + (expand_groups - expand_dim % expand_groups)
self.expand_dim = expand_dim
self.exp_conv = nn.Conv2d(dim, self.expand_dim, 1, 1, 0, groups=expand_groups)
self.dw_conv = nn.Conv2d(expand_dim, expand_dim, kernel_size, stride, kernel_size // 2, groups=expand_dim)
self.pw_conv = nn.Conv2d(expand_dim, self.out_dim, 1, 1, 0)
def forward(self, x):
x1 = self.act(self.exp_conv(x))
out = self.pw_conv(self.act(self.dw_conv(x1) + x1))
return out + x if self.dim == self.out_dim else out
def flops(self, resolutions):
H, W = resolutions
flops = H * W * 1 * 1 * (self.dim // self.expand_groups) * self.expand_dim # self.exp_conv
flops += H * W * self.kernel_size * self.kernel_size * self.expand_dim # self.dw_conv
flops += H * W * 1 * 1 * self.expand_dim * self.out_dim # self.pw_conv
return flops
class HRAMi(nn.Module):
def __init__(self, dim, kernel_size=3, stride=1, mv_ver=1, mv_act=nn.LeakyReLU, exp_factor=1.2, expand_groups=4):
super(HRAMi, self).__init__()
self.dim = dim
self.kernel_size = kernel_size
if mv_ver == 1:
self.mobivari = MobiVari1(dim, kernel_size, stride, act=mv_act, out_dim=dim)
elif mv_ver == 2:
self.mobivari = MobiVari2(dim, kernel_size, stride, act=mv_act, out_dim=dim,
exp_factor=2., expand_groups=1)
def forward(self, attn_list):
# for i, attn in enumerate(attn_list[:-1]):
# attn = F.pixel_shuffle(attn, 2 ** i)
# x = attn if i == 0 else torch.cat([x, attn], dim=1)
# x = torch.cat([attn_list[0], attn_list[1]], dim=1)
x = self.mobivari(attn_list)
return x
def flops(self, resolutions):
return self.mobivari.flops(resolutions)
def autopad(k, p=None, d=1): # kernel, padding, dilation
"""Pad to 'same' shape outputs."""
if d > 1:
k = d * (k - 1) + 1 if isinstance(k, int) else [d * (x - 1) + 1 for x in k] # actual kernel-size
if p is None:
p = k // 2 if isinstance(k, int) else [x // 2 for x in k] # auto-pad
return p
class Conv(nn.Module):
"""Standard convolution with args(ch_in, ch_out, kernel, stride, padding, groups, dilation, activation)."""
default_act = nn.SiLU() # default activation
def __init__(self, c1, c2, k=1, s=1, p=None, g=1, d=1, act=True):
"""Initialize Conv layer with given arguments including activation."""
super().__init__()
self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p, d), groups=g, dilation=d, bias=False)
self.bn = nn.BatchNorm2d(c2)
self.act = self.default_act if act is True else act if isinstance(act, nn.Module) else nn.Identity()
def forward(self, x):
"""Apply convolution, batch normalization and activation to input tensor."""
return self.act(self.bn(self.conv(x)))
def forward_fuse(self, x):
"""Perform transposed convolution of 2D data."""
return self.act(self.conv(x))
class ResNetBlock(nn.Module):
"""ResNet block with standard convolution layers."""
def __init__(self, c1, c2, s=1, e=4):
"""Initialize convolution with given parameters."""
super().__init__()
c3 = e * c2
self.cv1 = Conv(c1, c2, k=1, s=1, act=True)
self.cv2 = self.cv2 = Conv(c2, c2, k=3, s=s, p=1, act=True)
self.cv3 = Conv(c2, c3, k=1, act=False)
self.cv4 = HRAMi(c2)
self.shortcut = nn.Sequential(Conv(c1, c3, k=1, s=s, act=False)) if s != 1 or c1 != c3 else nn.Identity()
def forward(self, x):
"""Forward pass through the ResNet block."""
return F.relu(self.cv3(self.cv4(self.cv2(self.cv1(x)))) + self.shortcut(x))
class ResNetLayer_HRAMi(nn.Module):
"""ResNet layer with multiple ResNet blocks."""
def __init__(self, c1, c2, s=1, is_first=False, n=1, e=4):
"""Initializes the ResNetLayer given arguments."""
super().__init__()
self.is_first = is_first
if self.is_first:
self.layer = nn.Sequential(
Conv(c1, c2, k=7, s=2, p=3, act=True), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
)
else:
blocks = [ResNetBlock(c1, c2, s, e=e)]
blocks.extend([ResNetBlock(e * c2, c2, 1, e=e) for _ in range(n - 1)])
self.layer = nn.Sequential(*blocks)
def forward(self, x):
"""Forward pass through the ResNet layer."""
return self.layer(x)
四、创新模块
4.1 改进点1⭐
模块改进方法
：直接加入
HRAMi模块
（
第五节讲解添加步骤
）。
HRAMi模块
添加后如下：
4.2 改进点2⭐
模块改进方法
：基于
HRAMi模块
的
ResNetLayer
（
第五节讲解添加步骤
）。
第二种改进方法是对
RT-DETR
中的
ResNetLayer模块
进行改进，并将
HRAMi
在加入到
ResNetLayer
模块中。
改进代码如下：
在
ResNetLayer
模块中加入
HRAMi模块
，并重命名为：
ResNetLayer_HRAMi
。
class ResNetBlock(nn.Module):
"""ResNet block with standard convolution layers."""
def __init__(self, c1, c2, s=1, e=4):
"""Initialize convolution with given parameters."""
super().__init__()
c3 = e * c2
self.cv1 = Conv(c1, c2, k=1, s=1, act=True)
self.cv2 = self.cv2 = Conv(c2, c2, k=3, s=s, p=1, act=True)
self.cv3 = Conv(c2, c3, k=1, act=False)
self.cv4 = HRAMi(c2)
self.shortcut = nn.Sequential(Conv(c1, c3, k=1, s=s, act=False)) if s != 1 or c1 != c3 else nn.Identity()
def forward(self, x):
"""Forward pass through the ResNet block."""
return F.relu(self.cv3(self.cv4(self.cv2(self.cv1(x)))) + self.shortcut(x))
class ResNetLayer_HRAMi(nn.Module):
"""ResNet layer with multiple ResNet blocks."""
def __init__(self, c1, c2, s=1, is_first=False, n=1, e=4):
"""Initializes the ResNetLayer given arguments."""
super().__init__()
self.is_first = is_first
if self.is_first:
self.layer = nn.Sequential(
Conv(c1, c2, k=7, s=2, p=3, act=True), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
)
else:
blocks = [ResNetBlock(c1, c2, s, e=e)]
blocks.extend([ResNetBlock(e * c2, c2, 1, e=e) for _ in range(n - 1)])
self.layer = nn.Sequential(*blocks)
def forward(self, x):
"""Forward pass through the ResNet layer."""
return self.layer(x)
注意❗：在
第五小节
中需要声明的模块名称为：
ResNetLayer_HRAMi
。
五、添加步骤
5.1 修改一
① 在
ultralytics/nn/
目录下新建
AddModules
文件夹用于存放模块代码
② 在
AddModules
文件夹下新建
HRAMi.py
，将
第三节
中的代码粘贴到此处
5.2 修改二
在
AddModules
文件夹下新建
__init__.py
（已有则不用新建），在文件内导入模块：
from .HRAMi import *
5.3 修改三
在
ultralytics/nn/modules/tasks.py
文件中，需要在两处位置添加各模块类名称。
首先：导入模块
其次：在
parse_model函数
中注册
ResNetLayer_HRAMi
模块
最后：在
parse_model函数
中添加如下代码
elif m is HRAMi:
args = [ch[f]]
六、yaml模型文件
6.1 模型改进版本1
此处以
ultralytics/cfg/models/rt-detr/rtdetr-l.yaml
为例，在同目录下创建一个用于自己数据集训练的模型文件
rtdetr-l-HRAMi.yaml
。
将
rtdetr-l.yaml
中的内容复制到
rtdetr-l-HRAMi.yaml
文件下，修改
nc
数量等于自己数据中目标的数量。
📌 模型的修改方法是将
骨干网络
中的
HGBlock
替换成
HRAMi
。
# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-l object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/models/rtdetr
# Parameters
nc: 1 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
l: [1.00, 1.00, 1024]
backbone:
# [from, repeats, module, args]
- [-1, 1, HGStem, [32, 48]] # 0-P2/4
- [-1, 6, HGBlock, [48, 128, 3]] # stage 1
- [-1, 1, DWConv, [128, 3, 2, 1, False]] # 2-P3/8
- [-1, 6, HGBlock, [96, 512, 3]] # stage 2
- [-1, 1, DWConv, [512, 3, 2, 1, False]] # 4-P4/16
- [-1, 6, HRAMi, []] # cm, c2, k, light, shortcut
- [-1, 6, HRAMi, []]
- [-1, 6, HRAMi, []] # stage 3
- [-1, 1, DWConv, [1024, 3, 2, 1, False]] # 8-P5/32
- [-1, 6, HGBlock, [384, 2048, 5, True, False]] # stage 4
head:
- [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 10 input_proj.2
- [-1, 1, AIFI, [1024, 8]]
- [-1, 1, Conv, [256, 1, 1]] # 12, Y5, lateral_convs.0
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [7, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 14 input_proj.1
- [[-2, -1], 1, Concat, [1]]
- [-1, 3, RepC3, [256]] # 16, fpn_blocks.0
- [-1, 1, Conv, [256, 1, 1]] # 17, Y4, lateral_convs.1
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [3, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 19 input_proj.0
- [[-2, -1], 1, Concat, [1]] # cat backbone P4
- [-1, 3, RepC3, [256]] # X3 (21), fpn_blocks.1
- [-1, 1, Conv, [256, 3, 2]] # 22, downsample_convs.0
- [[-1, 17], 1, Concat, [1]] # cat Y4
- [-1, 3, RepC3, [256]] # F4 (24), pan_blocks.0
- [-1, 1, Conv, [256, 3, 2]] # 25, downsample_convs.1
- [[-1, 12], 1, Concat, [1]] # cat Y5
- [-1, 3, RepC3, [256]] # F5 (27), pan_blocks.1
- [[21, 24, 27], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)
6.2 模型改进版本2⭐
此处以
ultralytics/cfg/models/rt-detr/rtdetr-resnet50.yaml
为例，在同目录下创建一个用于自己数据集训练的模型文件
rtdetr-ResNetLayer_HRAMi.yaml
。
将
rtdetr-resnet50.yaml
中的内容复制到
rtdetr-ResNetLayer_HRAMi.yaml
文件下，修改
nc
数量等于自己数据中目标的数量。
📌 模型的修改方法是将
骨干网络
中的
ResNetLayer模块
替换成
ResNetLayer_HRAMi模块
。
# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-ResNet50 object detection model with P3-P5 outputs.
# Parameters
nc: 1 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
l: [1.00, 1.00, 1024]
backbone:
# [from, repeats, module, args]
- [-1, 1, ResNetLayer_HRAMi, [3, 64, 1, True, 1]] # 0
- [-1, 1, ResNetLayer_HRAMi, [64, 64, 1, False, 3]] # 1
- [-1, 1, ResNetLayer_HRAMi, [256, 128, 2, False, 4]] # 2
- [-1, 1, ResNetLayer_HRAMi, [512, 256, 2, False, 6]] # 3
- [-1, 1, ResNetLayer_HRAMi, [1024, 512, 2, False, 3]] # 4
head:
- [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 5
- [-1, 1, AIFI, [1024, 8]]
- [-1, 1, Conv, [256, 1, 1]] # 7
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [3, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 9
- [[-2, -1], 1, Concat, [1]]
- [-1, 3, RepC3, [256]] # 11
- [-1, 1, Conv, [256, 1, 1]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [2, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 14
- [[-2, -1], 1, Concat, [1]] # cat backbone P4
- [-1, 3, RepC3, [256]] # X3 (16), fpn_blocks.1
- [-1, 1, Conv, [256, 3, 2]] # 17, downsample_convs.0
- [[-1, 12], 1, Concat, [1]] # cat Y4
- [-1, 3, RepC3, [256]] # F4 (19), pan_blocks.0
- [-1, 1, Conv, [256, 3, 2]] # 20, downsample_convs.1
- [[-1, 7], 1, Concat, [1]] # cat Y5
- [-1, 3, RepC3, [256]] # F5 (22), pan_blocks.1
- [[16, 19, 22], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)
七、成功运行结果
打印网络模型可以看到
HRAMi
和
ResNetLayer_HRAMi
已经加入到模型中，并可以进行训练了。
rtdetr-l-HRAMi ：
rtdetr-l-HRAMi summary: 606 layers, 31,690,051 parameters, 31,690,051 gradients, 104.3 GFLOPs
from n params module arguments
0 -1 1 25248 ultralytics.nn.modules.block.HGStem [3, 32, 48]
1 -1 6 155072 ultralytics.nn.modules.block.HGBlock [48, 48, 128, 3, 6]
2 -1 1 1408 ultralytics.nn.modules.conv.DWConv [128, 128, 3, 2, 1, False]
3 -1 6 839296 ultralytics.nn.modules.block.HGBlock [128, 96, 512, 3, 6]
4 -1 1 5632 ultralytics.nn.modules.conv.DWConv [512, 512, 3, 2, 1, False]
5 -1 6 1606656 ultralytics.nn.AddModules.HRAMi.HRAMi [512]
6 -1 6 1606656 ultralytics.nn.AddModules.HRAMi.HRAMi [512]
7 -1 6 1606656 ultralytics.nn.AddModules.HRAMi.HRAMi [512]
8 -1 1 11264 ultralytics.nn.modules.conv.DWConv [512, 1024, 3, 2, 1, False]
9 -1 6 6708480 ultralytics.nn.modules.block.HGBlock [1024, 384, 2048, 5, 6, True, False]
10 -1 1 524800 ultralytics.nn.modules.conv.Conv [2048, 256, 1, 1, None, 1, 1, False]
11 -1 1 789760 ultralytics.nn.modules.transformer.AIFI [256, 1024, 8]
12 -1 1 66048 ultralytics.nn.modules.conv.Conv [256, 256, 1, 1]
13 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
14 7 1 131584 ultralytics.nn.modules.conv.Conv [512, 256, 1, 1, None, 1, 1, False]
15 [-2, -1] 1 0 ultralytics.nn.modules.conv.Concat [1]
16 -1 3 2232320 ultralytics.nn.modules.block.RepC3 [512, 256, 3]
17 -1 1 66048 ultralytics.nn.modules.conv.Conv [256, 256, 1, 1]
18 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
19 3 1 131584 ultralytics.nn.modules.conv.Conv [512, 256, 1, 1, None, 1, 1, False]
20 [-2, -1] 1 0 ultralytics.nn.modules.conv.Concat [1]
21 -1 3 2232320 ultralytics.nn.modules.block.RepC3 [512, 256, 3]
22 -1 1 590336 ultralytics.nn.modules.conv.Conv [256, 256, 3, 2]
23 [-1, 17] 1 0 ultralytics.nn.modules.conv.Concat [1]
24 -1 3 2232320 ultralytics.nn.modules.block.RepC3 [512, 256, 3]
25 -1 1 590336 ultralytics.nn.modules.conv.Conv [256, 256, 3, 2]
26 [-1, 12] 1 0 ultralytics.nn.modules.conv.Concat [1]
27 -1 3 2232320 ultralytics.nn.modules.block.RepC3 [512, 256, 3]
28 [21, 24, 27] 1 7303907 ultralytics.nn.modules.head.RTDETRDecoder [1, [256, 256, 256]]
rtdetr-l-HRAMi summary: 606 layers, 31,690,051 parameters, 31,690,051 gradients, 104.3 GFLOPs
rtdetr-ResNetLayer_HRAMi ：
rtdetr-ResNetLayer_HRAMi summary: 673 layers, 44,061,795 parameters, 44,061,795 gradients, 134.0 GFLOPs
from n params module arguments
0 -1 1 9536 ultralytics.nn.AddModules.HRAMi.ResNetLayer_HRAMi[3, 64, 1, True, 1]
1 -1 1 230208 ultralytics.nn.AddModules.HRAMi.ResNetLayer_HRAMi[64, 64, 1, False, 3]
2 -1 1 1290752 ultralytics.nn.AddModules.HRAMi.ResNetLayer_HRAMi[256, 128, 2, False, 4]
3 -1 1 7508480 ultralytics.nn.AddModules.HRAMi.ResNetLayer_HRAMi[512, 256, 2, False, 6]
4 -1 1 15768064 ultralytics.nn.AddModules.HRAMi.ResNetLayer_HRAMi[1024, 512, 2, False, 3]
5 -1 1 524800 ultralytics.nn.modules.conv.Conv [2048, 256, 1, 1, None, 1, 1, False]
6 -1 1 789760 ultralytics.nn.modules.transformer.AIFI [256, 1024, 8]
7 -1 1 66048 ultralytics.nn.modules.conv.Conv [256, 256, 1, 1]
8 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
9 3 1 262656 ultralytics.nn.modules.conv.Conv [1024, 256, 1, 1, None, 1, 1, False]
10 [-2, -1] 1 0 ultralytics.nn.modules.conv.Concat [1]
11 -1 3 2232320 ultralytics.nn.modules.block.RepC3 [512, 256, 3]
12 -1 1 66048 ultralytics.nn.modules.conv.Conv [256, 256, 1, 1]
13 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
14 2 1 131584 ultralytics.nn.modules.conv.Conv [512, 256, 1, 1, None, 1, 1, False]
15 [-2, -1] 1 0 ultralytics.nn.modules.conv.Concat [1]
16 -1 3 2232320 ultralytics.nn.modules.block.RepC3 [512, 256, 3]
17 -1 1 590336 ultralytics.nn.modules.conv.Conv [256, 256, 3, 2]
18 [-1, 12] 1 0 ultralytics.nn.modules.conv.Concat [1]
19 -1 3 2232320 ultralytics.nn.modules.block.RepC3 [512, 256, 3]
20 -1 1 590336 ultralytics.nn.modules.conv.Conv [256, 256, 3, 2]
21 [-1, 7] 1 0 ultralytics.nn.modules.conv.Concat [1]
22 -1 3 2232320 ultralytics.nn.modules.block.RepC3 [512, 256, 3]
23 [16, 19, 22] 1 7303907 ultralytics.nn.modules.head.RTDETRDecoder [1, [256, 256, 256]]
rtdetr-ResNetLayer_HRAMi summary: 673 layers, 44,061,795 parameters, 44,061,795 gradients, 134.0 GFLOPs