YOLO26改进| 特征融合 | 双向指导 + 自适应权重 + 低光降噪增强【CVPR】
·
💡💡💡本专栏所有程序均经过测试，可成功执行💡💡💡
本文给大家带来的教程是将YOLO26的特征融合替换为LCA来提取特征。文章在介绍主要的原理后，将手把手教学如何进行模块的代码添加和修改，并将修改后的完整代码放在文章的最后，方便大家一键运行，小白也可轻松上手实践。以帮助您更好地学习深度学习目标检测YOLO系列的挑战。
目录
1.论文
论文地址：HVI: ANewColor Space for Low-light Image Enhancement
官方代码：官方代码仓库点击即可跳转
2. LCA代码实现
2.1 将LCA添加到YOLO26中
关键步骤一：在ultralytics\ultralytics\nn\modules下面新建文件夹models，在文件夹下新建LCA.py，粘贴下面代码
import torch
import torch.nn as nn
import torch.functional as F
from einops import rearrange
from ultralytics.nn.modules.conv import Conv
class LayerNorm(nn.Module):
r""" LayerNorm that supports two data formats: channels_last (default) or channels_first.
The ordering of the dimensions in the inputs. channels_last corresponds to inputs with
shape (batch_size, height, width, channels) while channels_first corresponds to inputs
with shape (batch_size, channels, height, width).
"""
def __init__(self, normalized_shape, eps=1e-6, data_format="channels_first"):
super().__init__()
self.weight = nn.Parameter(torch.ones(normalized_shape))
self.bias = nn.Parameter(torch.zeros(normalized_shape))
self.eps = eps
self.data_format = data_format
if self.data_format not in ["channels_last", "channels_first"]:
raise NotImplementedError
self.normalized_shape = (normalized_shape, )
def forward(self, x):
if self.data_format == "channels_last":
return F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps)
elif self.data_format == "channels_first":
u = x.mean(1, keepdim=True)
s = (x - u).pow(2).mean(1, keepdim=True)
x = (x - u) / torch.sqrt(s + self.eps)
x = self.weight[:, None, None] * x + self.bias[:, None, None]
return x
class CAB(nn.Module):
def __init__(self, dim, num_heads, bias):
super(CAB, self).__init__()
self.num_heads = num_heads
self.temperature = nn.Parameter(torch.ones(num_heads, 1, 1))
self.q = nn.Conv2d(dim, dim, kernel_size=1, bias=bias)
self.q_dwconv = nn.Conv2d(dim, dim, kernel_size=3, stride=1, padding=1, groups=dim, bias=bias)
self.kv = nn.Conv2d(dim, dim*2, kernel_size=1, bias=bias)
self.kv_dwconv = nn.Conv2d(dim*2, dim*2, kernel_size=3, stride=1, padding=1, groups=dim*2, bias=bias)
self.project_out = nn.Conv2d(dim, dim, kernel_size=1, bias=bias)
def forward(self, x, y):
b, c, h, w = x.shape
q = self.q_dwconv(self.q(x))
kv = self.kv_dwconv(self.kv(y))
k, v = kv.chunk(2, dim=1)
q = rearrange(q, 'b (head c) h w -> b head c (h w)', head=self.num_heads)
k = rearrange(k, 'b (head c) h w -> b head c (h w)', head=self.num_heads)
v = rearrange(v, 'b (head c) h w -> b head c (h w)', head=self.num_heads)
q = torch.nn.functional.normalize(q, dim=-1)
k = torch.nn.functional.normalize(k, dim=-1)
attn = (q @ k.transpose(-2, -1)) * self.temperature
attn = nn.functional.softmax(attn,dim=-1)
out = (attn @ v)
out = rearrange(out, 'b head c (h w) -> b (head c) h w', head=self.num_heads, h=h, w=w)
out = self.project_out(out)
return out
class IEL(nn.Module):
def __init__(self, dim, ffn_expansion_factor=2.66, bias=False):
super(IEL, self).__init__()
hidden_features = int(dim*ffn_expansion_factor)
self.project_in = nn.Conv2d(dim, hidden_features*2, kernel_size=1, bias=bias)
self.dwconv = nn.Conv2d(hidden_features*2, hidden_features*2, kernel_size=3, stride=1, padding=1, groups=hidden_features*2, bias=bias)
self.dwconv1 = nn.Conv2d(hidden_features, hidden_features, kernel_size=3, stride=1, padding=1, groups=hidden_features, bias=bias)
self.dwconv2 = nn.Conv2d(hidden_features, hidden_features, kernel_size=3, stride=1, padding=1, groups=hidden_features, bias=bias)
self.project_out = nn.Conv2d(hidden_features, dim, kernel_size=1, bias=bias)
self.Tanh = nn.Tanh()
def forward(self, x):
x = self.project_in(x)
x1, x2 = self.dwconv(x).chunk(2, dim=1)
x1 = self.Tanh(self.dwconv1(x1)) + x1
x2 = self.Tanh(self.dwconv2(x2)) + x2
x = x1 * x2
x = self.project_out(x)
return x
class LCA(nn.Module):
def __init__(self, in_dim, out_dim, num_heads=8, bias=False):
super(LCA, self).__init__()
self.norm = LayerNorm(out_dim)
self.gdfn = IEL(out_dim)
self.ffn = CAB(out_dim, num_heads, bias=bias)
self.conv1x1 = nn.ModuleList([])
for i in in_dim:
if i != out_dim:
self.conv1x1.append(Conv(i, out_dim, 1))
else:
self.conv1x1.append(nn.Identity())
def forward(self, inputs):
x, y = inputs
x = self.conv1x1[0](x)
y = self.conv1x1[1](y)
x = x + self.ffn(self.norm(x),self.norm(y))
x = x + self.gdfn(self.norm(x))
return x
2.2 更改init.py文件
关键步骤二：在文件ultralytics\ultralytics\nn\modules\models文件夹下新建__init__.py文件，先导入函数
然后在下面的__all__中声明函数
2.3 添加yaml文件
关键步骤三：在/ultralytics/ultralytics/cfg/models/26下面新建文件yolo26_LCA.yaml文件，粘贴下面的内容
- 目标检测
# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
# Ultralytics YOLO26 object detection model with P3/8 - P5/32 outputs
# Model docs: https://docs.ultralytics.com/models/yolo26
# Task docs: https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
end2end: True # whether to use end-to-end mode
reg_max: 1 # DFL bins
scales: # model compound scaling constants, i.e. 'model=yolo26n.yaml' will call yolo26.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.50, 0.25, 1024] # summary: 260 layers, 2,572,280 parameters, 2,572,280 gradients, 6.1 GFLOPs
s: [0.50, 0.50, 1024] # summary: 260 layers, 10,009,784 parameters, 10,009,784 gradients, 22.8 GFLOPs
m: [0.50, 1.00, 512] # summary: 280 layers, 21,896,248 parameters, 21,896,248 gradients, 75.4 GFLOPs
l: [1.00, 1.00, 512] # summary: 392 layers, 26,299,704 parameters, 26,299,704 gradients, 93.8 GFLOPs
x: [1.00, 1.50, 512] # summary: 392 layers, 58,993,368 parameters, 58,993,368 gradients, 209.5 GFLOPs
# YOLO26n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 2, C3k2, [256, False, 0.25]] # 2-P2/4
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 2, C3k2, [512, False, 0.25]] # 4-P3/8
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 2, C3k2, [512, True]] # 6-P4/16
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 2, C3k2, [1024, True]] # 8-P5/32
- [-1, 1, SPPF, [1024, 5, 3, True]] # 9-P5/32
- [-1, 2, C2PSA, [1024]] # 10-P5/32
# YOLO26n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 11-P4/16
- [[-1, 6], 1, LCA, [512]] # 12-P4/16
- [-1, 2, C3k2, [512, True]] # 13-P4/16
- [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 14-P3/8
- [[-1, 4], 1, LCA, [256]] # 15-P3/8
- [-1, 2, C3k2, [256, True]] # 16-P3/8
- [-1, 1, Conv, [256, 3, 2]] # 17-P4/16
- [[-1, 13], 1, LCA, [512]] # 18-P4/16
- [-1, 2, C3k2, [512, True]] # 19-P4/16
- [-1, 1, Conv, [512, 3, 2]] # 20-P5/32
- [[-1, 10], 1, LCA, [1024]] # 21-P5/32
- [-1, 1, C3k2, [1024, True, 0.5, True]] # 22-P5/32
- [[16, 19, 22], 1, Detect, [nc]] # 23-P3/8,P4/16,P5/32- 语义分割
# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
# Ultralytics YOLO26 object detection model with P3/8 - P5/32 outputs
# Model docs: https://docs.ultralytics.com/models/yolo26
# Task docs: https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
end2end: True # whether to use end-to-end mode
reg_max: 1 # DFL bins
scales: # model compound scaling constants, i.e. 'model=yolo26n.yaml' will call yolo26.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.50, 0.25, 1024] # summary: 260 layers, 2,572,280 parameters, 2,572,280 gradients, 6.1 GFLOPs
s: [0.50, 0.50, 1024] # summary: 260 layers, 10,009,784 parameters, 10,009,784 gradients, 22.8 GFLOPs
m: [0.50, 1.00, 512] # summary: 280 layers, 21,896,248 parameters, 21,896,248 gradients, 75.4 GFLOPs
l: [1.00, 1.00, 512] # summary: 392 layers, 26,299,704 parameters, 26,299,704 gradients, 93.8 GFLOPs
x: [1.00, 1.50, 512] # summary: 392 layers, 58,993,368 parameters, 58,993,368 gradients, 209.5 GFLOPs
# YOLO26n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 2, C3k2, [256, False, 0.25]] # 2-P2/4
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 2, C3k2, [512, False, 0.25]] # 4-P3/8
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 2, C3k2, [512, True]] # 6-P4/16
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 2, C3k2, [1024, True]] # 8-P5/32
- [-1, 1, SPPF, [1024, 5, 3, True]] # 9-P5/32
- [-1, 2, C2PSA, [1024]] # 10-P5/32
# YOLO26n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 11-P4/16
- [[-1, 6], 1, LCA, [512]] # 12-P4/16
- [-1, 2, C3k2, [512, True]] # 13-P4/16
- [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 14-P3/8
- [[-1, 4], 1, LCA, [256]] # 15-P3/8
- [-1, 2, C3k2, [256, True]] # 16-P3/8
- [-1, 1, Conv, [256, 3, 2]] # 17-P4/16
- [[-1, 13], 1, LCA, [512]] # 18-P4/16
- [-1, 2, C3k2, [512, True]] # 19-P4/16
- [-1, 1, Conv, [512, 3, 2]] # 20-P5/32
- [[-1, 10], 1, LCA, [1024]] # 21-P5/32
- [-1, 1, C3k2, [1024, True, 0.5, True]] # 22-P5/32
- [[16, 19, 22], 1, Segment, [nc, 32, 256]]- 旋转目标检测
# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
# Ultralytics YOLO26 object detection model with P3/8 - P5/32 outputs
# Model docs: https://docs.ultralytics.com/models/yolo26
# Task docs: https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
end2end: True # whether to use end-to-end mode
reg_max: 1 # DFL bins
scales: # model compound scaling constants, i.e. 'model=yolo26n.yaml' will call yolo26.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.50, 0.25, 1024] # summary: 260 layers, 2,572,280 parameters, 2,572,280 gradients, 6.1 GFLOPs
s: [0.50, 0.50, 1024] # summary: 260 layers, 10,009,784 parameters, 10,009,784 gradients, 22.8 GFLOPs
m: [0.50, 1.00, 512] # summary: 280 layers, 21,896,248 parameters, 21,896,248 gradients, 75.4 GFLOPs
l: [1.00, 1.00, 512] # summary: 392 layers, 26,299,704 parameters, 26,299,704 gradients, 93.8 GFLOPs
x: [1.00, 1.50, 512] # summary: 392 layers, 58,993,368 parameters, 58,993,368 gradients, 209.5 GFLOPs
# YOLO26n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 2, C3k2, [256, False, 0.25]] # 2-P2/4
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 2, C3k2, [512, False, 0.25]] # 4-P3/8
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 2, C3k2, [512, True]] # 6-P4/16
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 2, C3k2, [1024, True]] # 8-P5/32
- [-1, 1, SPPF, [1024, 5, 3, True]] # 9-P5/32
- [-1, 2, C2PSA, [1024]] # 10-P5/32
# YOLO26n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 11-P4/16
- [[-1, 6], 1, LCA, [512]] # 12-P4/16
- [-1, 2, C3k2, [512, True]] # 13-P4/16
- [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 14-P3/8
- [[-1, 4], 1, LCA, [256]] # 15-P3/8
- [-1, 2, C3k2, [256, True]] # 16-P3/8
- [-1, 1, Conv, [256, 3, 2]] # 17-P4/16
- [[-1, 13], 1, LCA, [512]] # 18-P4/16
- [-1, 2, C3k2, [512, True]] # 19-P4/16
- [-1, 1, Conv, [512, 3, 2]] # 20-P5/32
- [[-1, 10], 1, LCA, [1024]] # 21-P5/32
- [-1, 1, C3k2, [1024, True, 0.5, True]] # 22-P5/32
- [[16, 19, 22], 1, OBB, [nc, 1]]温馨提示：本文只是对yolo26基础上添加模块，如果要对yolo26 n/l/m/x进行添加则只需要指定对应的depth_multiple 和 width_multiple
end2end: True # whether to use end-to-end mode
reg_max: 1 # DFL bins
scales: # model compound scaling constants, i.e. 'model=yolo26n.yaml' will call yolo26.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.50, 0.25, 1024] # summary: 260 layers, 2,572,280 parameters, 2,572,280 gradients, 6.1 GFLOPs
s: [0.50, 0.50, 1024] # summary: 260 layers, 10,009,784 parameters, 10,009,784 gradients, 22.8 GFLOPs
m: [0.50, 1.00, 512] # summary: 280 layers, 21,896,248 parameters, 21,896,248 gradients, 75.4 GFLOPs
l: [1.00, 1.00, 512] # summary: 392 layers, 26,299,704 parameters, 26,299,704 gradients, 93.8 GFLOPs
x: [1.00, 1.50, 512] # summary: 392 layers, 58,993,368 parameters, 58,993,368 gradients, 209.5 GFLOPs2.4 在task.py中进行注册
关键步骤四：在parse_model函数中进行注册，添加LCA
先在task.py导入函数
然后在task.py文件下找到parse_model这个函数，如下图，添加LCA
elif m in frozenset({LCA}):
c1, c2 = [ch[fi] for fi in f], args[0]
c2 = make_divisible(min(c2, max_channels) * width, 8)
args = [c1, c2, *args[1:]]2.5 执行程序
关键步骤五: 在ultralytics文件中新建train.py，将model的参数路径设置为yolo26_LCA.yaml的路径即可 【注意是在外边的Ultralytics下新建train.py】
from ultralytics import YOLO
import warnings
warnings.filterwarnings('ignore')
from pathlib import Path
if __name__ == '__main__':
# 加载模型
model = YOLO("ultralytics/cfg/26/yolo26.yaml") # 你要选择的模型yaml文件地址
# Use the model
results = model.train(data=r"你的数据集的yaml文件地址",
epochs=100, batch=16, imgsz=640, workers=4, name=Path(model.cfg).stem) # 训练模型🚀运行程序，如果出现下面的内容则说明添加成功🚀
from n params module arguments
0 -1 1 464 ultralytics.nn.modules.conv.Conv [3, 16, 3, 2]
1 -1 1 4672 ultralytics.nn.modules.conv.Conv [16, 32, 3, 2]
2 -1 1 6640 ultralytics.nn.modules.block.C3k2 [32, 64, 1, False, 0.25]
3 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
4 -1 1 26080 ultralytics.nn.modules.block.C3k2 [64, 128, 1, False, 0.25]
5 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
6 -1 1 87040 ultralytics.nn.modules.block.C3k2 [128, 128, 1, True]
7 -1 1 295424 ultralytics.nn.modules.conv.Conv [128, 256, 3, 2]
8 -1 1 346112 ultralytics.nn.modules.block.C3k2 [256, 256, 1, True]
9 -1 1 164608 ultralytics.nn.modules.block.SPPF [256, 256, 5, 3, True]
10 -1 1 249728 ultralytics.nn.modules.block.C2PSA [256, 256, 1]
11 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
12 [-1, 6] 1 245080 ultralytics.nn.models.LCA.LCA [[256, 128], 128]
13 -1 1 87040 ultralytics.nn.modules.block.C3k2 [128, 128, 1, True]
14 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
15 [-1, 4] 1 73648 ultralytics.nn.models.LCA.LCA [[128, 128], 64]
16 -1 1 22016 ultralytics.nn.modules.block.C3k2 [64, 64, 1, True]
17 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
18 [-1, 13] 1 220504 ultralytics.nn.models.LCA.LCA [[64, 128], 128]
19 -1 1 87040 ultralytics.nn.modules.block.C3k2 [128, 128, 1, True]
20 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
21 [-1, 10] 1 849576 ultralytics.nn.models.LCA.LCA [[128, 256], 256]
22 -1 1 430336 ultralytics.nn.modules.block.C3k2 [256, 256, 1, True, 0.5, True]
23 [16, 19, 22] 1 309656 ultralytics.nn.modules.head.Detect [80, 1, True, [64, 128, 256]]
YOLO26_LCA summary: 317 layers, 3,875,072 parameters, 3,875,072 gradients, 8.9 GFLOPs3. 完整代码分享
主页侧边
4. GFLOPs
关于GFLOPs的计算方式可以查看：百面算法工程师 | 卷积基础知识——Convolution
未改进的YOLO26n GFLOPs
改进后的GFLOPs
5. 进阶
可以与其他的注意力机制或者损失函数等结合，进一步提升检测效果
6.总结
通过以上的改进方法，我们成功提升了模型的表现。这只是一个开始，未来还有更多优化和技术深挖的空间。在这里，我想隆重向大家推荐我的专栏——<专栏地址：YOLO26改进-论文涨点——点击跳转看所有内容，关注不迷路！>。这个专栏专注于前沿的深度学习技术，特别是目标检测领域的最新进展，不仅包含对YOLO26的深入解析和改进策略，还会定期更新来自各大顶会（如CVPR、NeurIPS等）的论文复现和实战分享。
为什么订阅我的专栏？ ——专栏地址：YOLO26改进-论文涨点——点击跳转看所有内容，关注不迷路！
-
前沿技术解读：专栏不仅限于YOLO系列的改进，还会涵盖各类主流与新兴网络的最新研究成果，帮助你紧跟技术潮流。
-
详尽的实践分享：所有内容实践性也极强。每次更新都会附带代码和具体的改进步骤，保证每位读者都能迅速上手。
-
问题互动与答疑：订阅我的专栏后，你将可以随时向我提问，获取及时的答疑。
-
实时更新，紧跟行业动态：不定期发布来自全球顶会的最新研究方向和复现实验报告，让你时刻走在技术前沿。
专栏适合人群：
-
对目标检测、YOLO系列网络有深厚兴趣的同学
-
希望在用YOLO算法写论文的同学
-
对YOLO算法感兴趣的同学等
AtomGit 是由开放原子开源基金会联合 CSDN 等生态伙伴共同推出的新一代开源与人工智能协作平台。平台坚持“开放、中立、公益”的理念，把代码托管、模型共享、数据集托管、智能体开发体验和算力服务整合在一起，为开发者提供从开发、训练到部署的一站式体验。
更多推荐
31
0- 0
已为社区贡献20条内容
所有评论(0)