【YOLOv12单模态融合改进】普通数据集的双模型融合改进，涉及中期、中后期、后期融合方式的完整配置步骤以及二次改进方案
前言
主题： YOLOv12的单模态融合改进，普通数据集的双模型融合改进（双模型同步提升）
方式： 中期融合、中-后期融合、后期融合。
内容： 包含融合方式详解和完整配置步骤以及二次改进建议，通过融合多个模型的优势实现精度提升。
一、融合方式
输入的是一个模态的数据，所以没有早期的融合。
1.1 中期融合方法及结构图
定义： 在网络中间层（骨干网络与颈部网络之间）对多模态特征进行融合。
实现方式： 每个模态通过独立的骨干网络提取特征，融合时采用Concat操作合并特征图，送入颈部网络。
结构示意图：
1.2 中-后期融合方法及结构图
定义： 在颈部网络的上采样之后对多模态特征进行融合。
实现方式： 每个模态通过独立的骨干网络和颈部的FPN网络提取特征，融合时采用Concat操作合并特征图，送入检测头。
结构示意图：
1.3 后期融合方法及结构图
定义： 在网络输出阶段（如检测头或分类器前）对多模态特征进行融合。
实现方式： 每个模态通过独立的骨干网络和颈部网络提取特征，融合时采用Concat操作合并特征图，送入检测头。
结构示意图：
二、完整配置步骤
相关的配置只涉及单模态，只需在原本的项目包中配置运行即可，不需要使用我提供的多模态项目包。
①：在
ultralytics/nn/modules/block.py
中添加如下代码，并在
__all__
中添加
“IN”
class IN(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return x
②：在
ultralytics/nn/modules/__init__.py
中的
from .block import (...）
添加
IN
③：在
ultralytics/nn/tasks.py
中的
from ultralytics.nn.modules import (...)
中添加
IN
至此，添加完成。
三、YAML模型结构
此处以
ultralytics/cfg/models/v12/yolov12.yaml
为例，在同目录下创建一个用于自己数据集训练的双模型融合文件，并粘贴下方的模型训练即可。
3.1 中期融合
# YOLOv12 🚀, AGPL-3.0 license
# YOLOv12 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov12n.yaml' will call yolov12.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.50, 0.25, 1024] # summary: 465 layers, 2,603,056 parameters, 2,603,040 gradients, 6.7 GFLOPs
s: [0.50, 0.50, 1024] # summary: 465 layers, 9,285,632 parameters, 9,285,616 gradients, 21.7 GFLOPs
m: [0.50, 1.00, 512] # summary: 501 layers, 20,201,216 parameters, 20,201,200 gradients, 68.1 GFLOPs
l: [1.00, 1.00, 512] # summary: 831 layers, 26,454,880 parameters, 26,454,864 gradients, 89.7 GFLOPs
x: [1.00, 1.50, 512] # summary: 831 layers, 59,216,928 parameters, 59,216,912 gradients, 200.3 GFLOPs
# YOLO12n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, IN, []]
- [0, 1, Conv, [64, 3, 2]] # 1-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 2-P2/4
- [-1, 2, C3k2, [256, False, 0.25]]
- [-1, 1, Conv, [256, 3, 2]] # 4-P3/8
- [-1, 2, C3k2, [512, False, 0.25]]
- [-1, 1, Conv, [512, 3, 2]] # 6-P4/16
- [-1, 4, A2C2f, [512, True, 4]]
- [-1, 1, Conv, [1024, 3, 2]] # 8-P5/32
- [-1, 4, A2C2f, [1024, True, 1]] # 9
- [0, 1, Conv, [64, 3, 2]] # 10-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 11-P2/4
- [-1, 2, C3k2, [256, False, 0.25]]
- [-1, 1, Conv, [256, 3, 2]] # 13-P3/8
- [-1, 2, C3k2, [512, False, 0.25]]
- [-1, 1, Conv, [512, 3, 2]] # 15-P4/16
- [-1, 4, A2C2f, [512, True, 4]]
- [-1, 1, Conv, [1024, 3, 2]] # 17-P5/32
- [-1, 4, A2C2f, [1024, True, 1]] # 18
- [[5, 14], 1, Concat, [1]] # 19 cat backbone P3
- [[7, 16], 1, Concat, [1]] # 20 cat backbone P4
- [[9, 18], 1, Concat, [1]] # 21 cat backbone P5
# YOLO12n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 20], 1, Concat, [1]] # cat backbone P4
- [-1, 2, A2C2f, [512, False, -1]] # 24
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 19], 1, Concat, [1]] # cat backbone P3
- [-1, 2, A2C2f, [256, False, -1]] # 27
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 24], 1, Concat, [1]] # cat head P4
- [-1, 2, A2C2f, [512, False, -1]] # 30
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 21], 1, Concat, [1]] # cat head P5
- [-1, 2, C3k2, [1024, True]] # 33 (P5/32-large)
- [[27, 30, 33], 1, Detect, [nc]] # Detect(P3, P4, P5)
3.2 中-后期融合
# YOLOv12 🚀, AGPL-3.0 license
# YOLOv12 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov12n.yaml' will call yolov12.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.50, 0.25, 1024] # summary: 465 layers, 2,603,056 parameters, 2,603,040 gradients, 6.7 GFLOPs
s: [0.50, 0.50, 1024] # summary: 465 layers, 9,285,632 parameters, 9,285,616 gradients, 21.7 GFLOPs
m: [0.50, 1.00, 512] # summary: 501 layers, 20,201,216 parameters, 20,201,200 gradients, 68.1 GFLOPs
l: [1.00, 1.00, 512] # summary: 831 layers, 26,454,880 parameters, 26,454,864 gradients, 89.7 GFLOPs
x: [1.00, 1.50, 512] # summary: 831 layers, 59,216,928 parameters, 59,216,912 gradients, 200.3 GFLOPs
# YOLO12n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, IN, []]
- [0, 1, Conv, [64, 3, 2]] # 1-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 2-P2/4
- [-1, 2, C3k2, [256, False, 0.25]]
- [-1, 1, Conv, [256, 3, 2]] # 4-P3/8
- [-1, 2, C3k2, [512, False, 0.25]]
- [-1, 1, Conv, [512, 3, 2]] # 6-P4/16
- [-1, 4, A2C2f, [512, True, 4]]
- [-1, 1, Conv, [1024, 3, 2]] # 8-P5/32
- [-1, 4, A2C2f, [1024, True, 1]] # 9
- [0, 1, Conv, [64, 3, 2]] # 10-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 11-P2/4
- [-1, 2, C3k2, [256, False, 0.25]]
- [-1, 1, Conv, [256, 3, 2]] # 13-P3/8
- [-1, 2, C3k2, [512, False, 0.25]]
- [-1, 1, Conv, [512, 3, 2]] # 15-P4/16
- [-1, 4, A2C2f, [512, True, 4]]
- [-1, 1, Conv, [1024, 3, 2]] # 17-P5/32
- [-1, 4, A2C2f, [1024, True, 1]] # 18
# YOLO12n head
head:
- [9, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 7], 1, Concat, [1]] # cat backbone P4
- [-1, 2, A2C2f, [512, False, -1]] # 21
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 5], 1, Concat, [1]] # cat backbone P3
- [-1, 2, A2C2f, [256, False, -1]] # 24
- [18, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 16], 1, Concat, [1]] # cat backbone P4
- [-1, 2, A2C2f, [512, False, -1]] # 27
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 14], 1, Concat, [1]] # cat backbone P3
- [-1, 2, A2C2f, [256, False, -1]] # 30
- [[9, 18], 1, Concat, [1]] # 31 cat backbone P3
- [[21, 27], 1, Concat, [1]] # 32 cat backbone P4
- [[24, 30], 1, Concat, [1]] # 33 cat backbone P5
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 32], 1, Concat, [1]] # cat head P4
- [-1, 2, A2C2f, [512, False, -1]] # 36
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 31], 1, Concat, [1]] # cat head P5
- [-1, 2, C3k2, [1024, True]] # 39 (P5/32-large)
- [[33, 36, 39], 1, Detect, [nc]] # Detect(P3, P4, P5)
3.3 后期融合
# YOLOv12 🚀, AGPL-3.0 license
# YOLOv12 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov12n.yaml' will call yolov12.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.50, 0.25, 1024] # summary: 465 layers, 2,603,056 parameters, 2,603,040 gradients, 6.7 GFLOPs
s: [0.50, 0.50, 1024] # summary: 465 layers, 9,285,632 parameters, 9,285,616 gradients, 21.7 GFLOPs
m: [0.50, 1.00, 512] # summary: 501 layers, 20,201,216 parameters, 20,201,200 gradients, 68.1 GFLOPs
l: [1.00, 1.00, 512] # summary: 831 layers, 26,454,880 parameters, 26,454,864 gradients, 89.7 GFLOPs
x: [1.00, 1.50, 512] # summary: 831 layers, 59,216,928 parameters, 59,216,912 gradients, 200.3 GFLOPs
# YOLO12n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, IN, []]
- [0, 1, Conv, [64, 3, 2]] # 1-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 2-P2/4
- [-1, 2, C3k2, [256, False, 0.25]]
- [-1, 1, Conv, [256, 3, 2]] # 4-P3/8
- [-1, 2, C3k2, [512, False, 0.25]]
- [-1, 1, Conv, [512, 3, 2]] # 6-P4/16
- [-1, 4, A2C2f, [512, True, 4]]
- [-1, 1, Conv, [1024, 3, 2]] # 8-P5/32
- [-1, 4, A2C2f, [1024, True, 1]] # 9
- [0, 1, Conv, [64, 3, 2]] # 10-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 11-P2/4
- [-1, 2, C3k2, [256, False, 0.25]]
- [-1, 1, Conv, [256, 3, 2]] # 13-P3/8
- [-1, 2, C3k2, [512, False, 0.25]]
- [-1, 1, Conv, [512, 3, 2]] # 15-P4/16
- [-1, 4, A2C2f, [512, True, 4]]
- [-1, 1, Conv, [1024, 3, 2]] # 17-P5/32
- [-1, 4, A2C2f, [1024, True, 1]] # 18
# YOLO12n head
head:
- [9, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 7], 1, Concat, [1]] # cat backbone P4
- [-1, 2, A2C2f, [512, False, -1]] # 21
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 5], 1, Concat, [1]] # cat backbone P3
- [-1, 2, A2C2f, [256, False, -1]] # 24
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 21], 1, Concat, [1]] # cat head P4
- [-1, 2, A2C2f, [512, False, -1]] # 27
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 2, C3k2, [1024, True]] # 30 (P5/32-large)
- [18, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 16], 1, Concat, [1]] # cat backbone P4
- [-1, 2, A2C2f, [512, False, -1]] # 33
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 14], 1, Concat, [1]] # cat backbone P3
- [-1, 2, A2C2f, [256, False, -1]] # 36
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 33], 1, Concat, [1]] # cat head P4
- [-1, 2, A2C2f, [512, False, -1]] # 39
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 18], 1, Concat, [1]] # cat head P5
- [-1, 2, C3k2, [1024, True]] # 42 (P5/32-large)
- [[24, 36], 1, Concat, [1]] # 43 cat backbone P3
- [[27, 39], 1, Concat, [1]] # 44 cat backbone P4
- [[30, 42], 1, Concat, [1]] # 45 cat backbone P5
- [[43, 44, 45], 1, Detect, [nc]] # Detect(P3, P4, P5)
四、成功运行结果
中期融合结果：
YOLOv12-BiBackbone summary: 710 layers, 4,052,243 parameters, 4,052,227 gradients, 10.0 GFLOPs
from n params module arguments
0 -1 1 0 ultralytics.nn.modules.block.IN []
1 0 1 464 ultralytics.nn.modules.conv.Conv [3, 16, 3, 2]
2 -1 1 4672 ultralytics.nn.modules.conv.Conv [16, 32, 3, 2]
3 -1 1 6640 ultralytics.nn.modules.block.C3k2 [32, 64, 1, False, 0.25]
4 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
5 -1 1 26080 ultralytics.nn.modules.block.C3k2 [64, 128, 1, False, 0.25]
6 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
7 -1 2 181120 ultralytics.nn.modules.block.A2C2f [128, 128, 2, True, 4]
8 -1 1 295424 ultralytics.nn.modules.conv.Conv [128, 256, 3, 2]
9 -1 2 689920 ultralytics.nn.modules.block.A2C2f [256, 256, 2, True, 1]
10 0 1 464 ultralytics.nn.modules.conv.Conv [3, 16, 3, 2]
11 -1 1 4672 ultralytics.nn.modules.conv.Conv [16, 32, 3, 2]
12 -1 1 6640 ultralytics.nn.modules.block.C3k2 [32, 64, 1, False, 0.25]
13 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
14 -1 1 26080 ultralytics.nn.modules.block.C3k2 [64, 128, 1, False, 0.25]
15 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
16 -1 2 181120 ultralytics.nn.modules.block.A2C2f [128, 128, 2, True, 4]
17 -1 1 295424 ultralytics.nn.modules.conv.Conv [128, 256, 3, 2]
18 -1 2 689920 ultralytics.nn.modules.block.A2C2f [256, 256, 2, True, 1]
19 [5, 14] 1 0 ultralytics.nn.modules.conv.Concat [1]
20 [7, 16] 1 0 ultralytics.nn.modules.conv.Concat [1]
21 [9, 18] 1 0 ultralytics.nn.modules.conv.Concat [1]
22 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
23 [-1, 20] 1 0 ultralytics.nn.modules.conv.Concat [1]
24 -1 1 111488 ultralytics.nn.modules.block.A2C2f [768, 128, 1, False, -1]
25 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
26 [-1, 19] 1 0 ultralytics.nn.modules.conv.Concat [1]
27 -1 1 28096 ultralytics.nn.modules.block.A2C2f [384, 64, 1, False, -1]
28 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
29 [-1, 24] 1 0 ultralytics.nn.modules.conv.Concat [1]
30 -1 1 74624 ultralytics.nn.modules.block.A2C2f [192, 128, 1, False, -1]
31 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
32 [-1, 21] 1 0 ultralytics.nn.modules.conv.Concat [1]
33 -1 1 444416 ultralytics.nn.modules.block.C3k2 [640, 256, 1, True]
34 [27, 30, 33] 1 430867 ultralytics.nn.modules.head.Detect [1, [64, 128, 256]]
YOLOv12-BiBackbone summary: 710 layers, 4,052,243 parameters, 4,052,227 gradients, 10.0 GFLOPs
中-后期融合结果：
YOLOv12-BiBackbone summary: 780 layers, 4,293,907 parameters, 4,293,891 gradients, 11.6 GFLOPs
from n params module arguments
0 -1 1 0 ultralytics.nn.modules.block.IN []
1 0 1 464 ultralytics.nn.modules.conv.Conv [3, 16, 3, 2]
2 -1 1 4672 ultralytics.nn.modules.conv.Conv [16, 32, 3, 2]
3 -1 1 6640 ultralytics.nn.modules.block.C3k2 [32, 64, 1, False, 0.25]
4 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
5 -1 1 26080 ultralytics.nn.modules.block.C3k2 [64, 128, 1, False, 0.25]
6 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
7 -1 2 181120 ultralytics.nn.modules.block.A2C2f [128, 128, 2, True, 4]
8 -1 1 295424 ultralytics.nn.modules.conv.Conv [128, 256, 3, 2]
9 -1 2 689920 ultralytics.nn.modules.block.A2C2f [256, 256, 2, True, 1]
10 0 1 464 ultralytics.nn.modules.conv.Conv [3, 16, 3, 2]
11 -1 1 4672 ultralytics.nn.modules.conv.Conv [16, 32, 3, 2]
12 -1 1 6640 ultralytics.nn.modules.block.C3k2 [32, 64, 1, False, 0.25]
13 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
14 -1 1 26080 ultralytics.nn.modules.block.C3k2 [64, 128, 1, False, 0.25]
15 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
16 -1 2 181120 ultralytics.nn.modules.block.A2C2f [128, 128, 2, True, 4]
17 -1 1 295424 ultralytics.nn.modules.conv.Conv [128, 256, 3, 2]
18 -1 2 689920 ultralytics.nn.modules.block.A2C2f [256, 256, 2, True, 1]
19 9 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
20 [-1, 7] 1 0 ultralytics.nn.modules.conv.Concat [1]
21 -1 1 86912 ultralytics.nn.modules.block.A2C2f [384, 128, 1, False, -1]
22 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
23 [-1, 5] 1 0 ultralytics.nn.modules.conv.Concat [1]
24 -1 1 24000 ultralytics.nn.modules.block.A2C2f [256, 64, 1, False, -1]
25 18 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
26 [-1, 16] 1 0 ultralytics.nn.modules.conv.Concat [1]
27 -1 1 86912 ultralytics.nn.modules.block.A2C2f [384, 128, 1, False, -1]
28 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
29 [-1, 14] 1 0 ultralytics.nn.modules.conv.Concat [1]
30 -1 1 24000 ultralytics.nn.modules.block.A2C2f [256, 64, 1, False, -1]
31 [9, 18] 1 0 ultralytics.nn.modules.conv.Concat [1]
32 [21, 27] 1 0 ultralytics.nn.modules.conv.Concat [1]
33 [24, 30] 1 0 ultralytics.nn.modules.conv.Concat [1]
34 -1 1 73856 ultralytics.nn.modules.conv.Conv [128, 64, 3, 2]
35 [-1, 32] 1 0 ultralytics.nn.modules.conv.Concat [1]
36 -1 1 82816 ultralytics.nn.modules.block.A2C2f [320, 128, 1, False, -1]
37 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
38 [-1, 31] 1 0 ultralytics.nn.modules.conv.Concat [1]
39 -1 1 444416 ultralytics.nn.modules.block.C3k2 [640, 256, 1, True]
40 [33, 36, 39] 1 545235 ultralytics.nn.modules.head.Detect [1, [128, 128, 256]]
YOLOv12-BiBackbone summary: 780 layers, 4,293,907 parameters, 4,293,891 gradients, 11.6 GFLOPs
后期融合结果：
YOLOv12-BiBackbone summary: 854 layers, 5,096,083 parameters, 5,096,067 gradients, 12.6 GFLOPs
from n params module arguments
0 -1 1 0 ultralytics.nn.modules.block.IN []
1 0 1 464 ultralytics.nn.modules.conv.Conv [3, 16, 3, 2]
2 -1 1 4672 ultralytics.nn.modules.conv.Conv [16, 32, 3, 2]
3 -1 1 6640 ultralytics.nn.modules.block.C3k2 [32, 64, 1, False, 0.25]
4 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
5 -1 1 26080 ultralytics.nn.modules.block.C3k2 [64, 128, 1, False, 0.25]
6 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
7 -1 2 181120 ultralytics.nn.modules.block.A2C2f [128, 128, 2, True, 4]
8 -1 1 295424 ultralytics.nn.modules.conv.Conv [128, 256, 3, 2]
9 -1 2 689920 ultralytics.nn.modules.block.A2C2f [256, 256, 2, True, 1]
10 0 1 464 ultralytics.nn.modules.conv.Conv [3, 16, 3, 2]
11 -1 1 4672 ultralytics.nn.modules.conv.Conv [16, 32, 3, 2]
12 -1 1 6640 ultralytics.nn.modules.block.C3k2 [32, 64, 1, False, 0.25]
13 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
14 -1 1 26080 ultralytics.nn.modules.block.C3k2 [64, 128, 1, False, 0.25]
15 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
16 -1 2 181120 ultralytics.nn.modules.block.A2C2f [128, 128, 2, True, 4]
17 -1 1 295424 ultralytics.nn.modules.conv.Conv [128, 256, 3, 2]
18 -1 2 689920 ultralytics.nn.modules.block.A2C2f [256, 256, 2, True, 1]
19 9 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
20 [-1, 7] 1 0 ultralytics.nn.modules.conv.Concat [1]
21 -1 1 86912 ultralytics.nn.modules.block.A2C2f [384, 128, 1, False, -1]
22 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
23 [-1, 5] 1 0 ultralytics.nn.modules.conv.Concat [1]
24 -1 1 24000 ultralytics.nn.modules.block.A2C2f [256, 64, 1, False, -1]
25 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
26 [-1, 21] 1 0 ultralytics.nn.modules.conv.Concat [1]
27 -1 1 74624 ultralytics.nn.modules.block.A2C2f [192, 128, 1, False, -1]
28 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
29 [-1, 9] 1 0 ultralytics.nn.modules.conv.Concat [1]
30 -1 1 378880 ultralytics.nn.modules.block.C3k2 [384, 256, 1, True]
31 18 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
32 [-1, 16] 1 0 ultralytics.nn.modules.conv.Concat [1]
33 -1 1 86912 ultralytics.nn.modules.block.A2C2f [384, 128, 1, False, -1]
34 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
35 [-1, 14] 1 0 ultralytics.nn.modules.conv.Concat [1]
36 -1 1 24000 ultralytics.nn.modules.block.A2C2f [256, 64, 1, False, -1]
37 -1 1 36992 ultralytics.nn.modules.conv.Conv [64, 64, 3, 2]
38 [-1, 33] 1 0 ultralytics.nn.modules.conv.Concat [1]
39 -1 1 74624 ultralytics.nn.modules.block.A2C2f [192, 128, 1, False, -1]
40 -1 1 147712 ultralytics.nn.modules.conv.Conv [128, 128, 3, 2]
41 [-1, 18] 1 0 ultralytics.nn.modules.conv.Concat [1]
42 -1 1 378880 ultralytics.nn.modules.block.C3k2 [384, 256, 1, True]
43 [24, 36] 1 0 ultralytics.nn.modules.conv.Concat [1]
44 [27, 39] 1 0 ultralytics.nn.modules.conv.Concat [1]
45 [30, 42] 1 0 ultralytics.nn.modules.conv.Concat [1]
46 [43, 44, 45] 1 819795 ultralytics.nn.modules.head.Detect [1, [128, 256, 512]]
YOLOv12-BiBackbone summary: 854 layers, 5,096,083 parameters, 5,096,067 gradients, 12.6 GFLOPs
五、二次改进方案
-
双模型的二次改进和普通模型的改进一致，主要涉及到A2C2f、颈部结构、上采样、下采样等，可以增加或替换成其它模块，可以换成其它的颈部结构在进行融合。
-
两个骨干中均可以再次添加其它模块，需要注意的是融合的时候特征图大小要一致。