RTDETR融合MogaNet中的ChannelAggregationFFN模块

RT-DETR融合MogaNet的ChannelAggregationFFN模块

最新推荐文章于 2025-12-11 15:22:21 发布

原创最新推荐文章于 2025-12-11 15:22:21 发布 · 1.4k 阅读

11 ·

CC 4.0 BY-SA版权

文章标签：

#深度学习 #机器学习 #人工智能

RT-DETR涨点改进专栏专栏收录该内容

91 篇文章

订阅专栏

RT-DETR使用教程： RT-DETR使用教程

RT-DETR改进汇总贴：RT-DETR更新汇总贴

《MogaNet: Multi-order Gated Aggregation Network》

一、模块介绍

论文链接：https://arxiv.org/abs/2211.03295

代码链接：https://github.com/AIFengheshu/Plug-play-modules

论文速览：

通过将内核尽可能全局化，现代 ConvNet 在计算机视觉任务中显示出巨大的潜力。然而，深度神经网络（DNN）中多阶博弈论交互的最新进展揭示了现代ConvNet的表示瓶颈，即随着核大小的增加，表达互没有得到有效编码。为了应对这一挑战，我们提出了一个新的现代 ConvNet 系列，称为 MogaNet，用于在纯基于 ConvNet 的模型中进行判别视觉表示学习，并具有良好的复杂性-性能权衡。MogaNet将概念上简单而有效的卷积和门控聚合封装到一个紧凑的模块中，其中有效地收集了判别特征并自适应地进行情境化。与 ImageNet 上最先进的 ViT 和 ConvNet 以及各种下游视觉基准测试（包括 COCO 对象检测、ADE20K 语义分割、2D 和 3D 人体姿态估计和视频预测）相比，MogaNet 表现出强大的可扩展性、令人印象深刻的参数效率和具有竞争力的性能。

总结：本文更新其中ChannelAggregationFFN模块的

⭐⭐本文二创模块仅更新于付费群中，往期免费教程可看下方链接⭐⭐

RT-DETR更新汇总贴（含免费教程）文章浏览阅读264次。RT-DETR使用教程：缝合教程： RT-DETR中的yaml文件详解：labelimg使用教程：_rt-deterhttps://xy2668825911.blog.csdn.net/article/details/143696113https://xy2668825911.blog.csdn.net/article/details/143696113

二、二创融合模块

2.1 相关代码

# MogaNet: Multi-order Gated Aggregation Network
# https://arxiv.org/pdf/2211.03295
# https://blog.csdn.net/StopAndGoyyy?spm=1011.2124.3001.5343
# https://github.com/AIFengheshu/Plug-play-modules
def build_act_layer(act_type):
    #Build activation layer
    if act_type is None:
        return nn.Identity()
    assert act_type in ['GELU', 'ReLU', 'SiLU']
    if act_type == 'SiLU':
        return nn.SiLU()
    elif act_type == 'ReLU':
        return nn.ReLU()
    else:
        return nn.GELU()

class ElementScale(nn.Module):
    #A learnable element-wise scaler.

    def __init__(self, embed_dims, init_value=0., requires_grad=True):
        super(ElementScale, self).__init__()
        self.scale = nn.Parameter(
            init_value * torch.ones((1, embed_dims, 1, 1)),
            requires_grad=requires_grad
        )

    def forward(self, x):
        return x * self.scale

class ChannelAggregationFFN(nn.Module):
    """An implementation of FFN with Channel Aggregation.

    Args:
        embed_dims (int): The feature dimension. Same as
            `MultiheadAttention`.
        feedforward_channels (int): The hidden dimension of FFNs.
        kernel_size (int): The depth-wise conv kernel size as the
            depth-wise convolution. Defaults to 3.
        act_type (str): The type of activation. Defaults to 'GELU'.
        ffn_drop (float, optional): Probability of an element to be
            zeroed in FFN. Default 0.0.
    """

    def __init__(self,
                 embed_dims,
                 kernel_size=3,
                 act_type='GELU',
                 ffn_drop=0.):
        super(ChannelAggregationFFN, self).__init__()

        self.embed_dims = embed_dims
        self.feedforward_channels = int(embed_dims * 4)

        self.fc1 = nn.Conv2d(
            in_channels=embed_dims,
            out_channels=self.feedforward_channels,
            kernel_size=1)
        self.dwconv = nn.Conv2d(
            in_channels=self.feedforward_channels,
            out_channels=self.feedforward_channels,
            kernel_size=kernel_size,
            stride=1,
            padding=kernel_size // 2,
            bias=True,
            groups=self.feedforward_channels)
        self.act = build_act_layer(act_type)
        self.fc2 = nn.Conv2d(
            in_channels=self.feedforward_channels,
            out_channels=embed_dims,
            kernel_size=1)
        self.drop = nn.Dropout(ffn_drop)

        self.decompose = nn.Conv2d(
            in_channels=self.feedforward_channels,  # C -> 1
            out_channels=1, kernel_size=1,
        )
        self.sigma = ElementScale(
            self.feedforward_channels, init_value=1e-5, requires_grad=True)
        self.decompose_act = build_act_layer(act_type)

    def feat_decompose(self, x):
        # x_d: [B, C, H, W] -> [B, 1, H, W]
        x = x + self.sigma(x - self.decompose_act(self.decompose(x)))
        return x

    def forward(self, x):
        # proj 1
        x = self.fc1(x)
        x = self.dwconv(x)
        x = self.act(x)
        x = self.drop(x)
        # proj 2
        x = self.feat_decompose(x)
        x = self.fc2(x)
        x = self.drop(x)
        return x

2.2 更改yaml文件（以自研模型加入为例）

yam文件解读：YOLO系列 “.yaml“文件解读_yolo yaml文件-CSDN博客

打开更改ultralytics/cfg/models/rt-detr路径下的rtdetr-l.yaml文件，替换原有模块。

# 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
# ⭐⭐Powered by https://blog.csdn.net/StopAndGoyyy,  技术指导QQ:2668825911⭐⭐

# Parameters
nc: 80 # 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, 512]
#  n: [ 0.33, 0.25, 1024 ]
#  s: [ 0.33, 0.50, 1024 ]
#  m: [ 0.67, 0.75, 768 ]
#  l: [ 1.00, 1.00, 512 ]
#  x: [ 1.00, 1.25, 512 ]
# ⭐⭐Powered by https://blog.csdn.net/StopAndGoyyy,  技术指导QQ:2668825911⭐⭐

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, CCRI, [128, 5, True, False]]
  - [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
  - [-1, 4, CCRI, [256, 3, True, True]]
  - [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
  - [-1, 4, CCRI, [512, 3, True, True]]
  - [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
  - [-1, 2, CCRI, [1024, 3, True, False]]

head:
  - [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 9 input_proj.2
  - [-1, 1, ChannelAggregationFFN, []]
  - [-1, 1, Conv, [256, 1, 1]] # 11, Y5, lateral_convs.0

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
  - [6, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 13 input_proj.1
  - [[-2, -1], 1, Concat, [1]]
  - [-1, 2, RepC4, [256]] # 15, fpn_blocks.0
  - [-1, 1, Conv, [256, 1, 1]] # 16, Y4, lateral_convs.1

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
  - [4, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 18 input_proj.0
  - [[-2, -1], 1, Concat, [1]] # cat backbone P4
  - [-1, 2, RepC4, [256]] # X3 (20), fpn_blocks.1

  - [-1, 1, Conv, [256, 3, 2]] # 22, downsample_convs.0
  - [[-1, 16], 1, Concat, [1]] # cat Y4
  - [-1, 2, RepC4, [256]] # F4 (23), pan_blocks.0

  - [-1, 1, Conv, [256, 3, 2]] # 24, downsample_convs.1
  - [[-1, 11], 1, Concat, [1]] # cat Y5
  - [-1, 2, RepC4, [256]] # F5 (26), pan_blocks.1

  - [[20, 23, 26], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)
# ⭐⭐Powered by https://blog.csdn.net/StopAndGoyyy,  技术指导QQ:2668825911⭐⭐

2.2 修改train.py文件

创建Train_RT脚本用于训练。

from ultralytics.models import RTDETR
import os
os.environ['KMP_DUPLICATE_LIB_OK'] = 'True'

if __name__ == '__main__':
    model = RTDETR(model='ultralytics/cfg/models/rt-detr/rtdetr-l.yaml')
    # model.load('yolov8n.pt')
    model.train(data='./data.yaml', epochs=2, batch=1, device='0', imgsz=640, workers=2, cache=False,
                amp=True, mosaic=False, project='runs/train', name='exp')