RT-DETR改进策略【卷积层】| CVPR-2023 部分卷积 PConv 轻量化卷积，降低内存占用
一、本文介绍
本文记录的是
利用
部分卷积 Partial Conv
优化
RT-DETR
的目标检测方法研究
。
深度可分离卷积
可以减少FLOPs，但会导致更高的内存访问，引起延迟并减慢整体计算。
部分卷积利用逐点卷积处理通道冗余，减少模型计算量和内存访问量。
二、部分卷积原理介绍
Run, Don’t Walk: Chasing Higher FLOPS for Faster Neural Networks
2.1 出发点
- 解决FLOPS问题 ：在追求快速神经网络的过程中，许多工作致力于减少浮点运算（FLOPs），但研究者发现FLOPs的减少并不一定能带来延迟的同等降低，主要原因是每秒浮点运算次数（FLOPS）过低。通过对典型神经网络在Intel CPU上FLOPS的比较，发现很多现有神经网络FLOPS较低，其FLOPS普遍低于流行的ResNet50，导致“快速”神经网络实际上不够快，FLOPs的减少无法转化为延迟的准确降低。
- 深度可分离卷积的内存访问问题 ：深度可分离卷积（DWConv）是一种常用的减少FLOPs的方法，但它在实际应用中存在问题。当为了补偿精度下降而增加网络宽度（即DWConv的通道数c增加到c’）时，会导致更高的内存访问，从而引起不可忽视的延迟并减慢整体计算，特别是对于I/O受限的设备。
2.2 原理
-
利用特征图冗余
：观察到特征图在不同通道之间存在高度相似性（冗余），通过
部分卷积（PConv）来利用这种冗余。PConv不是对所有输入通道进行常规卷积，而是仅对一部分输入通道应用常规卷积进行空间特征提取，同时保持其余通道不变。 - 减少计算冗余和内存访问 ：通过这种方式，同时减少了计算冗余和内存访问。从计算量（FLOPs）来看，PConv的FLOPs仅为常规卷积的一部分（例如，当典型的部分比例 r = c p c = 1 4 r = \frac{c_{p}}{c}=\frac{1}{4} r = c c p = 4 1 时，PConv的FLOPs仅为常规Conv的 1 16 \frac{1}{16} 16 1 ）；从内存访问量来看，PConv的内存访问量也仅为常规卷积的一部分（同样在 r = 1 4 r=\frac{1}{4} r = 4 1 时，仅为常规Conv的 1 4 \frac{1}{4} 4 1 ）。
2.3 结构
- 基本结构 ：对于输入 I ∈ R c × h × w I\in\mathbb{R}^{c×h×w} I ∈ R c × h × w ，PConv选取连续的 c p c_{p} c p 个通道（例如可以是第一个或最后一个连续的通道作为代表）应用常规卷积，输出的特征图维度与输入特征图维度相同（即输出 O ∈ R c × h × w O\in\mathbb{R}^{c×h×w} O ∈ R c × h × w ）。
-
与PWConv结合
：为了充分利用所有通道的信息，在
PConv之后紧接着添加一个 逐点卷积（PWConv） 。它们在输入特征图上的有效感受野看起来像一个T形卷积，这种T形卷积更关注中心位置，与均匀处理一个区域的常规卷积不同。并且将T形卷积分解为PConv和PWConv可以进一步利用滤波器间的冗余，节省FLOPs。
2.4 优势
-
有效提取空间特征
：实验证明
PConv在提取空间特征方面是有效的。通过构建由PConv和PWConv组成的简单网络，并在从预训练ResNet50提取的特征图数据集上进行训练，结果表明PConv + PWConv能达到最低的测试损失，更好地近似常规卷积进行特征变换，说明仅从部分特征图中捕获空间特征是足够且高效的。 -
适用于构建快速神经网络
：
PConv为设计快速有效的神经网络提供了一种新的选择，具有很大潜力替代现有的DWConv等操作，并且基于PConv构建的FasterNet在各种设备上实现了快速运行，在分类、检测和分割任务上取得了良好的性能，验证了PConv的有效性。
论文： https://arxiv.org/pdf/2303.03667
源码： https://github.com/JierunChen/FasterNet
三、部分卷积的实现代码
Partial_conv3模块
及其改进的实现代码如下：
import numpy as np
import torch
import torch.nn as nn
class Partial_conv3(nn.Module):
def __init__(self, dim, n_div=2, forward='split_cat'):
super().__init__()
self.dim_conv3 = dim // n_div
self.dim_untouched = dim - self.dim_conv3
self.partial_conv3 = nn.Conv2d(self.dim_conv3, self.dim_conv3, 3, 1, 1, bias=False)
if forward == 'slicing':
self.forward = self.forward_slicing
elif forward == 'split_cat':
self.forward = self.forward_split_cat
else:
raise NotImplementedError
def forward_slicing(self, x):
# only for inference
x = x.clone() # !!! Keep the original input intact for the residual connection later
x[:, :self.dim_conv3, :, :] = self.partial_conv3(x[:, :self.dim_conv3, :, :])
return x
def forward_split_cat(self, x):
# for training/inference
x1, x2 = torch.split(x, [self.dim_conv3, self.dim_untouched], dim=1)
x1 = self.partial_conv3(x1)
x = torch.cat((x1, x2), 1)
return x
def drop_path(x, drop_prob: float = 0., training: bool = False):
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
'survival rate' as the argument.
"""
if drop_prob == 0. or not training:
return x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets
random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
random_tensor.floor_() # binarize
output = x.div(keep_prob) * random_tensor
return output
class DropPath(nn.Module):
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
"""
def __init__(self, drop_prob=None):
super(DropPath, self).__init__()
self.drop_prob = drop_prob
def forward(self, x):
return drop_path(x, self.drop_prob, self.training)
class Faster_Block(nn.Module):
def __init__(self,
inc,
dim,
n_div=4,
mlp_ratio=2,
drop_path=0.1,
layer_scale_init_value=0.0,
pconv_fw_type='split_cat'
):
super().__init__()
self.dim = dim
self.mlp_ratio = mlp_ratio
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
self.n_div = n_div
mlp_hidden_dim = int(dim * mlp_ratio)
mlp_layer = [
Conv(dim, mlp_hidden_dim, 1),
nn.Conv2d(mlp_hidden_dim, dim, 1, bias=False)
]
self.mlp = nn.Sequential(*mlp_layer)
self.spatial_mixing = Partial_conv3(
dim,
n_div,
pconv_fw_type
)
self.adjust_channel = None
if inc != dim:
self.adjust_channel = Conv(inc, dim, 1)
if layer_scale_init_value > 0:
self.layer_scale = nn.Parameter(layer_scale_init_value * torch.ones((dim)), requires_grad=True)
self.forward = self.forward_layer_scale
else:
self.forward = self.forward
def forward(self, x):
if self.adjust_channel is not None:
x = self.adjust_channel(x)
shortcut = x
x = self.spatial_mixing(x)
x = shortcut + self.drop_path(self.mlp(x))
return x
def forward_layer_scale(self, x):
shortcut = x
x = self.spatial_mixing(x)
x = shortcut + self.drop_path(
self.layer_scale.unsqueeze(-1).unsqueeze(-1) * self.mlp(x))
return x
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))
四、创新模块
4.1 改进点⭐
模块改进方法
：直接加入
Partial_conv3
（
第五节讲解添加步骤
）。
Partial_conv3
模块加入如下：
4.2 改进点⭐
模块改进方法
：加入
Faster_Block
（
第五节讲解添加步骤
）。
FasterNet Block内部结构
：每个
FasterNet Block
由一个
部分卷积（PConv）层
和
两个逐点卷积（PWConv）层
组成。
- PConv层 ：只在部分输入通道上应用常规卷积来提取空间特征，其余通道保持不变。
- PWConv层 ：在PConv层之后，用于进一步处理特征信息，使特征信息能够在所有通道间流动，提高特征利用效率。
代码如下：
class Faster_Block(nn.Module):
def __init__(self,
inc,
dim,
n_div=4,
mlp_ratio=2,
drop_path=0.1,
layer_scale_init_value=0.0,
pconv_fw_type='split_cat'
):
super().__init__()
self.dim = dim
self.mlp_ratio = mlp_ratio
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
self.n_div = n_div
mlp_hidden_dim = int(dim * mlp_ratio)
mlp_layer = [
Conv(dim, mlp_hidden_dim, 1),
nn.Conv2d(mlp_hidden_dim, dim, 1, bias=False)
]
self.mlp = nn.Sequential(*mlp_layer)
self.spatial_mixing = Partial_conv3(
dim,
n_div,
pconv_fw_type
)
self.adjust_channel = None
if inc != dim:
self.adjust_channel = Conv(inc, dim, 1)
if layer_scale_init_value > 0:
self.layer_scale = nn.Parameter(layer_scale_init_value * torch.ones((dim)), requires_grad=True)
self.forward = self.forward_layer_scale
else:
self.forward = self.forward
def forward(self, x):
if self.adjust_channel is not None:
x = self.adjust_channel(x)
shortcut = x
x = self.spatial_mixing(x)
x = shortcut + self.drop_path(self.mlp(x))
return x
def forward_layer_scale(self, x):
shortcut = x
x = self.spatial_mixing(x)
x = shortcut + self.drop_path(
self.layer_scale.unsqueeze(-1).unsqueeze(-1) * self.mlp(x))
return x
注意❗：在
第五小节
中需要声明的模块名称为：
Partial_conv3
和
Faster_Block
。
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
PConv.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 .PConv import *
5.3 修改三
在
ultralytics/nn/modules/tasks.py
文件中，需要在指定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
Faster_Block
和
Partial_conv3
模块
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-PConv.yaml
。
将
rtdetr-l.yaml
中的内容复制到
rtdetr-l-PConv.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
颈部网络
中的
RepC3模块
替换成
Partial_conv3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
# 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, HGBlock, [192, 1024, 5, True, False]] # cm, c2, k, light, shortcut
- [-1, 6, HGBlock, [192, 1024, 5, True, True]]
- [-1, 6, HGBlock, [192, 1024, 5, True, True]] # 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, Partial_conv3, [512]] # 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, Partial_conv3, [512]] # X3 (21), fpn_blocks.1
- [-1, 1, Conv, [256, 3, 2]] # 22, downsample_convs.0
- [[-1, 17], 1, Concat, [1]] # cat Y4
- [-1, 3, Partial_conv3, [512]] # F4 (24), pan_blocks.0
- [-1, 1, Conv, [256, 3, 2]] # 25, downsample_convs.1
- [[-1, 12], 1, Concat, [1]] # cat Y5
- [-1, 3, Partial_conv3, [512]] # 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-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-Faster.yaml
。
将
rtdetr-l.yaml
中的内容复制到
rtdetr-l-Faster.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模块
替换成
Faster_Block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
# 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, Faster_Block, [48]] # stage 1
- [-1, 1, DWConv, [128, 3, 2, 1, False]] # 2-P3/8
- [-1, 6, Faster_Block, [128]] # stage 2
- [-1, 1, DWConv, [512, 3, 2, 1, False]] # 4-P4/16
- [-1, 1, Faster_Block, [512]] # cm, c2, k, light, shortcut
- [-1, 1, Faster_Block, [512]]
- [-1, 1, Faster_Block, [512]] # stage 3
- [-1, 1, DWConv, [1024, 3, 2, 1, False]] # 8-P5/32
- [-1, 6, Faster_Block, [1024]] # 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)
七、成功运行结果
打印网络模型可以看到
Partial_conv3
和
C2fCIB_PConv
已经加入到模型中，并可以进行训练了。
rtdetr-l-PConv ：
rtdetr-l-PConv summary: 564 layers, 25,320,751 parameters, 25,320,751 gradients, 66.9 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 1695360 ultralytics.nn.modules.block.HGBlock [512, 192, 1024, 5, 6, True, False]
6 -1 6 2055808 ultralytics.nn.modules.block.HGBlock [1024, 192, 1024, 5, 6, True, True]
7 -1 6 2055808 ultralytics.nn.modules.block.HGBlock [1024, 192, 1024, 5, 6, True, True]
8 -1 1 11264 ultralytics.nn.modules.conv.DWConv [1024, 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 262656 ultralytics.nn.modules.conv.Conv [1024, 256, 1, 1, None, 1, 1, False]
15 [-2, -1] 1 0 ultralytics.nn.modules.conv.Concat [1]
16 -1 3 27 ultralytics.nn.AddModules.PConv.Partial_conv3[512, 512]
17 -1 1 131584 ultralytics.nn.modules.conv.Conv [512, 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 27 ultralytics.nn.AddModules.PConv.Partial_conv3[512, 512]
22 -1 1 1180160 ultralytics.nn.modules.conv.Conv [512, 256, 3, 2]
23 [-1, 17] 1 0 ultralytics.nn.modules.conv.Concat [1]
24 -1 3 27 ultralytics.nn.AddModules.PConv.Partial_conv3[512, 512]
25 -1 1 1180160 ultralytics.nn.modules.conv.Conv [512, 256, 3, 2]
26 [-1, 12] 1 0 ultralytics.nn.modules.conv.Concat [1]
27 -1 3 27 ultralytics.nn.AddModules.PConv.Partial_conv3[512, 512]
28 [21, 24, 27] 1 7500515 ultralytics.nn.modules.head.RTDETRDecoder [1, [512, 512, 512]]
rtdetr-l-PConv summary: 564 layers, 25,320,751 parameters, 25,320,751 gradients, 66.9 GFLOPs
rtdetr-l-Faster ：
**rtdetr-l-Faster summary: 597 layers, 51,648,483 parameters, 51,648,483 gradients, 107.0 GFLOPs **
from n params module arguments
0 -1 1 25248 ultralytics.nn.modules.block.HGStem [3, 32, 48]
1 -1 6 64224 ultralytics.nn.AddModules.PConv.Faster_Block [48, 48]
2 -1 1 3712 ultralytics.nn.modules.conv.DWConv [48, 128, 3, 2, 1, False]
3 -1 6 451584 ultralytics.nn.AddModules.PConv.Faster_Block [128, 128]
4 -1 1 5632 ultralytics.nn.modules.conv.DWConv [128, 512, 3, 2, 1, False]
5 -1 1 1198080 ultralytics.nn.AddModules.PConv.Faster_Block [512, 512]
6 -1 1 1198080 ultralytics.nn.AddModules.PConv.Faster_Block [512, 512]
7 -1 1 1198080 ultralytics.nn.AddModules.PConv.Faster_Block [512, 512]
8 -1 1 11264 ultralytics.nn.modules.conv.DWConv [512, 1024, 3, 2, 1, False]
9 -1 6 28729344 ultralytics.nn.AddModules.PConv.Faster_Block [1024, 1024]
10 -1 1 262656 ultralytics.nn.modules.conv.Conv [1024, 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 33280 ultralytics.nn.modules.conv.Conv [128, 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-Faster summary: 597 layers, 51,648,483 parameters, 51,648,483 gradients, 107.0 GFLOPs