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
《Rethinking Mobile Block for Efficient Attention-based Models》
一、 模块介绍
论文链接：https://arxiv.org/pdf/2301.01146
代码链接：https://github.com/zhangzjn/EMO
论文速览：
倒置残差块 （IRB） 是轻量级 CNN 的基础设施，但尚未得到基于注意力的研究的认可。这项工作从统一的角度从高效的 IRB 和 Transformer 的有效组件重新思考轻量级基础设施，将基于 CNN 的 IRB 扩展到基于注意力的模型，并抽象出一个单残差 Meta Mobile Block （MMB） 用于轻量级模型设计。遵循简单但有效的设计标准，我们推导出了一个现代的倒置残差移动块 （iRMB），并构建了一个仅用 iRMB 的 ResNetlike 高效模块 （EMO） 用于下游任务。在 ImageNet-1K、COCO2017 和 ADE20K 基准测试上的广泛实验表明，我们的 EMO 优于最先进的方法，例如，EMO-1M/2M/5M 达到 71.5、75.1 和 78.4 Top-1，超过了基于等阶 CNN/注意力的模型，同时很好地权衡了参数、效率和准确性。
总结：一种结合Resnet与Transformer的模块。
二、 加入到RT-DETR中
2.1 创建脚本文件
首先在ultralytics->nn路径下创建blocks.py脚本，用于存放模块代码。
2.2 复制代码
复制代码粘到刚刚创建的blocks.py脚本中，如下图所示：
class SE(nn.Module):
def __init__(self, channel=512, reduction=16):
super().__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.fc = nn.Sequential(
nn.Linear(channel, channel // reduction, bias=False),
nn.ReLU(inplace=True),
nn.Linear(channel // reduction, channel, bias=False),
nn.Sigmoid()
)
def forward(self, x):
b, c, _, _ = x.size()
y = self.avg_pool(x).view(b, c)
y = self.fc(y).view(b, c, 1, 1)
return x * y.expand_as(x)
class iRMB(nn.Module):
def __init__(self, dim_in, dim_out, dim_head=32, norm_in=True, has_skip=True, exp_ratio=1.0,
act_layer='relu', v_proj=True, dw_ks=3, stride=1, dilation=1, se_ratio=0.0, window_size=7,
attn_s=True, attn_drop=0., drop=0., drop_path=0., v_group=False, attn_pre=False):
super().__init__()
self.norm = nn.BatchNorm2d(dim_in) if norm_in else nn.Identity()
dim_mid = int(dim_in * exp_ratio)
self.has_skip = (dim_in == dim_out and stride == 1) and has_skip
self.attn_s = attn_s
if self.attn_s:
assert dim_in % dim_head == 0, 'dim should be divisible by num_heads'
self.dim_head = dim_head
self.window_size = window_size
self.num_head = dim_in // dim_head
self.scale = self.dim_head ** -0.5
self.attn_pre = attn_pre
self.qk = Conv(dim_in, int(dim_in * 2), k=1, act=False)
self.v = Conv(dim_in, dim_mid, k=1, g=self.num_head if v_group else 1, act=False)
self.attn_drop = nn.Dropout(attn_drop)
else:
if v_proj:
self.v = Conv(dim_in, dim_mid, k=1, act=act_layer)
else:
self.v = nn.Identity()
self.conv_local = Conv(dim_mid, dim_mid, k=dw_ks, s=stride, d=dilation, g=dim_mid,)
self.se = SE(dim_mid, reduction=se_ratio) if se_ratio > 0.0 else nn.Identity()
self.proj_drop = nn.Dropout(drop)
self.proj = Conv(dim_mid, dim_out, k=1, act=False)
self.drop_path = DropPath(drop_path) if drop_path else nn.Identity()
def forward(self, x):
shortcut = x
x = self.norm(x)
B, C, H, W = x.shape
if self.attn_s:
# padding
if self.window_size <= 0:
window_size_W, window_size_H = W, H
else:
window_size_W, window_size_H = self.window_size, self.window_size
pad_l, pad_t = 0, 0
pad_r = (window_size_W - W % window_size_W) % window_size_W
pad_b = (window_size_H - H % window_size_H) % window_size_H
x = F.pad(x, (pad_l, pad_r, pad_t, pad_b, 0, 0,))
n1, n2 = (H + pad_b) // window_size_H, (W + pad_r) // window_size_W
x = rearrange(x, 'b c (h1 n1) (w1 n2) -> (b n1 n2) c h1 w1', n1=n1, n2=n2).contiguous()
# attention
b, c, h, w = x.shape
qk = self.qk(x)
qk = rearrange(qk, 'b (qk heads dim_head) h w -> qk b heads (h w) dim_head', qk=2, heads=self.num_head,
dim_head=self.dim_head).contiguous()
q, k = qk[0], qk[1]
attn_spa = (q @ k.transpose(-2, -1)) * self.scale
attn_spa = attn_spa.softmax(dim=-1)
attn_spa = self.attn_drop(attn_spa)
if self.attn_pre:
x = rearrange(x, 'b (heads dim_head) h w -> b heads (h w) dim_head', heads=self.num_head).contiguous()
x_spa = attn_spa @ x
x_spa = rearrange(x_spa, 'b heads (h w) dim_head -> b (heads dim_head) h w', heads=self.num_head, h=h,
w=w).contiguous()
x_spa = self.v(x_spa)
else:
v = self.v(x)
v = rearrange(v, 'b (heads dim_head) h w -> b heads (h w) dim_head', heads=self.num_head).contiguous()
x_spa = attn_spa @ v
x_spa = rearrange(x_spa, 'b heads (h w) dim_head -> b (heads dim_head) h w', heads=self.num_head, h=h,
w=w).contiguous()
# unpadding
x = rearrange(x_spa, '(b n1 n2) c h1 w1 -> b c (h1 n1) (w1 n2)', n1=n1, n2=n2).contiguous()
if pad_r > 0 or pad_b > 0:
x = x[:, :, :H, :W].contiguous()
else:
x = self.v(x)
x = x + self.se(self.conv_local(x)) if self.has_skip else self.se(self.conv_local(x))
x = self.proj_drop(x)
x = self.proj(x)
x = (shortcut + self.drop_path(x)) if self.has_skip else x
return x
2.3 更改task.py文件
打开ultralytics->nn->modules->task.py，在脚本空白处导入函数。
from ultralytics.nn.blocks import *
之后找到模型解析函数parse_model（约在tasks.py脚本中940行左右位置，可能因代码版本不同变动），在该函数的最后一个else分支上面增加相关解析代码。
elif m is iRMB:
c2 = args[0]
args = [ch[f], *args]
2.4 更改yaml文件
yam文件解读：YOLO系列 “.yaml“文件解读_yolo yaml文件-****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。（放在该位置仅能插入该模块，具体效果未知。博主精力有限，仅完成与其他模块二次创新融合的测试，结构图见文末，代码见群文件更新。）
# 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: 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, 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-P3/16
- [-1, 2, iRMB, [512]] # 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-P4/32
- [-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)
2.5 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')
在train.py脚本中填入修改好的yaml路径，运行即可训。
三、相关改进思路（2024/11/16日群文件）
该模块可如图加入到HGBlock、RepNCSPELAN4、RepC3等模块中，代码见群文件，结构如图。自研模块与该模块融合代码及yaml文件见群文件。
⭐另外，融合上百种改进模块的YOLO项目仅79.9（含百种改进的v9），RTDETR79.9，含高性能自研模型，更易发论文，代码每周更新，欢迎点击下方小卡片加我了解。⭐
⭐⭐平均每个文章对应4-6个二创及自研融合模块⭐⭐
