<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[qq_35732321的博客]]></title><description><![CDATA[]]></description><link>https://blog.csdn.net/qq_35732321</link><language>zh-cn</language><generator>https://blog.csdn.net/</generator><copyright><![CDATA[Copyright &copy; qq_35732321]]></copyright><item><title><![CDATA[PaddleLabel标注注意事项_完整版]]></title><link>https://blog.csdn.net/qq_35732321/article/details/161721282</link><guid>https://blog.csdn.net/qq_35732321/article/details/161721282</guid><author>qq_35732321</author><pubDate>Fri, 05 Jun 2026 15:06:18 +0800</pubDate><description><![CDATA[123]]></description><category></category></item><item><title><![CDATA[paddlelabe标注注意事项]]></title><link>https://blog.csdn.net/qq_35732321/article/details/161546597</link><guid>https://blog.csdn.net/qq_35732321/article/details/161546597</guid><author>qq_35732321</author><pubDate>Sat, 30 May 2026 18:24:30 +0800</pubDate><description><![CDATA[必须用四点标注（按键盘 Q 进入四点模式），顺序为顺时针 / 逆时针（左上→右上→右下→左下）特殊符号（如 ●、○）：如果要作为有效字符识别，需在自定义字典中添加，不能用空格或其他符号代替。一个独立丝印单元一个框（如 R102 一个框，不拆 R/1/0/2）。倾斜 / 旋转字符必须画倾斜四边形框，不能画正矩形，否则矫正后会变形。每个文本框用一个 {}，多个框用逗号分隔，外层用 [] 包裹。四点顺序必须顺时针，不要乱点，否则矫正后字会歪 / 变形。模糊 / 反光 / 油污 → 统一 ###，不要猜字。]]></description><category></category></item><item><title><![CDATA[【无标题】RTX00安装paddle OCR，win11不能装最新的，也不能用GPU]]></title><link>https://blog.csdn.net/qq_35732321/article/details/161254860</link><guid>https://blog.csdn.net/qq_35732321/article/details/161254860</guid><author>qq_35732321</author><pubDate>Wed, 20 May 2026 13:24:59 +0800</pubDate><description><![CDATA[自动检测 CUDA /cuDNN → 自动装匹配的 PaddlePaddle 3.0 + PaddleOCR 3.5 + PP-OCRv5。RTX5080 不能用最新版的pp-V5。Windows 100% 兼容 ✅。]]></description><category></category></item><item><title><![CDATA[paddleocrlabel数据打标签]]></title><link>https://blog.csdn.net/qq_35732321/article/details/161167768</link><guid>https://blog.csdn.net/qq_35732321/article/details/161167768</guid><author>qq_35732321</author><pubDate>Sun, 17 May 2026 16:16:47 +0800</pubDate><description><![CDATA[由于元数据多，如果一个个打标签速度很慢，既要保证速度又要保证质量，因此把总的样本聚类，每个类别样本取一半制作标签训练模型，剩下的一半用训练好的打标签，然后用第二次的数据再训练模型。]]></description><category></category></item><item><title><![CDATA[根据颜色选择目标及测试效果]]></title><link>https://blog.csdn.net/qq_35732321/article/details/161121074</link><guid>https://blog.csdn.net/qq_35732321/article/details/161121074</guid><author>qq_35732321</author><pubDate>Fri, 15 May 2026 20:23:46 +0800</pubDate><description><![CDATA[【代码】根据颜色选择目标及测试效果。]]></description><category></category></item><item><title><![CDATA[安装paddleocr]]></title><link>https://blog.csdn.net/qq_35732321/article/details/160934232</link><guid>https://blog.csdn.net/qq_35732321/article/details/160934232</guid><author>qq_35732321</author><pubDate>Sat, 09 May 2026 20:50:21 +0800</pubDate><description><![CDATA[安装paddleocr。]]></description><category></category></item><item><title><![CDATA[安装dinov2]]></title><link>https://blog.csdn.net/qq_35732321/article/details/160930471</link><guid>https://blog.csdn.net/qq_35732321/article/details/160930471</guid><author>qq_35732321</author><pubDate>Sat, 09 May 2026 17:50:56 +0800</pubDate><description><![CDATA[Win11 + D 盘 Miniforge + RTX5080 + CUDA12.8。]]></description><category></category></item><item><title><![CDATA[判断物体方向]]></title><link>https://blog.csdn.net/qq_35732321/article/details/160869624</link><guid>https://blog.csdn.net/qq_35732321/article/details/160869624</guid><author>qq_35732321</author><pubDate>Thu, 07 May 2026 23:46:09 +0800</pubDate><description><![CDATA[【代码】判断物体方向。]]></description><category></category></item><item><title><![CDATA[对电感图像增量]]></title><link>https://blog.csdn.net/qq_35732321/article/details/160824644</link><guid>https://blog.csdn.net/qq_35732321/article/details/160824644</guid><author>qq_35732321</author><pubDate>Wed, 06 May 2026 14:38:10 +0800</pubDate><description><![CDATA[【代码】对电感图像增量。]]></description><category></category></item><item><title><![CDATA[【无标题】]]></title><link>https://blog.csdn.net/qq_35732321/article/details/160794856</link><guid>https://blog.csdn.net/qq_35732321/article/details/160794856</guid><author>qq_35732321</author><pubDate>Tue, 05 May 2026 15:24:37 +0800</pubDate><description><![CDATA[你好！ 这是你第一次使用 Markdown编辑器 所展示的欢迎页。如果你想学习如何使用Markdown编辑器, 可以仔细阅读这篇文章，了解一下Markdown的基本语法知识。我们对Markdown编辑器进行了一些功能拓展与语法支持，除了标准的Markdown编辑器功能，我们增加了如下几点新功能，帮助你用它写博客：撤销：Ctrl/Command + Z
重做：Ctrl/Command + Y
加粗：Ctrl/Command + B
斜体：Ctrl/Command + I
标题：Ctrl/Command + S]]></description><category></category></item><item><title><![CDATA[安装C++版opencv和opencv_contrib]]></title><link>https://blog.csdn.net/qq_35732321/article/details/160721120</link><guid>https://blog.csdn.net/qq_35732321/article/details/160721120</guid><author>qq_35732321</author><pubDate>Sun, 03 May 2026 01:12:23 +0800</pubDate><description><![CDATA[└── build/             # 专门用来编译（最干净，不乱）└── opencv_build/          # 你新建的总文件夹。├── opencv/            # 放 OpenCV 源码。├── opencv_contrib/    # 放扩展模块。]]></description><category></category></item><item><title><![CDATA[安装yolo26【无标题】]]></title><link>https://blog.csdn.net/qq_35732321/article/details/160720535</link><guid>https://blog.csdn.net/qq_35732321/article/details/160720535</guid><author>qq_35732321</author><pubDate>Sat, 02 May 2026 22:26:57 +0800</pubDate><description><![CDATA[✅ 系统： Ubuntu 26.04✅ 显卡： RTX 5080✅ CUDA： 12.8（已装好）✅ cuDNN： 9.21.1（已装好，完全匹配 CUDA 12.8）自动拉取最新编译的 torch/torchvision/torchaudio，支持 RTX 5080（sm_120）cu128：匹配你系统的 CUDA 12.8。–pre：装预发布（nightly）版本。]]></description><category></category></item><item><title><![CDATA[X-AnyLabelImg安装与使用]]></title><link>https://blog.csdn.net/qq_35732321/article/details/158920703</link><guid>https://blog.csdn.net/qq_35732321/article/details/158920703</guid><author>qq_35732321</author><pubDate>Wed, 11 Mar 2026 11:07:19 +0800</pubDate><description><![CDATA[X-AnyLabeling 是一款基于AI推理引擎和丰富功能特性于一体的强大辅助标注工具，其专注于实际应用，致力于为多模态数据工程师提供工业级的一站式解决方案，可自动快速进行各种复杂任务的标定。支持远程推理服务、支持GPU、FFmpeg加速推理、支持一键预测所有图像、支持图像和视频处理、支持自定义模型和二次开发、支持一键导入和导出多种标签格式，如 COCO\VOC\YOLO\DOTA\MOT\MASK\PPOCR\MMGF\VLM-R1 等；]]></description><category></category></item><item><title><![CDATA[CMake下载与安装]]></title><link>https://blog.csdn.net/qq_35732321/article/details/156546345</link><guid>https://blog.csdn.net/qq_35732321/article/details/156546345</guid><author>qq_35732321</author><pubDate>Sat, 03 Jan 2026 22:53:31 +0800</pubDate><description><![CDATA[1 官网下在最新版本cmake-4.2.1-linux-x86_64.sh或者使用命令安装下载完成后，你需要使脚本具有执行权限运行安装脚本选择安装目录‌按照提示选择安装目录，或者接受默认的安装目录。通常，默认目录是 /usr/local。验证 CMake 是否正确安装。]]></description><category></category></item><item><title><![CDATA[CFlow原理及代码解析]]></title><link>https://blog.csdn.net/qq_35732321/article/details/156023340</link><guid>https://blog.csdn.net/qq_35732321/article/details/156023340</guid><author>qq_35732321</author><pubDate>Wed, 17 Dec 2025 17:36:30 +0800</pubDate><description><![CDATA[MVTec AD 数据集
https://www.mvtec.com/company/research/datasets/mvtec-ad1.1 概念：Conditional Normalizing Flow，深度生成模型（Generative Model），无监督/弱监督异常检测（如产品表面缺陷识别）。 Google Research 在 2021 年提出（论文：CFlow-AD: Real-Time Industrial Anomaly Detection with Conditional Flows）]]></description><category></category></item><item><title><![CDATA[yolov13 tensorRT C++代码解析]]></title><link>https://blog.csdn.net/qq_35732321/article/details/155040574</link><guid>https://blog.csdn.net/qq_35732321/article/details/155040574</guid><author>qq_35732321</author><pubDate>Wed, 19 Nov 2025 23:45:00 +0800</pubDate><description><![CDATA[对检测结果做 NMS 后处理，并在原图上绘制边界框和标签；利用多线程实现“预处理”与“推理+后处理”的并行化；借助 TensorRT 加速深度学习模型推理；使用摄像头采集图像；]]></description><category></category></item><item><title><![CDATA[ONNX在C++上部署]]></title><link>https://blog.csdn.net/qq_35732321/article/details/155021344</link><guid>https://blog.csdn.net/qq_35732321/article/details/155021344</guid><author>qq_35732321</author><pubDate>Wed, 19 Nov 2025 12:54:37 +0800</pubDate><description><![CDATA[步骤 1：初始化 ONNX Runtime 环境。(2) 图片与处理preprocess.cpp。步骤 9：后处理（NMS + 解码）步骤 6：创建输入 Tensor。步骤 2：加载 ONNX 模型。步骤 3：获取模型输入输出信息。步骤 12：释放动态分配的内存。步骤 11：显示和保存结果。步骤 10：绘制检测结果。步骤 4：加载测试图像。步骤 8：获取输出数据。(3)预测结果后处理。]]></description><category></category></item><item><title><![CDATA[获取图像中制定颜色区域]]></title><link>https://blog.csdn.net/qq_35732321/article/details/154409629</link><guid>https://blog.csdn.net/qq_35732321/article/details/154409629</guid><author>qq_35732321</author><pubDate>Tue, 04 Nov 2025 22:44:28 +0800</pubDate><description><![CDATA[获取图像中制定颜色区域。]]></description><category></category></item><item><title><![CDATA[使用 C++ 部署 PyTorch 模型]]></title><link>https://blog.csdn.net/qq_35732321/article/details/154398830</link><guid>https://blog.csdn.net/qq_35732321/article/details/154398830</guid><author>qq_35732321</author><pubDate>Tue, 04 Nov 2025 15:51:24 +0800</pubDate><description><![CDATA[默认情况下会根据上一次训练保留的参数继续训练，如果要从头训练，需要删除./mlruns文件夹，新训练的模型回报存在 ./models路径。在build文件夹有可执行文件torch-cpp-deployment，运行torch-cpp-deployment。把模型转换为TorchScript格式，把训练好的模型放在 ./models/路径下。(1) 训练PyTorch model并保存，把模型转换成TorchScript。先用python训练网络，在转换模型格式，最后用C++部署模型。]]></description><category></category></item><item><title><![CDATA[LprNet车牌检测]]></title><link>https://blog.csdn.net/qq_35732321/article/details/154359635</link><guid>https://blog.csdn.net/qq_35732321/article/details/154359635</guid><author>qq_35732321</author><pubDate>Mon, 03 Nov 2025 20:28:38 +0800</pubDate><description><![CDATA[下载tlt（Transfer Learning Toolkit）格式预训练模型，参考网址：https://catalog.ngc.nvidia.com/orgs/nvidia/teams/tao/models/lprnet。CPPD（https://github.com/detectRecog/CCPD）数据集训练了一个车牌检测模型（337135张图片，YOLOv5s模型）参考代码：https://github.com/NVIDIA-AI-IOT/deepstream_lpr_app。]]></description><category></category></item></channel></rss>