<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"
	xmlns:media="http://search.yahoo.com/mrss/"
	xmlns:content="http://purl.org/rss/1.0/modules/content/"
	xmlns:wfw="http://wellformedweb.org/CommentAPI/"
	xmlns:dc="http://purl.org/dc/elements/1.1/"
	xmlns:atom="http://www.w3.org/2005/Atom"
	xmlns:sy="http://purl.org/rss/1.0/modules/syndication/"
	xmlns:slash="http://purl.org/rss/1.0/modules/slash/"
	>

<channel>
	<title>机器学习 - 四号程序员</title>
	<atom:link href="https://www.coder4.com/archives/category/ml/feed" rel="self" type="application/rss+xml" />
	<link>https://www.coder4.com</link>
	<description>Keep It Simple and Stupid</description>
	<lastBuildDate>Tue, 24 Mar 2026 21:40:30 +0000</lastBuildDate>
	<language>en-US</language>
	<sy:updatePeriod>
	hourly	</sy:updatePeriod>
	<sy:updateFrequency>
	1	</sy:updateFrequency>
	<generator>https://wordpress.org/?v=6.8.5</generator>
	<item>
		<title>ultralytics的yolov11模型直接转rknn运行</title>
		<link>https://www.coder4.com/archives/8280</link>
					<comments>https://www.coder4.com/archives/8280#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Thu, 27 Feb 2025 04:02:43 +0000</pubDate>
				<category><![CDATA[Linux]]></category>
		<category><![CDATA[机器学习]]></category>
		<category><![CDATA[npu]]></category>
		<category><![CDATA[radxa]]></category>
		<category><![CDATA[rk3566]]></category>
		<category><![CDATA[rknn]]></category>
		<category><![CDATA[yolo]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8280</guid>

					<description><![CDATA[ultralytics最近官方支持了rknn模型的导入，整体流程比用rknntool简单了不少，当然也有坑，记录下。我用的是yolov11，不确定对于v8等是否能用，大家可以评论区反馈我。 PS：如果你需要瑞莎radxa、香橙派orange pi的 屏幕、外壳、散热器，可以来我的咸鱼(coder4)看看，欢迎扫码关注 1 PC上模型转换 环境 python -m venv ./ultralytics-env source ./ultralytics-env/bin/[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8280/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>[转]记录如何在RK3588板子上跑通paddle的OCR模型</title>
		<link>https://www.coder4.com/archives/8262</link>
					<comments>https://www.coder4.com/archives/8262#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Fri, 06 Dec 2024 16:35:42 +0000</pubDate>
				<category><![CDATA[Linux]]></category>
		<category><![CDATA[机器学习]]></category>
		<category><![CDATA[OCR]]></category>
		<category><![CDATA[rknn]]></category>
		<category><![CDATA[深度学习]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8262</guid>

					<description><![CDATA[原文链接：https://blog.csdn.net/m0_60657960/article/details/143209851 参考链接：https://blog.csdn.net/Fzq1021/article/details/133508218 1 PC电脑是Ubuntu22.04系统中完成环境搭建(板子是20.04） 安装模型转换环境 conda create -n rknn2 python==3.10 conda activate rknn2 安装Ubuntu依[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8262/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>[转]rk3588部署yolov8视频目标检测教程</title>
		<link>https://www.coder4.com/archives/8236</link>
					<comments>https://www.coder4.com/archives/8236#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Tue, 29 Oct 2024 06:27:31 +0000</pubDate>
				<category><![CDATA[Linux]]></category>
		<category><![CDATA[机器学习]]></category>
		<category><![CDATA[npu]]></category>
		<category><![CDATA[rk]]></category>
		<category><![CDATA[yolo]]></category>
		<category><![CDATA[目标检测]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8236</guid>

					<description><![CDATA[转载自《rk3588部署yolov8视频目标检测教程》 1. 环境配置 1.1 训练环境和onnx（电脑端执行） #使用conda创建一个python环境 conda create -n torch python=3.9 #激活环境 conda activate torch #安装yolov8 pip install ultralytics -i https://pypi.tuna.tsinghua.edu.cn/simple #onnx安装 pip i[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8236/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>[转]rk3588对npu的再探索，yolov5使用rknn模型推理教程</title>
		<link>https://www.coder4.com/archives/8229</link>
					<comments>https://www.coder4.com/archives/8229#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Wed, 23 Oct 2024 04:53:11 +0000</pubDate>
				<category><![CDATA[机器学习]]></category>
		<category><![CDATA[npu]]></category>
		<category><![CDATA[rknn]]></category>
		<category><![CDATA[深度学习]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8229</guid>

					<description><![CDATA[原文转载自《rk3588对npu的再探索，yolov5使用rknn模型推理教程》 🍉零、引言 本文完成于2022-07-02 22:22:27。 博主刚开始在瑞芯微ITX-3588J-8K的开发板上跑了官方的yolov5目标检测算法，检测了ip相机rtsp视频流，但是每帧处理需要833ms左右，和放PPT一样。本来想使用tensorrt进行加速推理，但前提需要cuda，rk的板子上都是arm的手机gpu，没有nvidia的cuda，所以不能这样适配。那么转过来，使用开发板自带的NPU进行加速[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8229/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>[转]rk3588使用npu进行模型转换和推理，加速AI应用落地</title>
		<link>https://www.coder4.com/archives/8208</link>
					<comments>https://www.coder4.com/archives/8208#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Wed, 23 Oct 2024 04:11:47 +0000</pubDate>
				<category><![CDATA[Linux]]></category>
		<category><![CDATA[机器学习]]></category>
		<category><![CDATA[npu]]></category>
		<category><![CDATA[rknn]]></category>
		<category><![CDATA[推理]]></category>
		<category><![CDATA[模型]]></category>
		<category><![CDATA[转换]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8208</guid>

					<description><![CDATA[转载自：《rk3588使用npu进行模型转换和推理，加速AI应用落地》 🍉零、引言 博主在瑞芯微RK3588的开发板上跑了deepsort跟踪算法，从IP相机中的server拉取rtsp视频流，但是fps只有1.2，和放PPT一样卡顿，无法投入实际应用。本来想使用tensorrt进行加速推理，但是前提需要cuda，rk的板子上都是Arm的手机gpu，没有Nvidia的cuda，所以这条路行不通。那么转过来，使用开发板自带的NPU进行加速推理，岂不是更加可行，而且它本身就是深度学习嵌入式板子，[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8208/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>YoloV5-Lite目标检测之“微调 + 模型转换”</title>
		<link>https://www.coder4.com/archives/8201</link>
					<comments>https://www.coder4.com/archives/8201#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Thu, 12 Sep 2024 06:47:09 +0000</pubDate>
				<category><![CDATA[机器学习]]></category>
		<category><![CDATA[ncnn]]></category>
		<category><![CDATA[YoloV5-Lite]]></category>
		<category><![CDATA[微调]]></category>
		<category><![CDATA[模型转换]]></category>
		<category><![CDATA[自定义数据]]></category>
		<category><![CDATA[训练]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8201</guid>

					<description><![CDATA[在YoloV5-Lite目标检测之“安装推理”中，我们完成了安装和预训练权重的推理，下面介绍自定义训练数据、模型转换(ncnn) 1 训练数据准备 . ├── train │   ├── 000000000049.jpg │   ├── 000000000049.txt ...... │   ├── 000000581880.txt │   ├── 000000581900.jpg │   └── 000000581900.txt └── val ├── 00000[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8201/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>YoloV5-Lite目标检测之“安装推理”</title>
		<link>https://www.coder4.com/archives/8199</link>
					<comments>https://www.coder4.com/archives/8199#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Thu, 12 Sep 2024 04:34:24 +0000</pubDate>
				<category><![CDATA[心情随笔]]></category>
		<category><![CDATA[机器学习]]></category>
		<category><![CDATA[YoloV5-Lite]]></category>
		<category><![CDATA[深度学习]]></category>
		<category><![CDATA[目标检测]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8199</guid>

					<description><![CDATA[1 安装 conda activate py38 git clone https://github.com/ppogg/YOLOv5-Lite pip install -r requirements.txt 2 下载预训练的权重 预训练权重可以在官网下载，我这里下载的是v5lite-s 3 推理 图片推理 python3 ./detect.py --weights ./weights/v5lite-e.pt --source ././python_demo/openvino/[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8199/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>coco2017数据集标注转化为yolo的标注格式</title>
		<link>https://www.coder4.com/archives/8198</link>
					<comments>https://www.coder4.com/archives/8198#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Thu, 12 Sep 2024 04:19:39 +0000</pubDate>
				<category><![CDATA[机器学习]]></category>
		<category><![CDATA[coco]]></category>
		<category><![CDATA[yolo]]></category>
		<category><![CDATA[标注]]></category>
		<category><![CDATA[转换]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8198</guid>

					<description><![CDATA[我的需求是只保留人（和没有人的副样本），抽样，其他忽略 import json import os import shutil import random def coco_to_yolo_bbox(image_width, image_height, bbox): """ Convert bbox from COCO format to YOLO format. COCO format: [top_left_x, top_left_y, width,[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8198/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>NanoDet目标检测之&quot;微调训练&quot;</title>
		<link>https://www.coder4.com/archives/8194</link>
					<comments>https://www.coder4.com/archives/8194#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Wed, 11 Sep 2024 06:54:19 +0000</pubDate>
				<category><![CDATA[机器学习]]></category>
		<category><![CDATA[nanodet]]></category>
		<category><![CDATA[微调]]></category>
		<category><![CDATA[自定义数据]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8194</guid>

					<description><![CDATA[在 《NanoDet目标检测之"搭建预测篇"》 中，我们搭建了NanoDet的环境，并用默认权重做了简单的预测，本节我们继续用自己的数据做微调。 1 准备数据 NanoDet支持yolo或者coco格式的标注，自行准备即可，我这里以yolo为例，目录结构如下： TODO xxx 2 配置 我们需要融合下两个配置文件，复制： cp nanodet-plus-m_320.yml nanodet-plus-m_320_face.yml 修改1，保存位置： save_dir: wo[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8194/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>NanoDet目标检测之&quot;搭建预测篇&quot;</title>
		<link>https://www.coder4.com/archives/8192</link>
					<comments>https://www.coder4.com/archives/8192#respond</comments>
		
		<dc:creator><![CDATA[coder4]]></dc:creator>
		<pubDate>Wed, 11 Sep 2024 06:47:58 +0000</pubDate>
				<category><![CDATA[机器学习]]></category>
		<category><![CDATA[nanodet]]></category>
		<category><![CDATA[ncnn]]></category>
		<category><![CDATA[搭建]]></category>
		<category><![CDATA[目标检测]]></category>
		<guid isPermaLink="false">https://www.coder4.com/?p=8192</guid>

					<description><![CDATA[NanoDet是一个基于ShuffleNetV2的轻量级的目标检测模型，配合ncnn框架加速后，在中端Android机型能做到20fps+ 1 安装环境 为了避免冲突，需要安装好conda，Python 3.8+ 安装依赖包 git clone https://github.com/RangiLyu/nanodet.git cd nanodet pip install -r requirements.txta 安装代码 python setup.py develop 2 下[......] 继续阅读]]></description>
		
					<wfw:commentRss>https://www.coder4.com/archives/8192/feed</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
	</channel>
</rss>
