<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[qq_273058791的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/qq_42672745</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; qq_42672745]]></copyright><item><title><![CDATA[【Kaggle】House Prices - Advanced Regression Techniques]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/129003715</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/129003715</guid><author>qq_42672745</author><pubDate>Mon, 13 Feb 2023 09:50:51 +0800</pubDate><description><![CDATA[【代码】【Kaggle】House Prices - Advanced Regression Techniques。]]></description><category></category></item><item><title><![CDATA[【2021-CVPR】clDice - a Novel Topology-Preserving Loss Function for Tubular Structure Segmentation]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/128942197</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/128942197</guid><author>qq_42672745</author><pubDate>Wed, 08 Feb 2023 23:09:03 +0800</pubDate><description><![CDATA[保留拓扑精度对于一些任务很重要，比如对血管分割来分析血流。目前针对管状曲线结构的语义分网络，常用的两个定量评估标准为：1）基于重叠的度量，例如Dice；2）基于体积距离度量。目前大多问题关注点为局部管状结构和全局形成的网络，二者之中最重要的是全局网络拓扑的连通性。以血管数据为例，对全局平均损失的训练会导致对大血管的体积分割的强烈偏向。（如下紫色区域相对于红色区域）

针对二值几何图形，根据不同的拓扑定义，基于细化和中间区的各种算法已证明是可以保留拓扑关系的。然而目前还没提出端到端的损失函数。本文借此提出了一]]></description><category></category></item><item><title><![CDATA[【2021-IEEE】Image Segmentation Using Deep Learning: A Survey]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/128162221</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/128162221</guid><author>qq_42672745</author><pubDate>Sat, 03 Dec 2022 15:28:14 +0800</pubDate><description><![CDATA[作者在本次调查中，涵盖了语义和实例级分割的广泛开创性工作，包括全卷积像素标记网络，编码器-解码器体系结构，多尺度以及基于金字塔的方法，递归网络，视觉注意模型和对抗环境中的生成模型。研究了这些深度学习模型的相似性，优势和挑战，研究了使用最广泛的数据集，报告了性能，并讨论了该领域有希望的未来研究方向。结论：综述总结]]></description><category></category></item><item><title><![CDATA[【2021-TITS】Deep Learning in Lane Marking Detection: A Survey]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/127958790</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/127958790</guid><author>qq_42672745</author><pubDate>Tue, 22 Nov 2022 09:52:45 +0800</pubDate><description><![CDATA[创新点：1.总结深度学习网络架构、优化目标2.总结了相关现有数据集不足：分析：结论：综述总结。]]></description><category></category></item><item><title><![CDATA[【2019-TGRS】Aerial LaneNet: Lane-Marking Semantic Segmentation in Aerial Imagery Using Wavelet-Enhanc]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/127860729</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/127860729</guid><author>qq_42672745</author><pubDate>Wed, 16 Nov 2022 11:44:21 +0800</pubDate><description><![CDATA[创新点：1.建立了航片路面标志数据集2.在不同步骤中将输入图像的离散小波变换插入到FCNNs中来保存高频信息（包括标线），从而专注于小尺寸车道标记不足：分析：积结论：SOTA、end-to-end、pixels-to-pixels。]]></description><category></category></item><item><title><![CDATA[【2015-MICCAI】U-Net: Convolutional Networks for Biomedical Image Segmentation]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/127803543</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/127803543</guid><author>qq_42672745</author><pubDate>Fri, 11 Nov 2022 16:45:05 +0800</pubDate><description><![CDATA[精读论文U-Net: Convolutional Networks for Biomedical Image Segmentation（附翻译和代码）]]></description><category></category></item><item><title><![CDATA[【2015-CVPR】Fully Convolutional Networks for Semantic Segmentation]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/127773587</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/127773587</guid><author>qq_42672745</author><pubDate>Thu, 10 Nov 2022 17:02:30 +0800</pubDate><description><![CDATA[创新点：1.端到端，逐像素预测，自监督预训练2.接受任意大小输入，推理并学习产生相应大小输出3.将浅层信息与深层信息相结合不足：分析：结论：SOTA、end-to-end、pixels-to-pixels。]]></description><category></category></item><item><title><![CDATA[【2017-VCIP】LinkNet: Exploiting Encoder Representations for Efficient Semantic Segmentation]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/127733117</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/127733117</guid><author>qq_42672745</author><pubDate>Wed, 09 Nov 2022 10:51:09 +0800</pubDate><description><![CDATA[1.编码器和解码器直接相连，共享了学习信息，保留了通常丢失掉的空间信息2.编码器用ResNet18，编解码器参数更少不足：123结论：CamVid中的SOTA、]]></description><category></category></item><item><title><![CDATA[芝士目录]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/127624753</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/127624753</guid><author>qq_42672745</author><pubDate>Mon, 31 Oct 2022 22:24:08 +0800</pubDate><description><![CDATA[一些常识]]></description><category></category></item><item><title><![CDATA[【2018-AAAI】Spatial As Deep: Spatial CNN for Traffic Scene Understanding]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/127517224</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/127517224</guid><author>qq_42672745</author><pubDate>Mon, 31 Oct 2022 20:38:07 +0800</pubDate><description><![CDATA[提出面向RGB图像针对车道线的语义分割网络SCNN，实现行和列上的信息传递发布车道线检测数据集CULane Dataset（三次样条曲线标注四条线）不足：只有四条车道线；基于语义分割；代码配置？1.分割单通道多通道都还行2.细长结构信息，可解决外显不明显问题3.有益于大目标分割结论：传递空间信息 、易集成、端到端、1st in a challenge。]]></description><category></category></item><item><title><![CDATA[【2022-CVPR】Rethinking Efficient Lane Detection via Curve Modeling]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/127323953</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/127323953</guid><author>qq_42672745</author><pubDate>Fri, 14 Oct 2022 18:25:02 +0800</pubDate><description><![CDATA[提出用参数贝塞尔曲线模型进行端到端检测RGB图车道线的方法提出基于变形卷积的特征反转融合来利用对称性不足：特征翻转融合是专为前置相机设计的，很难集成到边缘设备上1.针对前置摄像头拍的RGB图算法能否应用到俯视单通道强度图上？2.能否将原来的(实线虚线)目标检测转化为关键点检测？可能后续精度要好些（关键点检测和目标检测和语义分割的关系）3.啥是可变形卷积？结论：a new SOTA；低延时，小模型，端到端，anchor-free，NMS-free。]]></description><category></category></item><item><title><![CDATA[【重装系统】正版Windows系统重装总结]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126466819</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126466819</guid><author>qq_42672745</author><pubDate>Mon, 22 Aug 2022 15:58:39 +0800</pubDate><description><![CDATA[闲来无事摸鱼一波~最近较为流行的Windows系统有和，根据安装后的系统组成又可分为和。但网上的指南鱼龙混杂且不乏好多破解版的教程，所以今天就此来总结一下，作为笔记供日后查阅。哈哈哈哈哈🐱‍🏍。]]></description><category></category></item><item><title><![CDATA[【总结】深度学习阶段性总结]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126451601</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126451601</guid><author>qq_42672745</author><pubDate>Mon, 22 Aug 2022 08:08:49 +0800</pubDate><description><![CDATA[深度学习神经网络学习的阶段性总结🐱‍🏍🐱‍🏍🐱‍🏍]]></description><category></category></item><item><title><![CDATA[【YOLO系列】YOLOv6,v7,X]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126451469</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126451469</guid><author>qq_42672745</author><pubDate>Mon, 22 Aug 2022 08:08:38 +0800</pubDate><description><![CDATA[YOLOv6由美团于2022年提出，并非官方，目前尚无论文，让子弹再飞一会✈️，挖个坑先⛳。YOLOv7由YOLOv4原班人马于2022年提出，让子弹再飞一会✈️，挖个坑先。YOLOX由旷世于2021年提出，挖个坑先。]]></description><category></category></item><item><title><![CDATA[【YOLO系列】YOLOv5]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126451373</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126451373</guid><author>qq_42672745</author><pubDate>Mon, 22 Aug 2022 08:08:20 +0800</pubDate><description><![CDATA[YOLOv5作者并未发表论文，紧跟YOLOv4发布，性能与YOLOv4不分伯仲，同样也是当时最先进的目标检测技术，并在推理速度上是当时最强。]]></description><category></category></item><item><title><![CDATA[【YOLO系列】YOLOv4]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126451258</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126451258</guid><author>qq_42672745</author><pubDate>Mon, 22 Aug 2022 08:08:04 +0800</pubDate><description><![CDATA[YOLOv4是对YOLOv3的一个改进，总结了几乎所有的检测技巧，引用文献量**超过100篇**，又提出了一点技巧，然后经过筛选，排列组合，挨个实验哪些方法有效。实际上是一种技巧的堆叠，没有提出实质性的新内容，但确实提高了性能。
]]></description><category></category></item><item><title><![CDATA[【YOLO系列】YOLOv3]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126451186</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126451186</guid><author>qq_42672745</author><pubDate>Mon, 22 Aug 2022 08:07:48 +0800</pubDate><description><![CDATA[YOLO系列的第三个版本，在YOLOv2基础上做了一些小改进，文章篇幅不长，核心思想和YOLOv2差不多。]]></description><category></category></item><item><title><![CDATA[【YOLO系列】YOLOv2]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126451100</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126451100</guid><author>qq_42672745</author><pubDate>Mon, 22 Aug 2022 08:07:33 +0800</pubDate><description><![CDATA[YOLO系列的第二个版本，可检测出超过9000种目标类别，并可运行在不同的尺寸之下，在保持原有速度优势的条件下使得精度得以提升。同时提出了在目标检测数据集和分类数据集上联合训练的方法，取得了很好的效果。]]></description><category></category></item><item><title><![CDATA[【YOLO系列】YOLOv1]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126450982</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126450982</guid><author>qq_42672745</author><pubDate>Mon, 22 Aug 2022 08:07:11 +0800</pubDate><description><![CDATA[以前的目标检测是利用分类器来执行检测任务，而YOLO将目标检测看作关于边界框和相关类别的回归问题，直接从完整的图像上预测边界框和类别概率，可以实现端到端。推理速度45FPS 448×448，准确率63.4mAP。]]></description><category></category></item><item><title><![CDATA[【RCNN系列】Mask RCNN]]></title><link>https://blog.youkuaiyun.com/qq_42672745/article/details/126450870</link><guid>https://blog.youkuaiyun.com/qq_42672745/article/details/126450870</guid><author>qq_42672745</author><pubDate>Sun, 21 Aug 2022 14:49:04 +0800</pubDate><description><![CDATA[Mask RCNN是在Faster RCNN的基础上提出的实例(语义)分割网络，它通过添加一个分支与现有的用于bbox识别的分支并行来预测对象的mask，可有效检测图中目标，并为每个实例生成高质量的分割mask。说明了不同任务之间互相借鉴的可能性，也启发我们在深钻某一领域的同时也要涉猎广泛，其他任务的方法也许是解决你问题的新思路。]]></description><category></category></item></channel></rss>