<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[Felaim的博客]]></title><description><![CDATA[尽人事，听天命！O(∩_∩)O哈哈~]]></description><link>https://blog.csdn.net/Felaim</link><language>zh-cn</language><generator>https://blog.csdn.net/</generator><copyright><![CDATA[Copyright &copy; Felaim]]></copyright><item><title><![CDATA[【World Model】世界模型：基本原理、技术路线与应用价值]]></title><link>https://blog.csdn.net/Felaim/article/details/169343484</link><guid>https://blog.csdn.net/Felaim/article/details/169343484</guid><author>Felaim</author><pubDate>Wed, 28 Oct 2026 10:21:24 +0800</pubDate><description><![CDATA[世界模型总览：基本原理、主要技术路线与在自动驾驶等场景的应用价值，梳理代表工作与用途。]]></description><category></category></item><item><title><![CDATA[【SLAM】NetVLAD论文精读]]></title><link>https://blog.csdn.net/Felaim/article/details/169343266</link><guid>https://blog.csdn.net/Felaim/article/details/169343266</guid><author>Felaim</author><pubDate>Wed, 28 Oct 2026 10:20:16 +0800</pubDate><description><![CDATA[NetVLAD 精读：可微 VLAD 池化嵌入 CNN，用弱监督做视觉地点识别，服务回环与重定位。]]></description><category></category></item><item><title><![CDATA[【VLA与模仿学习】LoRA低秩适配论文精读]]></title><link>https://blog.csdn.net/Felaim/article/details/169250060</link><guid>https://blog.csdn.net/Felaim/article/details/169250060</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 11:09:20 +0800</pubDate><description><![CDATA[LoRA 精读：用低秩旁路适配大模型，冻结底座只训少量参数，实现高效多任务微调与部署合并。]]></description><category></category></item><item><title><![CDATA[【导航与地图】DiffusionDrive截断扩散端到端驾驶]]></title><link>https://blog.csdn.net/Felaim/article/details/169249687</link><guid>https://blog.csdn.net/Felaim/article/details/169249687</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 11:04:09 +0800</pubDate><description><![CDATA[DiffusionDrive 精读：用截断扩散做端到端多模态轨迹规划，在实时性与多模态覆盖之间折中。]]></description><category></category></item><item><title><![CDATA[【VLA与模仿学习】VADv2论文精读]]></title><link>https://blog.csdn.net/Felaim/article/details/169249205</link><guid>https://blog.csdn.net/Felaim/article/details/169249205</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 10:59:52 +0800</pubDate><description><![CDATA[VADv2 精读：向量化端到端自动驾驶，用概率规划对动作分布建模，提升闭环驾驶表现。]]></description><category></category></item><item><title><![CDATA[【World Model】Drive-WM条件编码-视角分解-扩散与规划全流程详解]]></title><link>https://blog.csdn.net/Felaim/article/details/169248623</link><guid>https://blog.csdn.net/Felaim/article/details/169248623</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 10:55:11 +0800</pubDate><description><![CDATA[Drive-WM 全流程详解：条件编码、多视角分解、潜空间扩散与规划闭环，对照论文 §3–5 与相关基础工作。]]></description><category></category></item><item><title><![CDATA[【World Model】如何快速抓住论文重点-以Drive-WM为例]]></title><link>https://blog.csdn.net/Felaim/article/details/169248150</link><guid>https://blog.csdn.net/Felaim/article/details/169248150</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 10:50:16 +0800</pubDate><description><![CDATA[以 Drive-WM 为例，拆解「问题—方法—证据」卡片式读法：先建全局地图再深入细节，提高论文精读效率。]]></description><category></category></item><item><title><![CDATA[【SLAM】ConvNeXt论文精读]]></title><link>https://blog.csdn.net/Felaim/article/details/169247430</link><guid>https://blog.csdn.net/Felaim/article/details/169247430</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 10:45:06 +0800</pubDate><description><![CDATA[ConvNeXt 精读：把 Transformer 时代的训练与结构习惯迁回纯卷积，证明现代化 ConvNet 仍具竞争力。]]></description><category></category></item><item><title><![CDATA[【VLA与模仿学习】CLIP论文精读]]></title><link>https://blog.csdn.net/Felaim/article/details/169246824</link><guid>https://blog.csdn.net/Felaim/article/details/169246824</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 10:39:33 +0800</pubDate><description><![CDATA[CLIP 精读：用自然语言监督做大规模图文对比学习，实现可迁移的视觉表征与零样本分类/检索。]]></description><category></category></item><item><title><![CDATA[【SLAM】Swin-Transformer论文精读]]></title><link>https://blog.csdn.net/Felaim/article/details/169246325</link><guid>https://blog.csdn.net/Felaim/article/details/169246325</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 10:34:22 +0800</pubDate><description><![CDATA[Swin Transformer 精读：用移位窗口自注意力做分层视觉骨干，兼顾效率与多尺度特征，支撑检测与分割等下游任务。]]></description><category></category></item><item><title><![CDATA[【SLAM】VGGT-Diff几何路由扩散新视角合成]]></title><link>https://blog.csdn.net/Felaim/article/details/169245753</link><guid>https://blog.csdn.net/Felaim/article/details/169245753</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 10:30:03 +0800</pubDate><description><![CDATA[VGGT-Diff 精读：把视觉几何基础模型与视频扩散结合，用几何路由做稀疏视角新视角合成，解读方法、框架与实验。]]></description><category></category></item><item><title><![CDATA[【SLAM】YOLOE实时开提示检测分割]]></title><link>https://blog.csdn.net/Felaim/article/details/169245199</link><guid>https://blog.csdn.net/Felaim/article/details/169245199</guid><author>Felaim</author><pubDate>Tue, 27 Oct 2026 10:24:52 +0800</pubDate><description><![CDATA[YOLOE 精读：实时“看见一切”的开放提示检测与分割模型，支持文本、视觉与无提示三种模式，解读方法细节与实验结论。]]></description><category></category></item><item><title><![CDATA[【SLAM】RAFT光流原理白话深讲]]></title><link>https://blog.csdn.net/Felaim/article/details/169147490</link><guid>https://blog.csdn.net/Felaim/article/details/169147490</guid><author>Felaim</author><pubDate>Mon, 26 Oct 2026 10:46:28 +0800</pubDate><description><![CDATA[RAFT 光流原理白话深讲：特征编码、全点对相关体积与循环更新算子三块积木，并与 DROID-SLAM 对照。]]></description><category></category></item><item><title><![CDATA[【SLAM】DROID-SLAM原理白话深讲]]></title><link>https://blog.csdn.net/Felaim/article/details/169147279</link><guid>https://blog.csdn.net/Felaim/article/details/169147279</guid><author>Felaim</author><pubDate>Mon, 26 Oct 2026 10:44:22 +0800</pubDate><description><![CDATA[DROID-SLAM 原理白话深讲：从输入、术语、模型结构、损失到前后端推理，讲清循环光流更新与稠密光束法平差，以及单目/双目/RGB-D 接入方式。]]></description><category></category></item><item><title><![CDATA[【SLAM】SLAM数据集相机规格说明]]></title><link>https://blog.csdn.net/Felaim/article/details/169146729</link><guid>https://blog.csdn.net/Felaim/article/details/169146729</guid><author>Felaim</author><pubDate>Mon, 26 Oct 2026 10:41:27 +0800</pubDate><description><![CDATA[SLAM 数据集相机规格说明：整理 EuRoC MAV、TUM RGB-D、HILTI SLAM Challenge 三个常用基准的相机硬件、安装位置与标定参数，并给出使用建议。]]></description><category></category></item><item><title><![CDATA[【SLAM】MobileSAM]]></title><link>https://blog.csdn.net/Felaim/article/details/169146411</link><guid>https://blog.csdn.net/Felaim/article/details/169146411</guid><author>Felaim</author><pubDate>Mon, 26 Oct 2026 10:37:32 +0800</pubDate><description><![CDATA[MobileSAM 精读：通过解耦蒸馏把 SAM 的图像编码器换成轻量 ViT，让分割一切模型能在移动端实时运行。]]></description><category></category></item><item><title><![CDATA[【SLAM】3D-VLM-Grounding主流方案]]></title><link>https://blog.csdn.net/Felaim/article/details/169146133</link><guid>https://blog.csdn.net/Felaim/article/details/169146133</guid><author>Felaim</author><pubDate>Mon, 26 Oct 2026 10:34:12 +0800</pubDate><description><![CDATA[3D VLM Grounding 主流方案梳理：按 grounding 方式整理技术谱系、代表工作与基准，并讨论选型与 2D grounding 的关系。]]></description><category></category></item><item><title><![CDATA[【SLAM】SLAM-Former论文与代码导读]]></title><link>https://blog.csdn.net/Felaim/article/details/169145640</link><guid>https://blog.csdn.net/Felaim/article/details/169145640</guid><author>Felaim</author><pubDate>Mon, 26 Oct 2026 10:31:27 +0800</pubDate><description><![CDATA[SLAM-Former 论文与代码导读：从仓库结构、概念到代码符号、推理主循环、注意力分支到 DivPrune 实现，并与传统 SLAM 对照。]]></description><category></category></item><item><title><![CDATA[【SLAM】SLAM-Former机制FAQ深讲]]></title><link>https://blog.csdn.net/Felaim/article/details/169145407</link><guid>https://blog.csdn.net/Felaim/article/details/169145407</guid><author>Felaim</author><pubDate>Mon, 26 Oct 2026 10:27:33 +0800</pubDate><description><![CDATA[SLAM-Former 机制 FAQ：关键帧、地图 token、KV cache 增量更新、后端、建图与回环、重定位等问题逐一拆解。]]></description><category></category></item><item><title><![CDATA[【SLAM】SLAM-Former]]></title><link>https://blog.csdn.net/Felaim/article/details/169145207</link><guid>https://blog.csdn.net/Felaim/article/details/169145207</guid><author>Felaim</author><pubDate>Mon, 26 Oct 2026 10:25:33 +0800</pubDate><description><![CDATA[SLAM-Former 精读：把单目稠密 SLAM 的前端跟踪与后端全局优化放进同一个 Transformer，用地图 token 与 KV cache 取代传统后端，并报告在 Replica / ScanNet 上的表现。]]></description><category></category></item></channel></rss>