<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[presenttoday的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/presenttoday</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; presenttoday]]></copyright><item><title><![CDATA[ [CVPR 2024] FinePOSE 源码阅读(二)：训练流程、数据流水线与可视化]]></title><link>https://blog.youkuaiyun.com/presenttoday/article/details/155202113</link><guid>https://blog.youkuaiyun.com/presenttoday/article/details/155202113</guid><author>presenttoday</author><pubDate>Mon, 24 Nov 2025 19:24:32 +0800</pubDate><description><![CDATA[在FinePOSE 源码阅读(一)，我们分析了 FinePOSE 的核心模型代码（mixste_finepose.py）。这篇博客主要来读一下入口文件 main.py。这个文件负责整个项目的运行逻辑，包括怎么加载数据集、怎么构造 Prompt、以及训练和测试的具体流程。]]></description><category></category></item><item><title><![CDATA[ [CVPR 2024] FinePOSE 源码阅读(一)：核心模型架构与跨模态交互]]></title><link>https://blog.youkuaiyun.com/presenttoday/article/details/155168278</link><guid>https://blog.youkuaiyun.com/presenttoday/article/details/155168278</guid><author>presenttoday</author><pubDate>Mon, 24 Nov 2025 15:53:45 +0800</pubDate><description><![CDATA[在 3D 人体姿态估计领域，如何有效地利用文本语义信息一直是一个难点。CVPR 2024 的FinePOSE提出了一种基于扩散模型的细粒度提示驱动框架，通过引入 CLIP 和多层次的 Prompt 机制，显著提升了姿态估计的精度。本文将结合论文架构图，逐行解析FinePOSE 的核心代码实现（mixste_finepose.py），理解这个模型是如何一步步将文本、时间和空间信息融合在一起的。]]></description><category></category></item><item><title><![CDATA[【论文阅读笔记】Discovering “Semantics“ in Super-Resolution Networks]]></title><link>https://blog.youkuaiyun.com/presenttoday/article/details/143885387</link><guid>https://blog.youkuaiyun.com/presenttoday/article/details/143885387</guid><author>presenttoday</author><pubDate>Wed, 20 Nov 2024 11:36:36 +0800</pubDate><description><![CDATA[（如图b、图c）（作者由此。]]></description><category></category></item><item><title><![CDATA[【论文阅读笔记】TR-MISR: Multiimage Super-Resolution Based on Feature Fusion With Transformers]]></title><link>https://blog.youkuaiyun.com/presenttoday/article/details/143729068</link><guid>https://blog.youkuaiyun.com/presenttoday/article/details/143729068</guid><author>presenttoday</author><pubDate>Wed, 13 Nov 2024 10:02:41 +0800</pubDate><description><![CDATA[卫星采集的图像序列往往具有多视角和较长的时间跨度，因此将多个低分辨率视图融合为一张细节丰富的高分辨率图像是一个具有挑战性的问题。然而，大多数基于深度学习的MISR方法无法充分利用多张图像进行融合。它们的融合模块——无法很好地适应时间相关性较弱的图像序列多图像超分辨率——提出该模型(编码器、融合模块、解码器)。主要解决输入图像顺序敏感和特征融合不足的问题。]]></description><category></category></item><item><title><![CDATA[【论文阅读笔记】Accurate Image Super-Resolution Using Very Deep Convolutional Networks]]></title><link>https://blog.youkuaiyun.com/presenttoday/article/details/143555233</link><guid>https://blog.youkuaiyun.com/presenttoday/article/details/143555233</guid><author>presenttoday</author><pubDate>Wed, 06 Nov 2024 09:42:01 +0800</pubDate><description><![CDATA[VDSR模型学习]]></description><category></category></item></channel></rss>