<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[m0_60273616的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/m0_60273616</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; m0_60273616]]></copyright><item><title><![CDATA[BEVFormer论文总结]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/144676647</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/144676647</guid><author>m0_60273616</author><pubDate>Mon, 23 Dec 2024 20:28:43 +0800</pubDate><description><![CDATA[BEVFormer证明了利用多相机输入的时空信息可以显著提高视觉感知模型的性能。BEVFormer所展示的优势，如更准确的速度估计和在低可见物体上更高的召回率，对于构建更好和更安全的自动驾驶系统至关重要。]]></description><category></category></item><item><title><![CDATA[MambaMIL: Enhancing Long Sequence Modeling with Sequence Reordering in Computational Pathology（MICCA]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/141962010</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/141962010</guid><author>m0_60273616</author><pubDate>Fri, 06 Sep 2024 16:00:07 +0800</pubDate><description><![CDATA[为了有效地捕获长实例序列中的综合上下文信息，本文引入了一种新的方法，MambaMIL，通过将Mamba框架集成到MIL中，如图1所示。通过继承Mamba的属性，MambaMIL使得每个实例可以通过压缩的隐藏状态与先前扫描的任何实例进行交互，这有利于对长序列进行有效建模，同时降低了计算复杂度。]]></description><category></category></item><item><title><![CDATA[MamMIL: Multiple Instance Learning for Whole Slide Images with State Space Models论文总结]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/141866164</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/141866164</guid><author>m0_60273616</author><pubDate>Tue, 03 Sep 2024 16:53:58 +0800</pubDate><description><![CDATA[本文将选择性结构化状态空间模型(即Mamba)与MIL相结合，提出了一种用于WSI分类的MamMIL框架，在保持线性复杂度的同时，实现了实例依赖关系的建模。]]></description><category></category></item><item><title><![CDATA[视觉Mamba综述——Visual Mamba: A Survey and New Outlooks论文总结]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/141726031</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/141726031</guid><author>m0_60273616</author><pubDate>Fri, 30 Aug 2024 20:03:59 +0800</pubDate><description><![CDATA[本文对视觉Mamba方法进行了全面的综述。我们首先对Mamba架构进行深入的概述，然后对具有代表性的可视化Mamba骨干网及其广泛的应用进行详细的研究。这些应用按照不同的模态进行了系统的分类，包括图像、视频、点云和多模态数据。最后，我们分析了可视化Mamba面临的挑战，并描绘了未来的发展方向，提供了有价值的展望，可能会影响这个动态演化领域的持续和未来发展。]]></description><category></category></item><item><title><![CDATA[Proposal-based Multiple Instance Learning for Weakly-supervised Temporal Action Localization论文整理]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/139006704</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/139006704</guid><author>m0_60273616</author><pubDate>Fri, 17 May 2024 15:52:59 +0800</pubDate><description><![CDATA[研究团队：中国科学技术大学。]]></description><category></category></item><item><title><![CDATA[配置mamba环境指南]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/138801777</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/138801777</guid><author>m0_60273616</author><pubDate>Mon, 13 May 2024 14:33:43 +0800</pubDate><description><![CDATA[mamba环境配置]]></description><category></category></item><item><title><![CDATA[Video Mamba Suite: State Space Model as a Versatile Alternative for Video Understanding论文整理]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/138502633</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/138502633</guid><author>m0_60273616</author><pubDate>Mon, 06 May 2024 17:45:04 +0800</pubDate><description><![CDATA[研究团队将 Mamba 模型的应用划分为四种不同的角色，提出了针对 14 种模型 / 模块的 Video Mamba Suite，在 12 项视频理解任务中对其进行了深入评估。]]></description><category></category></item><item><title><![CDATA[Real-ESRGAN：Training Real-World Blind Super-Resolution with Pure Synthetic Data论文整理]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/137082566</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/137082566</guid><author>m0_60273616</author><pubDate>Wed, 27 Mar 2024 16:55:56 +0800</pubDate><description><![CDATA[使用合成数据训练的Real - ESRGAN能够增强细节，同时去除大多数真实世界图像的恼人伪影。]]></description><category></category></item><item><title><![CDATA[Pseudo-Label : The Simple and Efficient Semi-Supervised Learning Method for Deep Neural Networks论文解析]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/136884387</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/136884387</guid><author>m0_60273616</author><pubDate>Wed, 20 Mar 2024 17:57:14 +0800</pubDate><description><![CDATA[本文提出了一种简单高效的深度神经网络半监督学习方法。网络以有监督的方式同时使用有标记和无标记数据进行训练。对于未标记数据，使用伪标签，只挑选网络输出最大的类，当作它们是真实的标签。从原理上讲，该方法可以结合几乎所有的神经网络模型和训练方法。]]></description><category></category></item><item><title><![CDATA[Semi-supervised Learning by Entropy Minimization论文整理]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/136517134</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/136517134</guid><author>m0_60273616</author><pubDate>Wed, 06 Mar 2024 20:34:37 +0800</pubDate><description><![CDATA[该文章提出了一种适用于任何概率分类器的估计原则，旨在使未标记数据在有益的情况下得到充分利用，同时对它们的贡献进行控制，从而为学习方案提供鲁棒性。]]></description><category></category></item><item><title><![CDATA[ASM-Loc: Action-aware Segment Modeling for WTAL论文笔记（CVPR2022)]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/133961439</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/133961439</guid><author>m0_60273616</author><pubDate>Sat, 21 Oct 2023 15:03:08 +0800</pubDate><description><![CDATA[本文提出了一个新的WTAL框架ASM - Loc，它能够在以往基于MIL的方法之外实现显式的动作感知片段建模。引入了与WTAL管道的三个阶段相对应的三个新颖的以段segment为中心的模块，缩小了弱监督和强监督设置之间的性能差距。该方法进一步引入多步训练策略来逐步细化动作建议直到定位性能达到饱和。ASM - Loc在两个WTAL基准上取得了SOTA。]]></description><category></category></item><item><title><![CDATA[NMS非极大抑制-简单总结]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/133684705</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/133684705</guid><author>m0_60273616</author><pubDate>Sun, 08 Oct 2023 16:27:59 +0800</pubDate><description><![CDATA[关于NMS一些总结]]></description><category></category></item><item><title><![CDATA[（模型细节）ACM-Net: Action Context Modeling Network]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/133101894</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/133101894</guid><author>m0_60273616</author><pubDate>Wed, 20 Sep 2023 21:24:45 +0800</pubDate><description><![CDATA[Action Context Modeling Network for Weakly-Supervised Temporal Action Localization]]></description><category></category></item><item><title><![CDATA[Adversarial Background-Aware Loss for Weakly-Supervised Temporal Activity Localization论文笔记]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/131301176</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/131301176</guid><author>m0_60273616</author><pubDate>Tue, 20 Jun 2023 10:23:05 +0800</pubDate><description><![CDATA[Adversarial Background-Aware Loss for Weakly-Supervised Temporal Activity Localization论文笔记ECCV2020]]></description><category></category></item><item><title><![CDATA[Action Segmentation动作分割概念区别]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/131301094</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/131301094</guid><author>m0_60273616</author><pubDate>Tue, 20 Jun 2023 10:16:14 +0800</pubDate><description><![CDATA[Action Segmentation动作分割和动作定位概念区别]]></description><category></category></item><item><title><![CDATA[实战手写数字识别]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/131037268</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/131037268</guid><author>m0_60273616</author><pubDate>Sun, 04 Jun 2023 21:23:25 +0800</pubDate><description><![CDATA[我在CPU上训练的，有点慢，差不多一个小时，建议用GPU来训。]]></description><category></category></item><item><title><![CDATA[模型搭建和参数优化]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/131024707</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/131024707</guid><author>m0_60273616</author><pubDate>Sat, 03 Jun 2023 19:26:35 +0800</pubDate><description><![CDATA[PyTtorch之torch.nn]]></description><category></category></item><item><title><![CDATA[自动梯度（torch.autograd和Variable）]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/131022705</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/131022705</guid><author>m0_60273616</author><pubDate>Sat, 03 Jun 2023 16:26:20 +0800</pubDate><description><![CDATA[torch.autograd和Variable。]]></description><category></category></item><item><title><![CDATA[Python中的Tensor-简单神经网络示例]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/130978184</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/130978184</guid><author>m0_60273616</author><pubDate>Wed, 31 May 2023 22:14:04 +0800</pubDate><description><![CDATA[搭建一个简单的神经网络]]></description><category></category></item><item><title><![CDATA[JupyterNotebook连接不上kernel]]></title><link>https://blog.youkuaiyun.com/m0_60273616/article/details/130924361</link><guid>https://blog.youkuaiyun.com/m0_60273616/article/details/130924361</guid><author>m0_60273616</author><pubDate>Mon, 29 May 2023 11:14:04 +0800</pubDate><description><![CDATA[1.首先查看一下kernel里python解释器的路径是否有问题。重新安装了tornado之后，成功启动kernel。我按照这个博主的方法修改了路径之后还是无法启动。2.按照具体的报错来找解决方法。然后仔细看了一下终端的报错。]]></description><category></category></item></channel></rss>