<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/">
    <channel>
        <title>NotionNext BLOG</title>
        <link>https://www.frankxx.link/</link>
        <description>这是一个由NotionNext生成的站点</description>
        <lastBuildDate>Tue, 10 Sep 2024 12:34:42 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>https://github.com/jpmonette/feed</generator>
        <language>zh-CN</language>
        <copyright>All rights reserved 2024, Fr4nk</copyright>
        <item>
            <title><![CDATA[What Can I See Man[2408]]]></title>
            <link>https://www.frankxx.link/article/240803</link>
            <guid>https://www.frankxx.link/article/240803</guid>
            <pubDate>Sat, 03 Aug 2024 00:00:00 GMT</pubDate>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-full-width notion-block-24844c294a8f4e2295e93fd6465397ad"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-45563558300549d3b72d7e1ead7060be" data-id="45563558300549d3b72d7e1ead7060be"><span><div id="45563558300549d3b72d7e1ead7060be" class="notion-header-anchor"></div><a class="notion-hash-link" href="#45563558300549d3b72d7e1ead7060be" title="海的开始"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">海的开始</span></span></h2><div class="notion-row notion-block-93b26839fe084242ba61be57ec8c305e"><div class="notion-column notion-block-1849e00ad72f4d8b881f4554e97d4acb" style="width:calc((100% - (1 * min(32px, 4vw))) * 0.5)"><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-9955afcaffc84c37a68835dc857d4df3"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://www.themoviedb.org/t/p/w600_and_h900_bestv2/w8YbsCAv6p3ruag2SJWqp77pcTL.jpg?t=9955afca-ffc8-4c37-a688-35dc857d4df3" alt="notion image" loading="lazy" decoding="async"/></div></figure></div><div class="notion-spacer"></div><div class="notion-column notion-block-cba7509b39f8433cae78de002cb03952" style="width:calc((100% - (1 * min(32px, 4vw))) * 0.5)"><div class="notion-text notion-block-16ce6bc2086540c58a87a9abda07b0e6">主演：<b><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.themoviedb.org/person/3190917-ren-meguro?language=zh">目黑莲</a></b><b>、</b><b><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.themoviedb.org/person/4294998-izutani-rana?language=zh">泉谷星奈</a></b><b>、</b><b><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.themoviedb.org/person/1039305?language=zh">有村架纯</a></b></div><div class="notion-text notion-block-ac9e67481b2b4bcc8994c1b5d65494cb">关键词：日剧、家庭、剧情</div><div class="notion-text notion-block-4d00063c6d4945a1924c9f658994e388">时间：2024年第二季度</div><blockquote class="notion-quote notion-block-b17ee2fa0e0e420c8ccd649eab15dc1b"><div>简介</div><div class="notion-text notion-block-45c7a696ac964028a328257c98db9f36">本剧讲述了月冈夏在大学时代与同级生南云水季交往着，但在求职的某天，水季突然提出分手。7年后，夏从大学的朋友那得知水季去世的消息。被分手之后从未见面，对那个事实没有实感的夏去了葬礼，在那里遇见了名为海的小女孩，得知她是水季的女儿，夏难掩惊讶。然后，水季的母亲告知他就是海的父亲。水季在夏不知道的地方生了孩子，什么也不说就养育了她，夏想着水季和海度过的7年岁月…</div></blockquote><div class="notion-text notion-block-d67d908444794f7e90b88ead14464207">@Frank-GPT:故事起因略显狗血，但是卡司属实强大，前几集保持着惯有的细腻感，画面也比较舒服，进度不快但是给人一种润物细无声的感觉。</div></div><div class="notion-spacer"></div></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-76a2412278e6433ab30f438fb8535a01" data-id="76a2412278e6433ab30f438fb8535a01"><span><div id="76a2412278e6433ab30f438fb8535a01" class="notion-header-anchor"></div><a class="notion-hash-link" href="#76a2412278e6433ab30f438fb8535a01" title="黑色止血钳2"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">黑色止血钳2</span></span></h2><div class="notion-row notion-block-4176982ce04a4062b232396fbe33f45f"><div class="notion-column notion-block-3713d575585c477795888616b1a68a56" style="width:calc((100% - (1 * min(32px, 4vw))) * 0.5)"><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-b103922da1fd4278af99bd2e4931dd36"><div style="position:relative;display:flex;justify-content:center;align-self:start;width:100%;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://www.themoviedb.org/t/p/w600_and_h900_bestv2/q4es4RVBEb3hsx7KmBW9NgZZf8e.jpg?t=b103922d-a1fd-4278-af99-bd2e4931dd36" alt="notion image" loading="lazy" decoding="async"/></div></figure></div><div class="notion-spacer"></div><div class="notion-column notion-block-21ec2953e5b44454b08726a6832e47e1" style="width:calc((100% - (1 * min(32px, 4vw))) * 0.5)"><div class="notion-text notion-block-1f2db1b6ecd540cb9004a0fed551e53b">主演：<b><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.themoviedb.org/person/33515?language=zh">二宫和也</a></b><b>、</b><b><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.themoviedb.org/person/1384646?language=zh">竹内凉真</a></b></div><div class="notion-text notion-block-f5c941b110824a1e92949c25b1a5a0bb">关键词：日剧、悬疑、剧情</div><div class="notion-text notion-block-cdf84e2fe81e45c0a2e6e23be90cebdd">时间：2024年第二季度</div><blockquote class="notion-quote notion-block-fb47ac46b6fa407ab20e8ef0c4d8fb3a"><div>简介</div><div class="notion-text notion-block-894491a54b9a42b58a798a045a5878c5">本剧改编自海堂尊的《火焰手术刀1990》和《樱色心脏中心1991》，讲述了玩弄人与金钱的恶魔、世界级天才外科医生·天城雪彦（二宫和也 饰演），在澳大利亚黄金海岸心脏中心工作多年，手术技术堪称天才，是心脏冠状动脉搭桥术的世界级大师。另一方面，要接受天城的手术，只能在二选一中碰运气获胜，作为赌注，要求对方财产的一半，因此在医生之间，他被称为“恶魔”。他和6年前在东城大学医学部附属医院的渡海一模一样地站着…</div></blockquote><div class="notion-text notion-block-44eb2df101af4ed9ae0f58ed0be94acb">@Frank-GPT:还是当年内个味，不过挖的坑感觉越来越多了，第一季就有好多坑没填，颜值和演技都还行，起码不会有看国产小鲜肉的不适感。</div></div><div class="notion-spacer"></div></div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[家庭代理终极解决方案]]></title>
            <link>https://www.frankxx.link/article/240219</link>
            <guid>https://www.frankxx.link/article/240219</guid>
            <pubDate>Mon, 19 Feb 2024 00:00:00 GMT</pubDate>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-e9f8cb2142e6422a918a5740c05687c0"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div class="notion-text notion-block-51a07e1b1a154e3c96bd662729ec8dd2">拖拖拉拉几个月，准备写X86旁路由+OpenClash+SmartDNS的，结果用了几个月修了好几次，稳定性属实堪忧，而且人在外面的时候，家里网络炸了极其难修，因此不再推荐这种过于折腾的方案，毕竟这玩意儿稳定性才是首位，经过几个月的测试，<b>Mac Mini(Apple TV)+Surge</b>稳定性和功耗都非常满足我的需求。</div><div class="notion-blank notion-block-0eca91d3e9d34940be7c5704fd767681"> </div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[图像匹配挑战赛IMC 2024]]></title>
            <link>https://www.frankxx.link/article/240627</link>
            <guid>https://www.frankxx.link/article/240627</guid>
            <pubDate>Thu, 27 Jun 2024 00:00:00 GMT</pubDate>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-9c70b480df82474b96ae5ae52695a573"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-bf9d9b796a644ae4972665fafb176cc2" data-id="bf9d9b796a644ae4972665fafb176cc2"><span><div id="bf9d9b796a644ae4972665fafb176cc2" class="notion-header-anchor"></div><a class="notion-hash-link" href="#bf9d9b796a644ae4972665fafb176cc2" title="图像匹配挑战赛"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>图像匹配挑战赛</b></span></span></h2><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-72ae6b8051b249f8b5a07318fa64c35c" href="https://www.kaggle.com/competitions/image-matching-challenge-2024/overview"><div><div class="notion-bookmark-title">Image Matching Challenge 2024 - Hexathlon</div><div class="notion-bookmark-description">Reconstruct 3D scenes from 2D images over six different domains</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-icon"><img src="https://www.kaggle.com/static/images/favicon.ico?t=72ae6b80-51b2-49f8-b5a0-7318fa64c35c" alt="Image Matching Challenge 2024 - Hexathlon" loading="lazy" decoding="async"/></div><div class="notion-bookmark-link-text">https://www.kaggle.com/competitions/image-matching-challenge-2024/overview</div></div></div></a></div><div class="notion-text notion-block-011c4cceb0f44e24919ca98a4dc24697">该项比赛旨在利用计算机视觉技术，匹配不同视角下的同一个场景，从而恢复每张图像拍摄时的相机内外参矩阵。今年的比赛中主要对象有室外场景和透明弱纹理物体，这里主要关注室外场景的匹配策略，来总结一下前几名的思路。</div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-58fe806702a6428298d3a472a668b324" data-id="58fe806702a6428298d3a472a668b324"><span><div id="58fe806702a6428298d3a472a668b324" class="notion-header-anchor"></div><a class="notion-hash-link" href="#58fe806702a6428298d3a472a668b324" title="第一名：High Image Resolution ALIKED/LightGlue + Transparent Trick "><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">第一名：<b><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/510084">High Image Resolution ALIKED/LightGlue + Transparent Trick </a></b></span></span></h2><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-d7d43e1c91b841e4844f2e3686a3ccf1"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column;height:100%"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Feb510b60-7bdd-4d2d-8e33-90b9250fafd4%2Ff44d9654-cd3e-4e64-aa60-63958102fd14%2FUntitled.png?table=block&amp;id=d7d43e1c-91b8-41e4-844f-2e3686a3ccf1&amp;t=d7d43e1c-91b8-41e4-844f-2e3686a3ccf1" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-0811fb9333fd43808e6dc0b0769cf87e">总的来说，先通过一次稀疏的匹配，将部分旋转过的图像摆正，再将图像使用ALIKED和LightGlue进行匹配，并使用DBSCAN获取密集的匹配区域，并再一次进行密集匹配，将匹配点输入COLMAP进行位姿计算。第一名完全没有使用Dense-base的方法，他们认为这种方法获得的匹配点通常只在两张图之间，不具备传递性，会导致模型计算错误。</div><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-7cf2869f75224c66afe48bc4778f63ce" data-id="7cf2869f75224c66afe48bc4778f63ce"><span><div id="7cf2869f75224c66afe48bc4778f63ce" class="notion-header-anchor"></div><a class="notion-hash-link" href="#7cf2869f75224c66afe48bc4778f63ce" title="亮点："><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">亮点：</span></span></h3><ul class="notion-list notion-list-disc notion-block-416c0e449e9f4e6695c1d5f867e7afcc"><li>根据图像稀疏匹配确定待匹配对</li></ul><ul class="notion-list notion-list-disc notion-block-293a92ff864249449e8310419c818344"><li>微调过的ALIKED和LightGlue</li></ul><ul class="notion-list notion-list-disc notion-block-163d86b12a5f47b4952173805a8019bd"><li>一些运行效率上的优化，如缓存、分布式训练</li></ul><ul class="notion-list notion-list-disc notion-block-ed92eeea152b4d4ab9d7f34138392e14"><li>多尺度下的特征点匹配，在1280分辨率下匹配一次后，通过DBSCAN聚类获取<b>匹配密集区域</b>，剪裁图像后，在2048分辨率下再进行一次匹配，以获取更精细的匹配</li></ul><div class="notion-blank notion-block-cff002ea54a8494b8d979de14b816479"> </div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-e4a282eb840b482cab1f27d575313da1" data-id="e4a282eb840b482cab1f27d575313da1"><span><div id="e4a282eb840b482cab1f27d575313da1" class="notion-header-anchor"></div><a class="notion-hash-link" href="#e4a282eb840b482cab1f27d575313da1" title="第二名：MST-Aided SfM &amp; Transparent Scene Solution"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">第二名：<b><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/510499">MST-Aided SfM &amp; Transparent Scene Solution</a></b></span></span></h2><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-850efc0ca24f477ca1972ed0896236b0"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column;height:100%"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Feb510b60-7bdd-4d2d-8e33-90b9250fafd4%2Ffbd040f4-bee5-4020-af48-c1786cf28e40%2FUntitled.png?table=block&amp;id=850efc0c-a24f-477c-a197-2ed0896236b0&amp;t=850efc0c-a24f-477c-a197-2ed0896236b0" alt="notion image" loading="lazy" decoding="async"/></div></figure><ol start="1" class="notion-list notion-list-numbered notion-block-6a0897b0face43f1bcbf7b655119a078"><li>对图像进行了旋转检测和透明度检测，为了后面分别处理透明图像和非透明图像（大部分方法都将这两种图像分开处理）；</li></ol><ol start="2" class="notion-list notion-list-numbered notion-block-156ac2294318407fa2766c9c6cca3644"><li>对所有图像进行了全局特征提取，以生成更可靠的图像匹配对；</li></ol><ol start="3" class="notion-list notion-list-numbered notion-block-1d1e27862ff641b6be285124828f4b99"><li>根据图像匹配对，使用了<b>(Dedodev2+Dual SoftMax)、(DISK+LightGlue)、 (SIFT+NN)</b>三种方法获取匹配点；</li></ol><ol start="4" class="notion-list notion-list-numbered notion-block-50d873aea7ae41b7a2d110fafb26357f"><li>使用了一种基于最小生成树的策略获取各图像全局最优的相邻图，用于第一个粗略的SFM；</li></ol><ol start="5" class="notion-list notion-list-numbered notion-block-8132111ab5be4d0d8da06047e6d73308"><li>同时使用全局最优的相邻图和全连接图进行SFM，获得两个SFM结果，用粗略的SFM结果来过滤精细SFM结果中错误的部分，再使用PixSFM和HLoc中重定位模块来处理缺失的相机。</li></ol><ol start="6" class="notion-list notion-list-numbered notion-block-a7394a4ca2424e2aa6a9d524adcb20e8"><li>透明图像处理不做赘述。</li></ol><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-2e489cd9ddd141618e316780c00877fe" data-id="2e489cd9ddd141618e316780c00877fe"><span><div id="2e489cd9ddd141618e316780c00877fe" class="notion-header-anchor"></div><a class="notion-hash-link" href="#2e489cd9ddd141618e316780c00877fe" title="亮点："><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">亮点：</span></span></h3><ul class="notion-list notion-list-disc notion-block-d67aab28279b42529022952cf556f4fe"><li>主要在使用MST获取全局最优图以及使用粗略的SFM结果来过滤精细SFM结果中错误的部分</li></ul><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-68876163356f47ac872dbc153fcbe04e" data-id="68876163356f47ac872dbc153fcbe04e"><span><div id="68876163356f47ac872dbc153fcbe04e" class="notion-header-anchor"></div><a class="notion-hash-link" href="#68876163356f47ac872dbc153fcbe04e" title="第三名：Solution: VGGSfM"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">第三名：<b><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/510338">Solution: VGGSfM</a></b></span></span></h2><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1c372577f2914bc69174389122d4f44e"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column;height:100%"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Feb510b60-7bdd-4d2d-8e33-90b9250fafd4%2Fb2f40b7e-22af-4ea7-9f57-2c698608f504%2FUntitled.png?table=block&amp;id=1c372577-f291-4bc6-9174-389122d4f44e&amp;t=1c372577-f291-4bc6-9174-389122d4f44e" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-blank notion-block-79f6478d2592405dba9f5bd2423aaa92"> </div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ubuntu安装Colmap GPU]]></title>
            <link>https://www.frankxx.link/article/240508</link>
            <guid>https://www.frankxx.link/article/240508</guid>
            <pubDate>Wed, 08 May 2024 00:00:00 GMT</pubDate>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-f08d7b816add4bb4afad86dd59e81371"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div></div><div class="notion-text notion-block-50a969e94b304a7286ee1e7637a2b59c">如果没装conda，那么恭喜你，直接<a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://colmap.github.io/install.html">官网</a>教程安装，基本不会报错。</div><div class="notion-text notion-block-2b905f7f74ca41eb98782e1d1e1b61cd">如果装了conda，首先conda deactivate退出环境，并将conda安装文件夹改个名字，然后再进行<a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://colmap.github.io/install.html">官网</a>安装，通常来说会报错依赖问题，一个一个递推式安装，一般来说就能解决问题。</div><div class="notion-blank notion-block-8c2f249f566c4961b924ee194a691dcc"> </div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[高斯模型摆正 Python代码]]></title>
            <link>https://www.frankxx.link/article/240619</link>
            <guid>https://www.frankxx.link/article/240619</guid>
            <pubDate>Wed, 19 Jun 2024 00:00:00 GMT</pubDate>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-fd8b563d22f245d897409c147937980e"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-b85a2f9daf044847bfbda067981d9e7d" data-id="b85a2f9daf044847bfbda067981d9e7d"><span><div id="b85a2f9daf044847bfbda067981d9e7d" class="notion-header-anchor"></div><a class="notion-hash-link" href="#b85a2f9daf044847bfbda067981d9e7d" title="引言"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">引言</span></span></h2><div class="notion-text notion-block-eb963ec488754581b08956b7a626499d">通常基于colmap生成的模型，其坐标轴都是随机的，导致后面的模型会出现倾斜甚至是颠倒的情况，如果用SIBR查看不会有问题，但是导入到各种高斯编辑器中时问题会比较严重，因此，需要对模型进行摆正，这里采用平面拟合来确定整个模型的Z轴，计算模型质心作为模型的中心，并参考<a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://github.com/alvinliu0/HumanGaussian/blob/main/gs_renderer.py">alvinliu0</a>使用python读取编辑PLY文件，实现高斯模型摆正。</div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-06d4fa9fc1fd47f7923eb60e0b833628" data-id="06d4fa9fc1fd47f7923eb60e0b833628"><span><div id="06d4fa9fc1fd47f7923eb60e0b833628" class="notion-header-anchor"></div><a class="notion-hash-link" href="#06d4fa9fc1fd47f7923eb60e0b833628" title="平面拟合和质心计算"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">平面拟合和质心计算</span></span></h2><div class="notion-text notion-block-3b070a5c7b404d37948d7cec6d9c6530">使用open3d读取PLY点云，调用segment_plane计算场景法向量，并调用get_center计算场景质心。</div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-a3ec8c89d0a54b6d9f89218861b62502" data-id="a3ec8c89d0a54b6d9f89218861b62502"><span><div id="a3ec8c89d0a54b6d9f89218861b62502" class="notion-header-anchor"></div><a class="notion-hash-link" href="#a3ec8c89d0a54b6d9f89218861b62502" title="读取编辑PLY"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">读取编辑PLY</span></span></h2><div class="notion-text notion-block-f8635d2643e84775ab620d9ad946aaff">高斯模型的PLY文件格式比较特殊，无法直接调库编辑读写，主要是不能只旋转点云，还要处理球谐函数的对称轴向量（rotation这个值），参考源码，用Python和C++都能实现，这里给出Python的代码。</div><div class="notion-text notion-block-bd882b24f32648c280bdd3d105458bce">这里值得注意的是，rotation是一个四元数，其顺序是wxyz，计算旋转矩阵时不要弄错。</div><div class="notion-blank notion-block-e0f1842a695c4410a6274833de08584f"> </div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[3DGS训练类blender数据集]]></title>
            <link>https://www.frankxx.link/article/240523</link>
            <guid>https://www.frankxx.link/article/240523</guid>
            <pubDate>Thu, 23 May 2024 00:00:00 GMT</pubDate>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-456ce441e42b425aa39376e3f9641258"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div></div><div class="notion-text notion-block-08c798a800dd419d810dccd64148ec64"><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://github.com/graphdeco-inria/gaussian-splatting">3DGS</a>本身支持对Blender数据集的训练，其主要数据格式为：</div><h4 class="notion-h notion-h3 notion-h-indent-0 notion-block-3ab3ba63a49b47acb9786d640d6cad0c" data-id="3ab3ba63a49b47acb9786d640d6cad0c"><span><div id="3ab3ba63a49b47acb9786d640d6cad0c" class="notion-header-anchor"></div><a class="notion-hash-link" href="#3ab3ba63a49b47acb9786d640d6cad0c" title="数据准备"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">数据准备</span></span></h4><div class="notion-text notion-block-bf34913a33754050a7a9b4b3bed03833">通常我们自己采集的数据集来自激光扫描装置，提供了las格式的点云文件、json格式的相机位姿以及图片。</div><div class="notion-text notion-block-533f58dba10c4c0193b3825b012f4b57">首先处理las格式的点云文件，需要将其转化为二进制的ply文件。</div><div class="notion-text notion-block-410b6780d0f74dbe9d82cbad9f026c51">然后查看transformers_train.json文件，主要是针对性修改3dgs代码文件下/sence/dataset_readers.py这个文件。</div><div class="notion-text notion-block-a9006a005fbe4d319efc9158f4bc4dae">注释掉了Test Transforms相关的部分，不会进行eval，因此不需要test。修改了extension，主要是看自己的图像输入格式。</div><div class="notion-text notion-block-95ddee8061fc443a87e42caae6bce269">这里通常根据自己的json文件内容更改，几个主要需要注意的点是：</div><ul class="notion-list notion-list-disc notion-block-b598fd9db8ac4507ad3f74d30c38769d"><li><code class="notion-inline-code">fovx = contents[&quot;camera_angle_x&quot;]</code> </li></ul><ul class="notion-list notion-list-disc notion-block-d57c89ad5ece49368a97a058c148a084"><li><code class="notion-inline-code">cam_name = os.path.join(path,&quot;images&quot;, frame[&quot;file_path&quot;])</code></li></ul><div class="notion-text notion-block-0d48211faa164e6b815efc6b74a5deeb">
 </div><div class="notion-text notion-block-54b305d9b21b4830b24e971b71c1ec5e">这部分是ply文件相关的，通常自采集数据集不会有点云法向量normals，因此我们根据其colmap部分的代码，也将其设置为0。</div><div class="notion-text notion-block-34fff6b013a344fcb3fc12cce82b1caa">一般这样改完，代码就能跑通了，激光点云结合GPS高精度相机位姿，能够解决colmap无法获取相机位姿的问题。</div><div class="notion-blank notion-block-9f5c7409fe1b42bb8262048b4c492e1b"> </div></main></div>]]></content:encoded>
        </item>
    </channel>
</rss>