<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>3d on Alok Regmi</title>
		<link>https://alokregmi.com/tags/3d/</link>
		<description>Recent content in 3d on Alok Regmi</description>
		<generator>Hugo</generator>
		<language>en-us</language>
		
		
		
		
			<lastBuildDate>Sun, 19 Apr 2020 16:00:00 +0545</lastBuildDate>
		
			<atom:link href="https://alokregmi.com/tags/3d/index.xml" rel="self" type="application/rss+xml" />
			<item>
				<title>3ddatarepresentation</title>
				<link>https://alokregmi.com/braindump/2020/04/3ddatarepresentation/</link>
				<pubDate>Sun, 19 Apr 2020 16:00:00 +0545</pubDate>
				<guid>https://alokregmi.com/braindump/2020/04/3ddatarepresentation/</guid>
				<description>&lt;p&gt;3D data representations:&lt;/p&gt;&#xA;&lt;ol&gt;&#xA;&lt;li&gt;&#xA;&lt;p&gt;Point Cloud(PointNet):&#xA;&amp;ndash; More concretely, given points in a point cloud, the network needs to learn&#xA;unique features that are invariant to the permutations of this input data,&#xA;since the ordering of points fed into a network doesn&amp;rsquo;t impact the underlying geometry.&#xA;&amp;ndash; network should be robust to transformations of the point cloud —&#xA;rotations, translations and scaling should not impact prediction.&#xA;&amp;ndash; To ensure invariance across input ordering, the key insight behind PointNet is using&#xA;a simple symmetric function that produces a consistent output for any ordering of the input&#xA;(examples in this class of functions include addition and multiplication)&#xA;&amp;ndash; So, we need to have a symmetric function &amp;lsquo;f&amp;rsquo; that converts input points into k-dim vector&#xA;f = gamma * g(h(x1),&amp;hellip;. h(xn))&#xA;where h is hidden MLP conversion of input points&#xA;g is symmetric function too&#xA;The max-pooling operations serves as the symmetric function &amp;lsquo;g&amp;rsquo;,&#xA;aggregating learned features into a global descriptor for the point cloud.&#xA;gamma is also a MLP which predicts the output&lt;/p&gt;&#xA;&lt;p&gt;&amp;ndash;  PointNet employs a mini-network, called T-Net for geometric invariance&#xA;&amp;ndash; overall PointNet architecture integrates the vanilla approach and T-Net&#xA;with multiple MLP layers that create feature representations for the point clouds.&#xA;&amp;ndash; Also enables semantic segmentation of scenes and objects&#xA;&amp;ndash; To accomplish this, the architecture combines the global feature vector from the max pooling&#xA;symmetric function with per-point features learned after the input data is passed through a few MLPs.&#xA;By concatenating these two vectors, each point is aware of both its global semantics and its local features,&#xA;enabling the network to learn additional, more meaningful features that help with segmentation.&lt;/p&gt;&#xA;&lt;/li&gt;&#xA;&lt;/ol&gt;</description>
			</item>
	</channel>
</rss>
