<?xml version="1.0" encoding="UTF-8"?>
<?xml-stylesheet href="/atom.xsl" type="text/xsl"?>
<feed xmlns="http://www.w3.org/2005/Atom" xmlns:fh="http://purl.org/syndication/history/1.0">
  <title>Posts tagged: computer-vision Archive</title>
  <id>https://pype.dev/tags/computer-vision/archive/atom.xml</id>
  <updated>2025-05-31T08:18:55Z</updated>
  <subtitle>All posts with the tag &#34;computer-vision&#34;</subtitle>
  <link href="https://pype.dev/tags/computer-vision/" rel="alternate" type="text/html"></link>
  <link href="https://pype.dev/tags/computer-vision/archive/atom.xml" rel="self" type="application/atom+xml"></link>
  <link href="https://pype.dev/tags/computer-vision/atom.xml" rel="current" type="application/atom+xml"></link>
  <author>
    <name>Nic Payne</name>
  </author>
  <generator uri="https://github.com/WaylonWalker/markata-go">markata-go</generator>
  <fh:complete></fh:complete>
  <entry>
    <title>homelab-computer-vision-pipelines</title>
    <id>https://pype.dev/homelab-computer-vision-pipelines/</id>
    <updated>2025-05-31T08:18:55Z</updated>
    <published>2025-05-31T08:18:55Z</published>
    <link href="https://pype.dev/homelab-computer-vision-pipelines/" rel="alternate" type="text/html"></link>
    <summary type="text">Done in 11 seconds! Subtitle and audio files are in the outputs folder.</summary>
    <content type="html">&lt;p&gt;Done in 11 seconds! Subtitle and audio files are in the outputs folder.&lt;/p&gt;&#xA;&lt;p&gt;I wanted to talk through an idea I have for some computer vision pipelines at home.&#xA;I heard of a tool called Roboflow, which is like a no-code computer vision pipeline training&#xA;and inference tool.&#xA;It seems very cool.&#xA;I also use Frigate at home, and Frigate supports lots of detection on the NVR side.&#xA;And so deploying a model in Frigate or using Roboflow, depending on what &lt;a href=&#34;/kind/&#34; class=&#34;glossary-term&#34; title=&#34;kind - kubernetes in docker&#34; data-hover-title=&#34;kind&#34;&gt;kind&lt;/a&gt; of customization&#xA;you would need, you can have these detections potentially be as accurate as you want, at&#xA;least with Roboflow.&#xA;The thing is, I don&amp;rsquo;t know about Frigate, but my ideas are if I have a model at home&#xA;that knows me and my wife and my children by name, then when monitored areas issue detections,&#xA;for example, say I&amp;rsquo;m in the backyard and the temperature is above 80 degrees, have a home&#xA;assistant kick on a sprinkler system or something like that.&#xA;And if my wife is back there and it&amp;rsquo;s 80 degrees, don&amp;rsquo;t do that because she wouldn&amp;rsquo;t want that&#xA;sprinkler on or whatever.&#xA;I think that something like that would be very feasible once you have the model.&#xA;So you would need a way to train the model, which would require gathering data and labeling&#xA;it.&#xA;Then you could train a model in lots of ways, but you need to have the deployment in mind.&#xA;If the deployment was Frigate, you would need to be able to compile that model into some&#xA;TensorFlow Lite thing that can run on a TPU.&#xA;That sounds very possible.&#xA;Otherwise, the Roboflow would probably be the platform for training, and then they make&#xA;the deployment kind of a one-click thing.&#xA;From there, I guess either way, Frigate is running the inference, and so the home assistant&#xA;integration would be about the same.&#xA;Frigate sends notifications over MQTT, and based on the Frigate blueprint that I&amp;rsquo;m using&#xA;for home assistant anyway, based on the object and whatever other conditions you want to&#xA;set in an automation, you do whatever you want to do in home assistant.&#xA;That sounds neat.&lt;/p&gt;&#xA;</content>
    <author>
      <name>Nic Payne</name>
      <uri>https://pype.dev</uri>
    </author>
  </entry>
</feed>