<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>OCR on Blogs by Anil</title>
    <link>https://paudelanil9.com.np/categories/ocr/</link>
    <description>Recent content in OCR on Blogs by Anil</description>
    <generator>Hugo -- 0.138.0</generator>
    <language>en</language>
    <lastBuildDate>Sun, 16 Aug 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://paudelanil9.com.np/categories/ocr/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Building TrOCR from scratch: wiring a vision encoder to a text decoder</title>
      <link>https://paudelanil9.com.np/posts/trocrfromscratch/</link>
      <pubDate>Sun, 16 Aug 2026 00:00:00 +0000</pubDate>
      <guid>https://paudelanil9.com.np/posts/trocrfromscratch/</guid>
      <description>&lt;h1 id=&#34;building-trocr-from-scratch-what-wiring-a-vision-encoder-to-a-text-decoder-actually-taught-me&#34;&gt;Building TrOCR from scratch: what wiring a vision encoder to a text decoder actually taught me&lt;/h1&gt;
&lt;p&gt;I spent a weekend answering one question end to end: &lt;em&gt;how does a model look at a picture of text and type out the characters?&lt;/em&gt; I already knew the NLP Transformer — Q/K/V, multi-head attention, positional encodings — so I didn&amp;rsquo;t want a tutorial that treats the Vision Transformer as a magic box. I wanted to build the whole thing from scratch in PyTorch: the ViT encoder, the text decoder, and the cross-attention that joins them, aiming eventually at handwritten Devanagari OCR.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
