Photo OCR

Problem Description and Pipeline

  • OCR Problem

    • Recognising text from the photo

    OCR Problem.png

    • OCR Pipeline

      OCR Pipeline.png OCR Pipeline 1.png

Sliding Windows

  • Text and Pedestrian Detection

    Text and Pedestrian Detection.png

  • Pedestrian Detection

    Pedestrian Detection.png

  • Sliding Window Detection

    Sliding Window Detection.png

  • Text Detection

    Text Detection.png Text Detection 1.png

Getting Lots of Data and Artificial Data

  • Artificial Data Synthesis for Photo OCR

    Artificial Data Synthesis.png

  • Introducing Distortions

    Introducing Distortions.png

    • Introducing Meaningful Distortions

      Introducing Distortions 1.png

  • Speech Synthesis

    Speech Synthesis.png

  • More Discussions on getting more data

    More Discussions.png

Ceiling Analysis: What Part of the Pipeline to Work on Next

  • Ceiling Analysis

    • Check which component of the pipeline affects the accuracy most

    • Allocate your resources to that component

    Ceiling Analysis.png

  • Example

    • Ceiling Analysis on Face Recognition

    Example.png

Lecture Presentations