Skip to content
Labeled images of the Where's Waldo puzzle for use in classification and image recognition problems.
Branch: master
Clone or download
vc1492a Merge pull request #1 from ubalklen/patch-1
Update image_processing.py. Thanks for noticing that and submitting the PR!
Latest commit 326707c Feb 6, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
128-bw intial commit Jul 20, 2016
128-gray intial commit Jul 20, 2016
128 intial commit Jul 20, 2016
256-bw intial commit Jul 20, 2016
256-gray intial commit Jul 20, 2016
256 intial commit Jul 20, 2016
64-bw intial commit Jul 20, 2016
64-gray intial commit Jul 20, 2016
64 intial commit Jul 20, 2016
original-images intial commit Jul 20, 2016
.gitignore intial commit Jul 20, 2016
LICENSE.txt readme update, ObdL license) Aug 8, 2017
image_processing.py Update image_processing.py Feb 6, 2019
readme.md added example Jul 3, 2018

readme.md

Hey Waldo

License

This repository contains labeled images of the Where's Waldo puzzle for use in classification and image recognition problems, which I painstakingly hand labeled. Along with the labeled images are 19 original images, as well as the Python script used to create the smaller labeled images. Color, grayscale, and black or white versions of the labeled images are included as part of the data set.

Image Formats

  • 256 x 256 pixels (317 images)
  • 128 x 128 pixels (1344 images)
  • 64 x 64 pixels (5376 images)

Discussion

I decided to label images as containing Waldo regardless of where Waldo is located in the image. As such, there are images that are labeled as Waldo that are not strictly part of the puzzle. Waldo look-alikes, such as those with different hair styles or glasses, were also labeled as Waldo. This was just due to the way in which I decided to approach the problem, so feel free to relabel the images according to your needs and interpretation.

Classification with these images is a difficult problem due to the Waldo's variable size (scaling issue), repeating patterns (red/white stripes present on other objects), occlusion (Waldo is often blocked by other people or objects), and the nature of the data, which is unbalanced (most images do not contain Waldo). It's a tough but fun application and I invite others to give it a try and share their results.

Fun Stuff

You can’t perform that action at this time.