Skip to content

roboflow/neuralhash-collisions

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

neuralhash-collisions

A catalog of naturally occurring images whose Apple NeuralHash is identical.

See this NeuralHash collision blog post with full details and background.

Contributing

PRs are welcome and encouraged! You can find the code to run NeuralHash on images in this repo. Please do not submit artificially generated adversarial examples. This repo is meant to document natural collisions found in the wild.

Please follow the example format (add a folder in collisions named with the colliding hash and containing the example images with a README cataloging information about their provenance). Ideally provide evidence that these images existed prior to the public divulgence of the NeuralHash weights to prove that they are not artificial.

Suggested Methodology

There are many open source image datasets which contain a wide variety of images.

If you generate a NeuralHash for each one, store the image path / hash pairs in a text file, sort them by hash, and compare adjacent lines' hashes, you can quickly check n^2 image pairs (where n is the size of the dataset).

There are some example scripts you may adapt in the utils directory.

Hash Lists

You may also compare your hashes against the ImageNet hashes using the txt file archived in the imagenet directory.

If you check a large, public corpus please PR the hash list so others can easily compare them for collisions across sources.

Releases

No releases published

Packages

No packages published