Skip to content
This repository has been archived by the owner on Feb 28, 2018. It is now read-only.

Drawing - Damage and Distortion #8

Open
makeyourownneuralnetwork opened this issue Apr 12, 2017 · 1 comment
Open

Drawing - Damage and Distortion #8

makeyourownneuralnetwork opened this issue Apr 12, 2017 · 1 comment

Comments

@makeyourownneuralnetwork

One of the key nice things about neural networks is that they are resilient to some damage.

That means either damage to the network itself (e.g. broken links), or .. more interestingly .. damage to the input image.

It might be interesting for users to draw a digit, then click "damage" and "distort" buttons which would blow holes in the image, or stretch/wobble/twist the image a bit.

This would allow users to see for themselves that you can damage an input image quite a bit before overall performance degrades. I did some experiments myself at http://makeyourownneuralnetwork.blogspot.co.uk/2016/03/your-own-handwriting-real-test.html

my_own_images

@shiffman
Copy link
Owner

This is a great suggestion, thank you!

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants