1
$\begingroup$

I want to use the MNIST dataset to teach my neural network to recognise numbers. My problem is: The data I am working with is "european". What i mean by that is: the seven always has a dash (it often has a dash in MNIST too so that is more unlikely to cause a problem) and more importantly: the 1 is always written with this hook in Europe, whereas Americans tend to use a straight line (almost 100% of MNIST 1s are a straight line). So it seems to me that training on MNIST can cause problems. Have you ever ran into that problem? And if so, how did you solve it? I haven't found good solutions online.

$\endgroup$
0
$\begingroup$

for sure this is too late for you but as i used mnist some days ago and could not get good results for my own handwriting... maybe someone else is in search for a solution:

MNIST does represent american handwriting and will not achive good results in recognition of european writing.

| improve this answer | |
$\endgroup$
  • $\begingroup$ source might be helpful $\endgroup$ – oW_ Mar 30 at 23:24
  • $\begingroup$ Hey! Thanks for the comment. Do you have a solution yet? $\endgroup$ – tömmel Mar 31 at 19:06
  • $\begingroup$ after i realized that my net does not beat ~ 66% performance i plottet a random load of numbers from mnist_train and could clearly see that the handwriting all in all is different to the European style. so just a bit of google... as mnist was created by american high-school students and employees of the us census bureau it's very origin is america and american handwriting. wiki for mnist $\endgroup$ – bitdruid Apr 2 at 18:42
  • $\begingroup$ How do you do if you use the same architecture but mix in your personal handwriting with the training data, too, so the model has some exposure to it? $\endgroup$ – Dave Aug 30 at 14:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.