For example there is the MNIST database which is used to test ANN, however it's not so challenging, because some hierarchical systems of convolutional neural networks manages to get an error rate of 0.23 percent and the same with other OCR tests.

Therefore OCR datasets are 'no longer perceived as an exemplar of "artificial intelligence"'wiki.

Are there any similar equivalent image-recognition tests, especially these which have the most challenging tasks with dataset which are commonly used as benchmark tests to challenge the AI which are fairly reliable and they're possible to pass (e.g. by humans), but most AAN are struggling to achieve the lower error rate?

  • Re-posted from AI, then from stats in hope that's the right place for this question.
    – kenorb
    Aug 7 '16 at 16:29

I'm not sure if this is what you're looking for, but I've recently published the HASYv2 dataset. It contains 369 classes and 168233 recordings in total. The best classification result I could get was about 82% accuracy.

The data is extremely similar to MNIST. For more information, I recommend reading the paper. But if the paper doesn't answer questions, feel free to ask me :-)

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.