What dataset were these networks trained on?

[livnev]

What dataset were these networks trained on? There is no mention in the readme/scripts what dataset you used to achieve the results that you are demonstrating. I am curious: is there a publicly available handwritten kanji dataset?

[KyotoSunshine]

Sorry, the dataset is not publicly available. But you can see some samples in the Dataset-Augmentation folder. So if you have images that look like that, the model should work on them.

[huangwaylon]

Is there a public list of the kanji characters that are contained within the dataset used to train the model?

[KyotoSunshine]

I am afraid there is not, but there were about 500 characters if I remember correctly. So rather than using the pretrained model, I would suggest you just train this model on your own data.

[huangwaylon]

I see, thank you for the information!

[r3m4k3]

@huangwaylon You can use dataset from this site: http://etlcdb.db.aist.go.jp/