What about not embossed card's?
dimiroll opened this issue · 5 comments
I have tested a lot of card's, but i surprised why card.io doesn't support not embossed (flat) cards?
I would like to do training of CNN myself, but i don't have whole training set. The training one of existed net's with small learning rate is not good idea, i think.
So, can you support my begining?
@dimiroll thanks for your interest in improving card.io!
I have tested a lot of card's, but i surprised why card.io doesn't support not embossed (flat) cards?
Due to their many differences (in font-size, typefaces, digit color, lack of shadows, etc.), non-embossed cards require different digit-categorizing models than embossed cards. There is a great deal of variety in non-embossed cards, so supporting them is a lot of work. And non-embossed cards are still rare relative to embossed cards. As a result, so far, it has not been enough of a priority to justify the work that supporting them would require.
I would like to do training of CNN myself, but i don't have whole training set.
Unfortunately, our training set contains lots of personal credit card images, which we cannot release.
The training one of existed net's with small learning rate is not good idea, i think.
We are always open to suggestions. However, we have already experimented with a very wide variety of hyper parameters and model types during training.
Hi David,
Can card io able to scan cards given in the below link.
https://usa.visa.com/dam/VCOM/download/merchants/New_VBM_Acq_Merchant_62714_v5.pdf
@shaileshnamjoshi card.io can handle most "traditional" cards (although the presence of the hologram behind some of the numbers, as shown there, might be problematic). However, any non-traditional layouts, such as the "Quick Read Card Design" shown there, will not work with card.io.
@dgoldman-ebay how would i train with my own images?
@ericlewis -- I've responded in your other issue.