0

I want to create a Keras/TF model to recognise Magic the Gathering Cards. The cards could be in different languages. Does it make more sense to recognise the card name on top and the edition of the card, or just the artwork and edition? What is more efficient?

c111
  • 37
  • 6

2 Answers2

0

Or try HMS scan kit for card recognition. HMS Core Scan Kit supports most standard formats, providing long range detetction and improving the succcess rate of code scanning.

enter image description here

For more information, you can visit https://forums.developer.huawei.com/forumPortal/en/topicview?tid=0201236478929600089&fid=0101187876626530001

סטנלי גרונן
  • 2,917
  • 23
  • 46
  • 68
Joyce
  • 1
0

You may want to do both. Use a set of masks for the artwork, edition and extract image features and use text extraction and text features. Your pipeline would be something like:

image --> artwkmask1 | --> artwork_feat_xtract --> [...] : artwork_feat_vec
      |   artwkmask2 |
      |   ...        |
      |   artwkmaskn |
      |
      +-> edtnmask1  | --> edition_feat_xtract --> [...] : edition_feat_vec
      |   edtnmask2  |
      |   ...        |
      |   edtnmaskm  |
      |    
      +-> txtmask1   | --> txt_feat_xtract ------> [...] : text_feat_vec 
          txtmask2   |
          ...        |
          txtmaskk   |

You can then combine the three feature vectors into one wide vector and then use them for classification.

vvg
  • 1,010
  • 7
  • 25