search

FangShancheng / ABINet

Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition
Other
420 stars 72 forks source link

Evaluation dataset contain corrupted image #94

Open Yosu26 opened 1 year ago

Yosu26 commented 1 year ago

When I try to pre-train the vision model and ABINet I found that there are 3 evaluation datasets that contain corrupted image. Any fix for this ? Or do I just remove it from the configs ?

Below are the evaluation datasets that contain corrupted image:

Does it affect the training in any way ? If so, is it significant ? Thanks

WenYanZhang99 commented 1 year ago

When I try to pre-train the vision model and ABINet I found that there are 3 evaluation datasets that contain corrupted image. Any fix for this ? Or do I just remove it from the configs ?

Below are the evaluation datasets that contain corrupted image:

  • SV
  • SVT
  • IC15_1811

Does it affect the training in any way ? If so, is it significant ? Thanks

same problem. Have you solved it?