search

RubanSeven / CRAFT_keras

Keras implementation of Character Region Awareness for Text Detection (CRAFT)
Apache License 2.0
166 stars 57 forks source link

Why did you modified the original Affinity Box generation method? #10

Closed Antonio-hi closed 2 years ago

Antonio-hi commented 5 years ago

From my point of view, the original method is simpler and easier to understand. Of course, your implement is also great, but it is not particularly easy to understand. Is there a proof that your method is better than papaer proposed. Thanks for your contribution

Antonio-hi commented 5 years ago

update: I found two issue below The first is that the confidence score of weak supervised training has some bugs The second is that the generation of weak supervised labels has not been modified like your readme description

Antonio-hi commented 2 years ago

close this work