The label of threshold map? about db HOT 12 CLOSED

@MhLiao Thanks, I have solved by adding sigmod to threshold map, by the way, have you try supervised normal border map as threshold map like 'self organized text detection', and compare it in ablation study?

from db.

very good work , but i have a question:
why the label of threshold map is bigger closing to G, smaller closing to Gs and Gd, rather than bigger closing to Gs and smaller closing to G.

from db.

@saicoco You are correct. But I am not sure why your loss cannot be converged.

from db.

@saicoco We did not try it.
The threshold map would highlight the text border region even without supervision for the threshold map. This indicates that the border-like threshold map is beneficial to the final results. Thus, we apply border-like supervision on the threshold map for better guidance.
The threshold supervision makes the training more stable and obtains minor performance gain on some specific datasets. However, we think that it is not the key element of DB since the DB is effective even the threshold supervision is removed.

from db.

@MhLiao Thanks for you detail reply, I am understand it clearly that why you add a threshold supervision😊

from db.

@saicoco How abount your implementation result?

from db.

@Banyueqin I implement it based psenet codebase, just add a label for threshold map, two heads for threshold map and binarization map, I dont try deformablev2.

from db.

@saicoco Thanks for your reply, I have some problems in implementation. Can you give me some advice? I contact you with the email. Or do you have QQ?

from db.

@saicoco did you post the code for your implementation

from db.

@MhLiao
When two texts are close to each other, threshold map gt will also overlap, what to do at this time?

from db.

@saicoco @Banyueqin can we communication with qq？my qq is 572459439, please note DBNet

from db.

@MhLiao When two texts are close to each other, threshold map gt will also overlap, what to do at this time?

hi,can you please what i do for displaying the image of threshold map and probability map in the code

from db.