search

namtuanly / MTL-TabNet

MTL-TabNet: Multi-task Learning based Model for Image-based Table Recognition
Apache License 2.0
78 stars 12 forks source link

[Question] When model trains, do I need HTML tag? #15

Open LightAllWorld opened 2 months ago

LightAllWorld commented 2 months ago

Hello,

I interest of your paper. And I have some questions.

  1. When your model trains, HTML tag data is optional? Because, fig. 2 of your paper is written shared decoder's input is HTML tags. But Fig 1. is tell me need only table image.
  2. I don`t need structure decoder. How to run model without that decoder?

Thanks,

namtuanly commented 1 month ago

Hi, Thank you for your interest in our work.

The following are the answers to your questions. I hope they are useful for you.

  1. You need the table images and the corresponding HTML tags for training the model (Fig. 1 just show the overview of the model in the inference phase).

  2. You can change the code in master_decoder.py (mmocr/models/textrecog/decoders/master_decoder.py) to remove the structure decoder (note that if you remove the structure decoder, you can not get the result from other decoders, because the other decoders work based on the results of the structure decoder).

lerndeep commented 1 month ago

is it possible to train this module only for cell and table structure recognition without cell contain? if yes please let me know how? THank you