search

xuewenyuan / ReS2TIM

ReS2TIM: Reconstruct Syntactic Structures from Table Images
Apache License 2.0
22 stars 9 forks source link

Generating structures for new images #2

Open Kehindeajayi01 opened 2 years ago

Kehindeajayi01 commented 2 years ago

Hello, Please can your model be used to generate table structures for the ICDAR 2019 Modern table images?

Thanks

xuewenyuan commented 2 years ago

That can be difficult. ReS2TIM is time-consuming and a lot tables in ICDAR 2019 has a large of cells. Another reason is that ReS2TIM cannot well handle graphs with circle. You can try TGRNet that is more robust.

Kehindeajayi01 commented 2 years ago

TGRNet only includes ICDAR 2019 historical tables but does not include the modern tables. Also, how can I use TGRNet to generate structures for new table images? Can you point me to the right script to do that?

Thanks

xuewenyuan commented 2 years ago

TGRNet only includes ICDAR 2019 historical tables but does not include the modern tables.

TableGraph-24K contains tables that are converted from Latex files.

Also, how can I use TGRNet to generate structures for new table images? Can you point me to the right script to do that?

If you just use the pretrained model to infer on new images, you can prepare a test.txt file as other datasets.

Kehindeajayi01 commented 2 years ago

TGRNet only includes ICDAR 2019 historical tables but does not include the modern tables.

TableGraph-24K contains tables that are converted from Latex files.

Also, how can I use TGRNet to generate structures for new table images? Can you point me to the right script to do that?

If you just use the pretrained model to infer on new images, you can prepare a test.txt file as other datasets.

It appears that the results for ReS2TIM are for the threshold of 0.5 IOU, please can you point me to how I can vary this threshold for ReS2TIM? I could do that for TGRNet in the evaluator_vis.py script

Thanks