search

IBM / SynthTabNet

Dataset of PNG images from synthetically generated table layouts with annotations in JSONL files
Other
128 stars 11 forks source link

Could you provide details about inference? #13

Closed ejlee95 closed 2 years ago

ejlee95 commented 2 years ago

Thanks for uploading SynthTabNet.

I have a question about your paper, TableFormer.

Could you provide more details about table recognition of large table (with more than 512 tokens) during test time?

Do you split the input image and run the code multiple times? or just ignore the tokens longer than 512 tokens?

nassarofficial commented 2 years ago

Thank you for your interest in our work. The input image size, and token size is only limited during training for memory issues. During evaluation, and testing, it is unconstrained. In the paper, it's ignored.