Batched/parallel implementation of GriTS

microsoft / table-transformer

Table Transformer (TATR) is a deep learning model for extracting tables from unstructured documents (PDFs and images). This is also the official repository for the PubTables-1M dataset and GriTS evaluation metric.

MIT License

2.31k stars 257 forks source link

Batched/parallel implementation of GriTS #12

Closed MrinalJain17 closed 2 years ago

MrinalJain17 commented 2 years ago

Hi @bsmock @rohithpv ,

I was wondering if you explored implementing a batched/parallel version of GriTS. It might be quite helpful when tuning a bunch of model hyperparameters on (relatively) set of large document images.

Or, if not, could you provide any suggestions on how to approach this? I'd be happy to try and work on a PR.

bsmock commented 2 years ago

Thanks for the suggestion! We added this to our roadmap several months ago and started working on it in a branch that we recently merged into main. The new code supports GPU batched inference and parallelization of GriTS on CPU. It should provide a pretty significant speed-up over the first version of the evaluation code.