sign-language-processing / transcription

Text to pose model for sign language pose generation from a text sequence
33 stars 16 forks source link

command not found: text_to_text #12

Open Yaffa16 opened 1 year ago

Yaffa16 commented 1 year ago

I tried to riun the model but I keep. getting the :command not found: text_to_text: error

AmitMY commented 1 year ago

Similar to #8

The text_to_text step is not currently done, and so the binary does not actually work.

Please note that the README "hides" it under "next steps" for that reason. Help in developing this is always much appreciated

aarush-dev commented 8 months ago

is it completed now? What needs to be done as i would be happy to help.

AmitMY commented 8 months ago

Hi Aarush - matter of fact, there is progress! Over the last few months I have cleaned the data for this step (https://github.com/sign-language-processing/signbank-plus) Over the last 2 weeks I moved the bergamot training pipeline here and have made great progress -

  1. got the full pipeline to work (after solving issues with https://github.com/mozilla/firefox-translations-training)
  2. around 1 hour ago, managed to get training to run on multiple gpus https://github.com/sign-language-processing/signbank-plus/commit/8299fb591e3552c9e89fc18f612b1d37e060fba4

This means that now we are at the experimentation phase - and the next phase would be to redo the text_to_text part of this repository to use the other one as a submodule and create a binary executable.

If you want to help with this specific part of the pipeline, my first recommendation is to check out the open issues - mainly regarding data generation https://github.com/sign-language-processing/signbank-plus/issues Better data, better models.