Fine Tune Training by BryonLewis · Pull Request #1376 · Kitware/dive

Skip to content

Navigation Menu

Sign in

Appearance settings

Conversation

@BryonLewis

Will allow users to choose a model to allow for fine-tune training.
resolves #1375

  • Logs the model files that are available after downloading the add-ons for the web version
  • The endpoint for training configs has been updated to show the models as well as the configs.

TODO:

  • Update front-end to utilize modified Training configs
  • Add optional checkbox on front-end to fine-tune model
  • Update the endpoint to run training so that a model can be passed in as well
  • Add in argument to the training task in celery to utilize new input

Desktop Version

  • Find a way to search and find available models in the desktop version
  • Update the interface to show the models that can be fine tuned
  • Update the call to running training to add in the new optional fine-tune training model.

@BryonLewis

Everything is templated out for the fine-tune training just need some executables (docker container and VIAME desktop) to test with.

@BryonLewis

Tested the following:

  • Web version will display the --init-weights command when configured in the UI.
  • Desktop version does the same and displays --init-weights.

Possibly Todo:

  • Better arrangement of values (like dividing into system models and user models)
  • Styling for the training menu so it expands better, possibility having a max height and scrolling.

naglepuff

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Full disclosure - I did not get a chance to do any functional testing of this branch, but I did give the code a look, and found some suggestions.

Co-authored-by: Michael Nagler <mike.nagler@kitware.com>
Co-authored-by: Michael Nagler <mike.nagler@kitware.com>

naglepuff

2 participants

@BryonLewis @naglepuff