Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Make AI Model Training work with tasks #8265

Closed
3 tasks done
daniel-wer opened this issue Dec 5, 2024 · 6 comments · Fixed by #8310
Closed
3 tasks done

Make AI Model Training work with tasks #8265

daniel-wer opened this issue Dec 5, 2024 · 6 comments · Fixed by #8310

Comments

@daniel-wer
Copy link
Member

daniel-wer commented Dec 5, 2024

Detailed Description

As far as I can see two things are missing:

  • Allow to specify task IDs instead of annotation IDs as the training data source. To do this I think it would be easiest to look up the (or are there multiple?) annotation ID for a specific task ID and continue to use that. This way, no changes in the worker should be needed. (@fm3 Could you comment on whether that's possible?)
  • Take into account task bounding boxes and not only user bounding boxes

Context

  • Specific to long-running jobs (set jobsEnabled=true in application.conf)

Training data for model trainings is often created using tasks. It should be possible to start a training from these.

@fm3
Copy link
Member

fm3 commented Dec 5, 2024

Sounds good! Yes, there may be multiple annotation ids per task. We should also decide if only “finished” instances should be used, or also “active” ones.

@MichaelBuessemeyer
Copy link
Contributor

We should also decide if only “finished” instances should be used, or also “active” ones.

Maybe add a warning for each unfinished task? This way we can allow both and thus this feature is more flexible 🤔

@daniel-wer
Copy link
Member Author

I would vote to only use finished instances (and show a warning that there are x unfinished instances that are not used). It usually does not make sense to use unfinished training data.

@MichaelBuessemeyer
Copy link
Contributor

MichaelBuessemeyer commented Jan 8, 2025

This way, no changes in the worker should be needed.

Sadly, the worker needs adjustments as the bounding boxes collected by the frontend are not passed to the job and thus the worker collect the bounding boxes itself.

Edit: Therefore, the worker now needs to be adjusted to also include the task bounding box when collecting the bounding boxes

@daniel-wer
Copy link
Member Author

Sadly, the worker needs adjustments as the bounding boxes collected by the frontend are not passed to the job and thus the worker collect the bounding boxes itself.

But the worker previously collected the bounding boxes as well. Or is the needed change that the taskBoundingBox in addition to userBoundingBoxes are collected by the worker?

@MichaelBuessemeyer
Copy link
Contributor

MichaelBuessemeyer commented Jan 8, 2025

Or is the needed change that the taskBoundingBox in addition to userBoundingBoxes are collected by the worker?

Yes, that is what I meant. Sorry for being unclear here 🙈

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging a pull request may close this issue.

3 participants