Skip to content

Conversation

@Wiebke
Copy link
Member

@Wiebke Wiebke commented Nov 20, 2025

This pull request introduces improvements to the result selection and display. This includes:

  • Restructuring of the render_image callback and callbacks that store training and inference result to ensure that shown data and shown inference result match (Segmentation results remain on display after switching dataset #188 is partially addressed, resolves Inference results persist as overlay even after new training job submission #190). The refactor of render_image included separating out the jumping to annotated slices into a dedicated callback
  • Improved logic in populate_segmentation_results to provide user notifications when segmentation results are unavailable, either due to failed training/inference or mismatched image URIs. (resolves Result retrieval does not properly handle failed flows #203)
  • Improved logic for result overlay toggle (is no longer automatically deselected, when the user selects a failed/incomplete job)
  • Added clearable and searchable options to both train-job-selector and inference-job-selector to allow users to easily clear and search jobs in the dropdowns. The clearable option is a bit of a work-around to a user having selected running job, that then completes. This doesn't trigger any result overlay callbacks (since the data changes, but not the value, as the value is equal to the flow uuid), as we do not explicitly monitor and act on state changes in the currently selected value

Interestingly, I can now no longer reproduce #204.

This does not change functionality but simplifies logic of `render_image` callback
…ent data

Previously, `render_image` was only called when:
- the image slice index changed (this is also triggered when a new data set is loaded, even if the slide index remains the same)
- the overlay toggle is used
Since this was not dependent on changes of results (e.g. a inference or training job is selected), it can happen that outdated result information is shown if the overlay toggle was not toggled in between switching jobs. Inference results from other data sets, also
Partially adresses mlexchange#188, adresses mlexchange#190
The drop-downs being used both as result selection and as state trackers for jobs complicates this.
@gitnotebooks
Copy link

gitnotebooks bot commented Nov 20, 2025

@Wiebke Wiebke requested review from taxe10 and xiaoyachong November 20, 2025 16:30
Copy link
Contributor

@xiaoyachong xiaoyachong left a comment

Choose a reason for hiding this comment

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

Thanks for your PR. I tested it on my end and confirmed that all the issues you mentioned no longer occur. I also haven’t seen #204 on my side. This PR has no conflicts with the main branch or with the upcoming Prefect upgrade PR (#201). I think it’s ready to merge.

@Wiebke Wiebke merged commit d7b011b into mlexchange:main Nov 26, 2025
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Result retrieval does not properly handle failed flows Inference results persist as overlay even after new training job submission

2 participants