Skip to content

A vectorized approach to coord conversion. Provides a significant speed up.#44

Open
mathiser wants to merge 1 commit into
Sikerdebaard:masterfrom
Aarhus-RadOnc-AI:VectorizeMemoryOptimized
Open

A vectorized approach to coord conversion. Provides a significant speed up.#44
mathiser wants to merge 1 commit into
Sikerdebaard:masterfrom
Aarhus-RadOnc-AI:VectorizeMemoryOptimized

Conversation

@mathiser

Copy link
Copy Markdown

Hi @Sikerdebaard
Another PR with vectorized approach. A little tweaked compared to the other PR I closed yesterday. It provides a significant speed up by approximately 50 % (1 min 3 sec vs 28 sec in my test)

It does two things:

  1. doing all coord transforms in one numpy call
  2. only drawing contour for the area and resampling the contour back to the original image again.

Mathis

@Sikerdebaard

Sikerdebaard commented Mar 7, 2023

Copy link
Copy Markdown
Owner

Hi @mathiser

I have tried to run your vectorized solution and I am running into two issues:

  • There is a merge conflict when trying to merge this PR with the master branch
  • The IoU scores on the unit tests are reported as 0, which means no overlap

Especially the last issue seems problematic. It shows that the resulting masks in the nii files have no overlap with the expected masks. Upon manual inspection this seems to be because the mask is empty. Is it possible for you to have a look at this issue?

@mathiser

mathiser commented Apr 6, 2023

Copy link
Copy Markdown
Author

Yirks, sorry about that. I'll review it ASAP...

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.

2 participants