Repository navigation
Fix intensity scaling range in MAISI VAE transform - #2084
Conversation
RandScaleIntensityd scales by (1 + factor), so factors=(0.9, 1.1) multiplied images by 1.9-2.1. Use (-0.1, 0.1) for a 0.9-1.1x range. Signed-off-by: Don Yin <Don_Yin@outlook.com>
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
WalkthroughTraining-time ChangesIntensity scaling augmentation
Priority: ➖ Normal Estimated code review effort: 2 (Simple) | ~5 minutes Change: Bug fix Suggested reviewers: Merge Risk: ⚪ Minimal · up to
|
| Check name | Status | Explanation |
|---|---|---|
| Title check |  |
The title clearly and concisely identifies the main change: correcting the intensity scaling range in the MAISI VAE transform. |
| Description check | |
The description explains the issue, proposed fix, observed impact, validation, and relevant checks. It omits the template's issue reference, but the required technical content is present. |
| Docstring Coverage | |
Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 1 functions across 1 files. |
| Linked Issues check | |
Check skipped because no linked issues were found for this pull request. |
| Out of Scope Changes check | |
Check skipped because no linked issues were found for this pull request. |
✨ Finishing Touches
-
🧪 Generate unit tests (beta)
-
- Create a new PR
-
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
Comment @coderabbitai help to get the list of available commands.
|
Thank you! This is very helpful! |
2 participants
Description
RandScaleIntensitymultiplies by1 + factor, sofactors=(0.9, 1.1)indefine_vae_transformscales images by 1.9-2.1x. I think(-0.1, 0.1)(0.9-1.1x) was the intent, which is also how the transform is used elsewhere in the tutorials, but please correct me if I've misread it.I ran the transform on a public CT and MRI scan to check. With the current value, fat in CT lands roughly where bone normally sits, and soft tissue goes above 1.
As far as I can tell this doesn't break anything, since the VAE reconstructs the same scaled patch, and the released weights and reported results shouldn't be affected. It may cost a little reconstruction quality, but I haven't measured that. If the 2x range is intentional, I'm happy to turn this into a comment instead. I've also raised it in NVIDIA-Medtech/NV-Generate-CTMR#53.
Checks