Closed jadie1 closed 1 year ago
@jadie1, I have re-uploaded the DeepSSM tiny test use case to the new data portal. Can you please check the image quality of the downloaded images now?
@zahidemon These look correct now! Closing the issue
@zahidemon , @jadie1, sorry I didn't reply sooner.
I resized/resampled the DeepSSM tiny test femur images to be 100 times smaller (<1mb per image vs 100mb per image) to save both time, bandwidth and disk space for PR unit testing. We are extremely limited on disk space on the Github Runners and are having to cut corners in many places. The data download for DeepSSM tiny test is now 617MB with the full size images. It was <10mb previously.
The purpose of the automated unit test is really only to confirm that the code runs, e.g. no syntax errors, api call errors, etc. We cannot really check the accuracy of the execution.
The full DeepSSM use case should be the full size images. If that was in error, then yes, we must fix it, however the tiny test really needs to be kept small.
I didn't find any error with the full DeepSSM use case. @akenmorris, If you have the resampled images for the tiny test use case, I can upload them again.
@akenmorris I apologize when I first made the issue I thought it was a problem with the full use case. Later I realized it was only the tiny test. It makes sense to use smaller images for the tiny test
@zahidemon Here are the reduced size images:
http://www.sci.utah.edu/~amorris/shapeworks/deep_ssm.tar.gz
Do we have a problem that the download for both the regular and tiny test go to the same place? Output/deep_ssm/femur
? Do we need to further separate the tiny tests?
@akenmorris I have re-uploaded the reduced size images and tested the use case. It's working as expected.
I don't think we need separate paths for regular and tiny test dataset downloads. It's already handled inside the use case.
Resolved.
I noticed the femur images on the portal are about half the size memory-wise of the originals. Here is the original m03 image:
And here is the one from the portal:
![image](https://github.com/SCIInstitute/ShapeWorks/assets/54382613/392ce388-15e5-4b7e-bbfe-23f0572bfbb6)
I'm not sure if this is a compression issue or what, but it could impact the accuracy of the DeepSSM use case. I've put the original images here: /home/sci/jadie/Public/femur_images/