GSoC'23 | Add Unit Tests | Continuing Discussion from Issues

[skat00sh]

Hello @paulojamorim
Continuing the discussion from #496 (comment)
I've a few queries:

1. You said "the segmentation can have pre-segmented files to verify that the methods are working correctly." : I don't really understand what does pre-segmented file mean here? Does it mean Ground-Truth segmentation masks or something else?

2. What sort of code-coverage are you aiming for? Is the idea ultimately to move somewhat in the direction of Test-Driven Development in future? I'm asking that since that would mean to me that Adding Unit tests could be treated as some sort of technical debt and then moving forward TDD becomes a norm

3. If you please share what is expected out of this project at the least and what would be good to have?

Thanks in advance

P.S: Apologies that I never really introduced myself (I just got veru intrigued when I saw this project and instantly wanted to understand better).
I'm Devendra, final year master's student at TU Munich. I've worked in Object Detection, Tracking and Segmentation both in an academic/research and corporate setting. I've worked both with C++ and Python. In Deep Learning, I'd say my skill-set is in Computer Vision based tasks like Detection/Tracking/Segmentation, 3-D reconstruction and robustness and privacy of networks. From the software engineering side, I've worked as a Self-Drving Car Engineer in a project where required Test Driven Development and Clean coding was a must.

The project sounds really interesting and I'd love to contribute to invesalius.

[paulojamorim]

Hi @skat00sh

Thank you for introducing yourself.

About your questions:

1. Yes, it's the ground truth.

2. We are not looking for TDD. We want the unit tests to know that everything is working correctly before of release.

3. It's interesting to have the tests of basic functions that compare the output of the functions, tests with the ground truth (To verify that segmentation methods are working as expected). We could have tests for the volume rendering method, triangular mesh generation.

Best regards

[skat00sh]

Understood!

1. So for comparison with the Ground Truth we are essentially looking to check some metric value then right? Like Intersection over Unioni (IOU) or mean-IOU, right?

2. I was going through the application process described here : Write a working code sample relevant for the proposed project. Is there a preferred module that you'd like me to write some unit-tests on in a secret gist or any module works?

3. So one idea could be that there is a bash script or a task runner like go-task, that'll run and report on all sort of tests before release. (I worked on making something similar in my last student job and it worked quite well)

[paulojamorim]

Hi @skat00sh ,

1 - Yes, you can use one of these metrics.

2 - We don't have a preferred module, it can be anyone.

3 - Yes, that's the idea but use some tool available on GitHub Action to run the tests every day. Because later we intend to generate releases every day (aka nightly-build).

[skat00sh]

Thank you @paulojamorim!
For point no. 3, that makes more sense to just let Github actions do it on every PR may be.
This does look like a project that I'll be interested in in. I'll try to finish the example task soon and share the proposal. Would be great if you could please review it once :)