The mask of the patch classification inference result cannot cover the exact position. Is there a problem with PatchGenerator or PatchStitcher?

[liuyichong2250]

Hello! I tried running the example code given on https://fast.eriksmistad.no/neuralNetworkWSIClassification_8cpp-example.html, and I found that when displaying the inference mask for the patch, the mask is not correctly placed on the corresponding patch. Could this be an issue with PatchGenerator or PatchStitcher? Both the code and the test data are the ones provided in the GitHub repository.

Additionally, there are some warnings appearing in the terminal:
2025-02-21 18:26:05.030 show_fast[36066:12271637] 2025-02-21 18:26:05.030096 [W:onnxruntime:, helper.cc:61 IsInputSupported] Dynamic shape is not supported for now, for input:input_1:0 'Tissue Segmentation Model' has already been downloaded from FAST Data Hub 2025-02-21 18:26:07.311 show_fast[36066:12271637] 2025-02-21 18:26:07.311221 [W:onnxruntime:, helper.cc:61 IsInputSupported] Dynamic shape is not supported for now, for input:args_0 WARNING [0x16fe6f000] Patch size must be a multiple of 16 (TIFF limitation). Adding some overlap (56, 56) to fix.

[smistad]

Thanks for reporting, I am able to reproduce the error and I will investigate it.

[smistad]

There seems to be a bug with TissueSegmentation which was updated in version 4.10.
You can set the first argument in TissueSegmentation::create to true to use the old tissue segmentation method, while we try to fix the error:

auto tissueSegmentation = TissueSegmentation::create(true)->connect(importer);

[liuyichong2250]

Thank you for your help! I appreciate the quick response and look forward to the bug fix in future updates!

[smistad]

The bug has now been fixed: 598e65f
New packages with the fix are being built on GitHub now: https://github.com/smistad/FAST/actions
You can download the new packages when its done by pressing on the CI workflow for your operating system and downloading it under "artifacts".

[liuyichong2250]

The issue has been fixed. Thank you for your careful help and support!