Output issue

[MLamk-CLB]

Hi,
I’m currently using miphei, and I had a few questions about the output.
When visualizing the output in qupath,i can clearly distinguish the processed tiles (256px résolution). I tried to overlap tiles, still it remains heterogeneous.
Is there a way to homogenize the virtual staining across all tiles from a single sample?
Best regards
Mehdi Lamkhioued, MD PhD

[Guillaume-Balezo]

Hi, thanks for pointing that out.

run_wsi_inference has been fixed. There was an issue with vertical misalignment (see #7) and with overlap handling, where overlap regions were not correctly cropped. We also increased the default tile size from 256 to 512 during inference, which helps reduce tile artifacts, but they can still be visible.

A possible improvement would be to merge overlapping predictions during inference, but this would increase code complexity, RAM usage, and inference time.

Is it better with the latest update?