how can Patch-based 3d segmentation get full heatmap

[Da1daidaidai]

Thanks for your geart work.

Some 3d image is too large that it uses patch to crop the whole image (200200300) into patches(969696) . when using medcam, how can we get one full heat map of such pathch-based inference ?

[Karol-G]

Hey,

you will need to aggregate the heatmap patches back to the original image size. You can use the GridSampler and Grid Aggregator from TorchIO for that (https://torchio.readthedocs.io/patches/patch_inference.html).

Best, Karol

[Da1daidaidai]

Thank you for your quick reply :)

I tried the sliding_window_inference in MONAI, it give a full return, while the whole heatmap is black I have tried different model like unet dyunet , the result is the same. Is there any solution?

Thanks again :）