phzwart / qlty

A utility to patch pytorch tensors into smaller, overlapping bits and stitch them together again.
Other
0 stars 3 forks source link

bug fix to numba implementation #5

Closed Giselleu closed 2 months ago

Giselleu commented 2 months ago

Changed tmp @ weight to tmp weight. The previous one caused pixelated segmentation results (not noticeable for small qlty window sizes (<= 64), but prominent for larger window sizes. Changing to gives identical result as the original implementation without numba.