Open hannah-rae opened 3 years ago
@bdubayah Have you made this change for the raster pixel loader too? Are you planning to PR/merge or are you still making changes?
Still need to do it for raster pixels. I think we also need to decide the merge strategy for overlapping patches - whether to stick with mean or do max instead. I'm going to try with max for the vtfs experiment and see how it works out.
Rasters often have nodata values that should be ignored in analysis (e.g., edges missing data due to registration). These pixels should be given a specific sentinel value (e.g., -9999) instead of being scored.