This script uses 50% overlap in 2 dimensions, with a 'mode' resampling algorithm, to attempt to smooth over noise.
However, labels have error and post-processing options would also be nice. This would require reading the label mosaic tiff using rasterio, and carrying out one of a set of pre-defined filters. In the past, I have used morphological operators from scikit-image to remove small holes and blobs. This could be done on each binary mask (each individual class map), and any gaps can be filled in using inpainting (of which there is a quick method inside doodleverse_utils already)
This script uses 50% overlap in 2 dimensions, with a 'mode' resampling algorithm, to attempt to smooth over noise.
However, labels have error and post-processing options would also be nice. This would require reading the label mosaic tiff using rasterio, and carrying out one of a set of pre-defined filters. In the past, I have used morphological operators from scikit-image to remove small holes and blobs. This could be done on each binary mask (each individual class map), and any gaps can be filled in using inpainting (of which there is a quick method inside doodleverse_utils already)