Currently planned implementations of smoothing operations ( and dilating) require iteration over the entire rectangular volume that encompasses the target label. This frequently results in many blocks being unnecessarily accessed. A solution to this would be to reduce the target image by the blocksize in each dimension, and setting the result to a binary mask, depending on the target label. You can then apply the desired operation to much smaller image, and determine which blocks need to be accessed for the original operation on the target image.
Specifically, for use with #472 when it is implemented
Currently planned implementations of smoothing operations ( and dilating) require iteration over the entire rectangular volume that encompasses the target label. This frequently results in many blocks being unnecessarily accessed. A solution to this would be to reduce the target image by the blocksize in each dimension, and setting the result to a binary mask, depending on the target label. You can then apply the desired operation to much smaller image, and determine which blocks need to be accessed for the original operation on the target image.
Specifically, for use with #472 when it is implemented