Current implementation requires each workflow {} to perform the matching of its inputs. This seems to create unnecessary code bloat in downstream modules, because the matching requires inferring which image and segmentation method were used upstream from the filenames. Consider the following alternative:
All entities are always in a tuple with their tag(s). For example, (image tag, image), (image tag, method tag, segmentation mask), (image tag, method tag, SFT), etc.
The tags for pre-computed intermediates are determined at load time.
Every process should accept tag(s) in the input tuples and return (possibly unmodified) tags in their outputs
Current implementation requires each
workflow {}to perform the matching of its inputs. This seems to create unnecessary code bloat in downstream modules, because the matching requires inferring which image and segmentation method were used upstream from the filenames. Consider the following alternative:(image tag, image),(image tag, method tag, segmentation mask),(image tag, method tag, SFT), etc.