Rastervision is a pretty extensive dependency because it in turn pulls in pytorch. I've replace the aggregation functionality provided by rastervision in #37 so it's now less important. However, it's still used for metric computation and to generate training chips. If we want to replace these with our own code this might simplify things a lot.
Rastervision is a pretty extensive dependency because it in turn pulls in pytorch. I've replace the aggregation functionality provided by rastervision in #37 so it's now less important. However, it's still used for metric computation and to generate training chips. If we want to replace these with our own code this might simplify things a lot.