When running `Detection` -> `Detect`, would it be possible to support batched inference? Detection each image individually seems rather slow. Let me know Refers https://github.com/openjournals/joss-reviews/issues/5986