I would still ask the authors to cross-check with implementation. Replicating papers is hard. I also suggest
See also