Indeed. This is why data needs to be open, and the pipelines locally reproducible. All papers must include their data and training code and weights.