I came away from this convinced that somebody needs to build a data sharing system for medical images. If there was a system for sharing images that had existe

I came away from this convinced that somebody needs to build a data sharing system for medical images. If there was a system for sharing images that had existed before Covid and enough data had been contributed that reviewers could demand evaluation on some predefined test sets, then a lot of these inconsistent evaluation issues could be weeded out. The article mentions federated learning, but I feel like that's solving a different problem than reliable evaluation.