Comment by muds

I admit that Imagenet isn't the best analogy here. But I'm pretty confident that this data cleaning issue would be caught in peer review. The biggest issue which I still don't understand was the removal of the test set. That was bad practice on the authors' part.