wvangansbeke/Unsupervised-Classification

Is knn for validation?? (memory bank, target(label))

edwardcho opened this issue · 2 comments

Hello Sir,

I still have some problem.
Is knn for validation (pretext, scan, selflabeling)??

If so, memory bank and targets are do not need for training??

Thanks,
Edward Cho.

Hello Sir,

I found that knn needs for scan clustering. ('anchor' and 'neighbor')
So memory bank and target(label) do need. Right?

Thanks,
Edward Cho.

We only use labels during evaluation. You can't compute the knn accuracy without targets.