How did you use unlabeled dataset? about scrl HOT 2 CLOSED

SCRL doesn't need labels to train a model since it is a self-supervised learning method.

I think you probably found an online evaluation block that uses labels. We've implemented a code to evaluate the model's linear evaluation result on the fly. It means if you disable the online evaluation mode, you don't need labels to train a model.

Even though you enable the online evaluation mode, the loss from the online evaluation block is detached. It means it does not affect the model training.

from scrl.

답변 감사합니다 :)

I thought this method was using an unlabeled dataset. But looking at the 88 line of the trainer.py, it needs labels. So I misunderstood. But after watching your answer, I saw the 234 line part of trainer.py, the label will only be used for the evaluation part. From now on, I will read the code in more detail. Thank u 👍

from scrl.