the train_mean_loss

[tht106]

Hello, thank you for presenting your excellent work!

While pretraining aimclr, how do you measure the progress with respect to the loss function train_mean_loss, what is a good value demonstrating that we can stop the aimclr pretraining to yield a good encoder?

Thanks