How to get latent representations of all training samples during pretraining

[zhaozh10]

I hope to observe the change in each sample's latent representation every epoch, and I hope to get the representations obeying the order of the dataset. How can I implement it in mmselfsup?

[YuanLiuuuuuu]

In loss func (which is defined in the algorithm module in MMSelfSup 1.x branch), after you obtain these features, you can directly save them to your disk for further analysis.