Imagenet results reported from Linear Eval or Fine tuning?

vturrisi / solo-learn

solo-learn: a library of self-supervised methods for visual representation learning powered by Pytorch Lightning

MIT License

1.38k stars 181 forks source link

Imagenet results reported from Linear Eval or Fine tuning? #372

Closed LightingMc closed 8 months ago

LightingMc commented 8 months ago

vturrisi commented 8 months ago

Only MAE was finetuned.

LightingMc commented 8 months ago

What is the LR rate that you used for the linear Eval? I thought it was supposed to be a fixed LR of 0.1. But the scripts in the repo say otherwise. I am having a hard time reproducing the results at a fixed LR of 0.1

vturrisi commented 8 months ago

It depends on the method. We used the setting in our config files.