aCASTLE icon indicating copy to clipboard operation
aCASTLE copied to clipboard

Trying to run eval_gfsl

Open chinmayembedded opened this issue 4 years ago • 2 comments

Can I run eval_gfsl.py without training the model? I think I need to train the model first but if its possible can you please provide trained model?

chinmayembedded avatar Mar 13 '21 15:03 chinmayembedded

Hi,

We are working on releasing the models trained for GFSL. Will keep you posted once we released the models.

To quickly assess the GFSL performances, one way is to directly evaluate the pre-trained model weights in terms of GFSL (https://drive.google.com/open?id=14Jn1t9JxH-CxjfWy4JmVpCxkC9cDqqfE). Note that these weights are obtained from supervised pre-training on each dataset we marked and should give reasonable performances (but will have a gap comparing to our reported ones).

Best,

hexiang-hu avatar Mar 15 '21 17:03 hexiang-hu

Thank you!

chinmayembedded avatar Mar 19 '21 22:03 chinmayembedded