I re-run the experiment based on VGG-19 backbone network. However, the results on VGG-19 is rather far less than the ones reported on the original paper. It might be caused by the hyperparameters setting?
If so, then how to set the parameters different from those based on CNN-F.
Thank you very much!
I re-run the experiment based on VGG-19 backbone network. However, the results on VGG-19 is rather far less than the ones reported on the original paper. It might be caused by the hyperparameters setting? If so, then how to set the parameters different from those based on CNN-F. Thank you very much!