Perceptual Loss

[vishnusanjayrs]

Thanks for sharing your amazing work.

In the paper you mentioned that " adversarial loss for the generator is combined with the Laplacian pyramid loss to
form the perceptual loss for ST-MFNet fine-tuning" can you please let us know the value for lambda used and after how many epochs of training with just Laplacian pyramid loss this perceptual loss term was introduced.

Thanks

[danier97]

Hi,

Thank you for your interest in the work!

The model was trained with the Laplacian loss for around 70 epochs. Then fine-tuning was performed for a further 10 epochs with $\lambda$=100.

Thanks.

[vishnusanjayrs]

Thanks for the quick reply