refined_loss_1 is nan

[fanglinpu]

The refined_loss_1 is nan when I train the model using vgg16 backbone.

[vasgaowei]

Maybe the learning rate is larger. And you can try learning rate 0.001 which is the same as my settings'.

[fanglinpu]

I use the learning rate 0.001, but the refined_loss_1 is nan.

[vasgaowei]

Make sure you download the right version of pretrained 'vgg16.pth' file. And I give the link of the version of pretrained 'vgg16.pth' file which woks fine.