search

vasgaowei / pytorch_MELM

The pytorch implementation of the Min-Entropy Latent Model for Weakly Supervised Object Detection
104 stars 19 forks source link

refined_loss_1 is nan #7

Open fanglinpu opened 5 years ago

fanglinpu commented 5 years ago

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

vasgaowei commented 5 years ago

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

fanglinpu commented 5 years ago

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

vasgaowei commented 5 years ago

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.