The value of loss is unnormal with EuclideanLoss Layer

BVLC / caffe

Caffe: a fast open framework for deep learning.

http://caffe.berkeleyvision.org/

Other

34.12k stars 18.69k forks source link

The value of loss is unnormal with EuclideanLoss Layer #4614

Open yechaochen opened 8 years ago

yechaochen commented 8 years ago

I'm doing a multi-label regression training.The loss value is keep in 0-10 and I think it's normal because my label is in the former like 8.454 or 10.264. But the loss value suddenly bacome very high to the 1.*****e+13. My base_lr in solver.prototxt is 0.0001

Can someone tell me how can I solve this problem.Thanks

lingerois commented 8 years ago

Can you tell me how to use EuclideanLoss Layer? I doing what you are doing. Does EuclideanLoss Layer needs up to make every label of a image an ground truth label? I mean if I have 10 different labels, I have to pass 10 ground truth images to EuclideanLoss Layer right?