search

hli1221 / imagefusion_densefuse

DenseFuse (IEEE TIP 2019, Highly Cited Paper) - Python 3.6, TensorFlow 1.8.0
244 stars 62 forks source link

the loss is nan #14

Open cena001plus opened 3 years ago

cena001plus commented 3 years ago

i have try different lr from 0.00001 to 0.1 , but thr loss is nan , why? image

LuoXubo commented 2 years ago

same problem, have you solved it?

GHGluck commented 2 years ago

I also had the same problem. 1648638909(1)

hli1221 commented 2 years ago

Hi, thank you very much for report this problem.
Normally, the pixel loss should be decreased with increasing iteration number. However, the pixel loss was increasing. I suggest you can check the input and the output, this problem may occur if they are not in the same domain.

GHGluck commented 2 years ago

Thank you for your reply. I tried to modify the input data and found that sometimes the training can be successful, but when I increase the data, the loss will still be none. I don't know if it is because the network cannot fit my training data.