How to adjust the weight of loss functions

wasidennis / AdaptSegNet

Learning to Adapt Structured Output Space for Semantic Segmentation, CVPR 2018 (spotlight)

847 stars 203 forks source link

How to adjust the weight of loss functions #86

Closed dh23333 closed 4 years ago

dh23333 commented 4 years ago

You said that modifying the model requires fine-tuning the parameters. Is it convenient for you to remind about the tricks about adjusting the weight of loss functions?

wasidennis commented 4 years ago

We just try different weights in a reasonable rage, e.g., [5e-3, 2e-3, 1e-3, 5e-4, 2e-4, 1e-4] and so on.