search

wasidennis / AdaptSegNet

Learning to Adapt Structured Output Space for Semantic Segmentation, CVPR 2018 (spotlight)
847 stars 203 forks source link

why stopping trainning at num-steps-stop(120000) ? #48

Closed idealwei closed 5 years ago

idealwei commented 5 years ago

I see you set two iteration controller(num-step&num-steps-stop) in your code. Could you explain why you finish trainning at iteration num-steps-stop rather than iteration num-steps.

wasidennis commented 5 years ago

It is just for an early stopping strategy to save the resource, as the model is often over-trained (over-fitting to the source data) under the domain adaptation setting. Of course, you can still complete the training, but those models would not be good anyway.