search

wasidennis / AdaptSegNet

Learning to Adapt Structured Output Space for Semantic Segmentation, CVPR 2018 (spotlight)
845 stars 205 forks source link

Question about weight decay #85

Closed taeyeopl closed 4 years ago

taeyeopl commented 4 years ago

Thanks for sharing good work. I have one simple question about your implementation.

https://github.com/wasidennis/AdaptSegNet/blob/025a7f54dca681e30fe02327bac46c19dfd8c27c/train_gta2cityscapes_multi.py#L48

Can I asked why you used the 0.0005 instead of 1e-4 which mentioned in the paper section 5.

wasidennis commented 4 years ago

Sorry for the confusion. The weight decay we are using should be the one in the implementation. We will fix the typo in the paper.

taeyeopl commented 4 years ago

Thanks !!