zijundeng/pytorch-semantic-segmentation

Typos in the FCN model

JiahuiSun opened this issue · 0 comments

I found some strange settings:
score_pool4 = self.score_pool4(0.01 * pool4)
score_pool3 = self.score_pool3(0.0001 * pool3)
but in my experiments, the model without 0.01 and 0.0001 performs better than the original one.