search

wuhuikai / FastFCN

FastFCN: Rethinking Dilated Convolution in the Backbone for Semantic Segmentation.
http://wuhuikai.me/FastFCNProject
Other
838 stars 148 forks source link

ImageNet pre-trained model #60

Closed SeokjuLee closed 4 years ago

SeokjuLee commented 4 years ago

Hi, thanks for your work. Did you initialize your backbone architecture (ResNet50 or 101) with the ImageNet pre-trained model? Have you tried some ablation experiments, such as training with or without the ImageNet pre-trained model?

wuhuikai commented 4 years ago

All the results are initialized with models pre-trained on ImageNet; We do not have any experiments w/o ImageNet pretraining since it will lead to poor performance as other papers state.