search

PingoLH / FCHarDNet

Fully Convolutional HarDNet for Segmentation in Pytorch
MIT License
195 stars 52 forks source link

Model architecture #9

Open khiemkhanh98 opened 4 years ago

khiemkhanh98 commented 4 years ago

Thank you for your implementation. I just wonder that where does the FC-HarDNet70 architecture come from as i don't see it being described in the paper?

PingoLH commented 4 years ago

Hi, thanks for the question. FC-HarDNet70 is a modified version of FC-HarDNet68/84 from the paper. In the paper, we focus on comparing with FC-DenseNet trained with CamVid dataset for the segmentation part so we didn't include the FC-HarDNet70 in the paper due to the page limitation. Nevertheless, architectures for them are very similar. They are all based on U-Net and the only difference is with the HarDNet blocks and the size configuration for each of the blocks. You can easily create another version on your own. Thanks!