Resnet101-FCN - Githubissues

zijundeng / pytorch-semantic-segmentation

PyTorch for Semantic Segmentation

MIT License

1.73k stars 393 forks source link

Open shahabty opened 6 years ago

shahabty commented 6 years ago

Hello, I cannot find the implementation of resnet 101 FCN in this repo. Is there any code?

Thanks,

guanfuchen commented 6 years ago

I either can not find this version of FCN.

shahabty commented 6 years ago

Hello @guanfuchen, Here is a link for detailed info of Res101-FCN. It is used in NIPS 2017 paper "Predicting Scene Parsing and Motion Dynamics in the Future" Link: https://papers.nips.cc/paper/7267-predicting-scene-parsing-and-motion-dynamics-in-the-future-supplemental.zip

I think the original version is implemented in Caffe.

guanfuchen commented 6 years ago

@shahabty Thanks for your paper for the detail about ResNet101-FCN，I will try my code fcn_resnet.py to experiment. Here is some doubt:

the upsampling in the origin paper choice, bi-linear-upsample or convtranspose.
the dilation rate 2 in the conv5 block means all resnet block, or just the first conv in the block.