search

voldemortX / pytorch-auto-drive

PytorchAutoDrive: Segmentation models (ERFNet, ENet, DeepLab, FCN...) and Lane detection models (SCNN, RESA, LSTR, LaneATT, BézierLaneNet...) based on PyTorch with fast training, visualization, benchmarking & deployment help
BSD 3-Clause "New" or "Revised" License
824 stars 137 forks source link

About the model size #134

Closed huanFigure closed 1 year ago

huanFigure commented 1 year ago

Hello: Your paper is very good, but I have a question about the size of the model parameters in your experiment, the baseline (resnet-101) size is 43.56 ,but the RESA(ResNet-101) is just 31.46, and LaneATT(ResNet-122) size is 8.55? and the pretrained resnet-18 is 44.7M, RESA(Res18) is only 6.61M,

voldemortX commented 1 year ago

@huanFigure It is because lots of papers like resa only use stage3 of reset feature.

huanFigure commented 1 year ago

ok,thanks for your reply

voldemortX commented 1 year ago

@huanFigure As for LaneATT, resnet122 is a special architecture, which has very few parameters.