search

Eric-mingjie / rethinking-network-pruning

Rethinking the Value of Network Pruning (Pytorch) (ICLR 2019)
MIT License
1.51k stars 293 forks source link

about train() and eval() #16

Open leonzgtee opened 5 years ago

leonzgtee commented 5 years ago

anyone konws what the line128 model.tarin() and line 154 model.eval() in network-slimming/main.py are meant for. i did not find the defination of these two function,can i just delete them,thanks

Eric-mingjie commented 5 years ago

This is to ensure that the batch norm layer are set accordingly. These two lines should be kept.

leonzgtee commented 5 years ago

thanks for your reply.

abhigoku10 commented 5 years ago

@leonzgtee did u perform the network slimming, on which architecture how mucha was the performance gain from the main model . did ur accuracy drop ? by how much

leonzgtee commented 3 years ago

@abhigoku10 yes i did the network slimming, in our own training architecture with our own data, map drops about 0.01 from 0.38 to 0.37