search

TimoSaemann / ENet

ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation
577 stars 276 forks source link

PythonLayer is not imlemented in Multi-GPU training #53

Closed An-Pan closed 6 years ago

An-Pan commented 6 years ago

Hi, I have try to training using multi-gpu . I got this error

[python_layer.hpp:25] PythonLayer is not implemented in Multi-GPU training

So, does any one have experience on change the code spatial_dropout.py to use multi-gpu ?

ewen1024 commented 6 years ago

As far as I know, python layer is computed through cpu. So forget about multi-gpu

An-Pan commented 6 years ago

@ewen1024 Tanks~

ewen1024 commented 6 years ago

@An-Pan Meanwhile, I found the spatial drop taking so much time. I simply replaced it with the original verision of dropout and got nearly 2X speed and no noticeable performance drop.