Open hunterlew opened 7 years ago
I've trained based on ZF net on my GTX965 2G memory and failed even the batch size was set to 2. Any other tricks to address it?
Maybe you could try the newest CUDNN, it will save GPU memory. And I rewrite faster rcnn in C++ and keep synchronization with official caffe, you may try https://github.com/D-X-Y/caffe-faster-rcnn/tree/dev
I've trained based on ZF net on my GTX965 2G memory and failed even the batch size was set to 2. Any other tricks to address it?