Caffe2 discussion thread


(Karthik Kannan) #1

Hi guys,

I came across Caffe2 while browsing HN right now. The makers are pushing it as a mobile-first DL alternative to Tensorflow.

Would love to discuss more about Caffe2 and experiments anyone has tried with it! I will update this thread with results from my own experiments with it later this week.


(melissa.fabros) #2

There should be a tutorial on nvidia.qwiklab.com soonish? At least, that’s what the flyers at Facebook’s developer conference says. I didn’t see up when I just checked.

The network is pretrained on ImageNet and the demos look like the framework can do image segmentation and semantic labelling on the fly.


(Karthik Kannan) #3

So, you can apparently convert PyTorch code to Caffe2 and then deploy it on mobile.


(Karthik Kannan) #4

And according to this reddit thread the backends of both PyTorch and Caffe2 would be merged over time.