Closed KleinYuan closed 7 years ago
CUDA is a requirement, it was really slow on CPU so we ended up not finishing the CPU code so far. What do you mean by dependencies on multiple caffe models?
OK sorry, I got confused when you said multiple because right now only 2 are available (COCO/MPI). Forget hands and face so far, they are not available yet (we are working on them).
We are interested in supporting also Caffe2, but I am not familiar with it yet. But I thought it would have by default some kind of tool to convert models from Caffe, is there any? If so, then it should be easy to convert the COCO-MPI models and relatively easy to create a basic wrapper for Caffe2 and run OpenPose on top of it as we are doing with Caffe.
@gineshidalgo99 Thanks for the clarification and that's quite helpful since I didn't dig too deep into the src code.
Yeah, Caffe2 offers a translator to automate translating Caffe model to Caffe2 pb files. I did a bunch of tests and it works pretty good. I would like to give it a try to see whether it works with Caffe2.
Alright. Good chat. I think we can close this issue and looking forwards to you guys coming work. Good job!
Please @KleinYuan, let us know if you are able to compile it with Caffe2, or make a pull request as a new branch, so we can add it! Best
Issue summary
Firstly, congratulations on your great job! I managed to run it on my GeForce 980 ti card and it's quite smooth (like 7fps).
I am working on a Caffe2-iOS project and thinking it will be super awesome if we can come up with a wrapper or Cocoa Pod wrapping Caffe2 models to run it on iOS.
However, it seems that you guys have a dependencies on multiple caffe models, which got it not that straightforward.
Do you guys have any plan to get a Caffe2 support (I would like to have a try but it seems lots of work) ?
And also btw is this project running on CUDA a requirement or just because it's slow on CPU ?
Type of issue
Your system configuration
Operating system (
lsb_release -a
on Ubuntu): iOS CUDA version (cat /usr/local/cuda/version.txt
in most cases): X cuDNN version: X GPU model (nvidia-smi
): X Compiler (gcc --version
on Ubuntu): X