I am trying to run Idein/chainer-pose-proposal-net on a Raspberry Pi 3 B+. This is a pose estimation model (specifically Pose Proposal Networks by Taiki Sekii) in Chainer.
In the README it is suggested that it can use the Raspberry's VideoCore IV GPU to achieve up to 10fps. My problem is that I do not know how to make it use the GPU. What has to be installed?
I looked at CuPy which would be the way to go if it's a Nvidia GPU and CUDA installed. I could not find an alternative for the Raspberry.
I also had a look at Actcast but there is no documentation on how to use it.
python3 video.py trained/mobilenetv2_224x224_1.0_mpii/
Desired result is to get close to 10fps, but currently without GPU I only get ~0.4fps.
Edit 2019-05-28:
In the meantime tried the following. Unfortunately I am still unable to achieve the desired frame rate:
- I successfully installed QMKL with all dependencies py-videocore etc.
- I installed VC4CL OpenCL 1.2 standard for the VideoCore IV GPU
- I found clpy which is a is CuPy but using OpenCL backend. I successfully installed and tested it and import it, but I do not get improved performance.
