用普通的安装方式走了不少弯路，感觉还是用docker方便：
参考的是https://hub.docker.com/r/caffe2ai/caffe2/
Latest
docker pull caffe2ai/caffe2
Comes with GPU support, CUDA 8.0, cuDNN 7, all options, and tutorial files. Uses Caffe2 v0.8.1.
GPU images (for use with nvidia-docker)
All GPU builds come with Ubuntu 16.04 and fully loaded with optional dependencies like OpenCV. Minimal builds and Ubuntu 14.04 can be found in the CPU section.
- Caffe2 v0.8.1, CUDA 8, cuDNN 7:
docker pull caffe2ai/caffe2:c2v0.8.1.cuda8.cudnn7.ubuntu16.04 - Caffe2 v0.8.0, CUDA 8, cuDNN 6:
docker pull caffe2ai/c2v0.8.0.cuda8.cudnn6.ubuntu16.04 - Caffe2 v0.7.0, CUDA 8, cuDNN 6:
docker pull caffe2ai/c2.cuda8.cudnn6.ubuntu16.04 - Caffe2 v0.7.0, CUDA 8, cuDNN 5:
docker pull caffe2ai/c2.cuda8.cudnn5.ubuntu16.04
CPU Images
- Caffe2 v0.8.1 on Ubuntu 16.04 and full options:
docker pull caffe2ai/caffe2:c2v0.8.1.full.ubuntu16.04 - Caffe2 v0.8.1 on Ubuntu 16.04 and minimal options:
docker pull caffe2ai/caffe2:c2v0.8.1.min.ubuntu16.04 - Caffe2 v0.8.1 on Ubuntu 14.04 and full options:
docker pull caffe2ai/caffe2:c2v0.8.1.full.ubuntu14.04 - Caffe2 v0.8.1 on Ubuntu 14.04 and minimal options:
docker pull caffe2ai/caffe2:c2v0.8.1.min.ubuntu14.04