Skip to content

Setup guide on GCP

Jump to bottom
Fei Xia edited this page May 12, 2021 · 7 revisions

Here are the steps to run gibson on GCP,

Provisioning

Use a similar configuration as the following to create a virtual machine, choose K80, P100 or V100 GPU. I choose c1-deeplearning-pytorch-0-4-cu92-20181009 image because it has CUDA and nvidia-docker already installed so requires minimal setup.

Setup

  1. Install nvidia-driver:

  2. Setup firewall rules to allow traffic (the web UI uses flask, the port is configurable)

  3. Download dataset and pull gibson docker:

wget https://storage.googleapis.com/gibson_scenes/dataset.tar.gz
tar -zxf dataset.tar.gz
docker pull xf1280/gibson:0.3.1

Running

docker run --runtime=nvidia -ti -v <dataset root>:/root/mount/gibson/gibson/assets/dataset -p 5001:5001 xf1280/gibson:0.3.1

In docker container, open tmux and run python examples/demo/benchmark_fps.py in one tab and python gibson/utils/web_ui.py in another.

To order to use web UI, you would need to change mode into web_ui in yaml file.

You should be visit gcp-external-ip:5001 to see the web UI rendered.

Clone this wiki locally