Before you can run experiments on MTurk, you need to obtain an MTurk account and an SSL certificate.
You need an Amazon MTurk Requester account from https://requester.mturk.com/. Place your secret key info in MTURK_AWS_ACCESS_KEY_ID and MTURK_AWS_SECRET_ACCESS_KEY in the file server/config/settings_local.py.
Note that there is room for two sets of keys: we allow for you to use different keys for sandbox mode. While you can make the two sets the same, you my find it useful to use an entirely different account for the sandbox.
Unfortunately you need to set up SSL in order to properly use MTurk as a nested iframe (this wasn’t always the case). Without a certificate, the browsers of your workers will notice that mturk.com is served over HTTPS, but that your task is served over HTTP, and refuse to serve your task.
If you already have an SSL certificate for your server (with the appropriate hostname), you can use that. Otherwise, I had good success obtaining a reasonably cheap SSL certificate with Gandi. Any other provider also works.
Once you have the certificate, update nginx to have a path to your certificate by editing /etc/nginx/sites-available/PROJECT_NAME, where PROJECT_NAME is specified in scripts/config.sh (default: opensurfaces).
Update server/config/settings_local.py with ENABLE_SSL = True.
Restart both nginx and gunicorn (see Sytem architecture). You should now be able to visit the site with the https: prefix.
If you want to use our segmentation task ("segment_material"), then: you need to install CGAL (version >= 4.1, < 5, from http://www.cgal.org) and build the triangulate program (cd triangulate; ./build.sh). The Ubuntu version of CGAL is too old. Note that you may have to add /usr/local/lib to your LD_LIBRARY_PATH if it not already there.