A web-based image landmarking tool intended for use with Amazon Mechanical Turk. Currently this is not very general and will actually only work for one specific set of images, but it would not be too difficult to adapt it to your own uses.
- app.html - entry point into the application.
- js/ - JavaScript files. Includes Bootstrap, jQuery, the JSON library, and the Amazon mTurk library.
- js/base.js - all custom JavaScript functionality.
- css/ - CSS sheets, including Bootstrap.
- css/base.css - all custom CSS style information.
- img/ - images for context-sensitive help. If these change then helpimgs.js needs to be updated as well.
- protocol/ - files that generate the full help file that contains instructions for landmarkers
You can deploy these files to Amazon as-is, though you will need to use the Amazon Mechanical Turk Command Line Tools, or the API, to format these HITs as an ExternalQuestion. They are too large (>128kb or so) to be inlined on the mTurk site.
For extra speed, you can compress and inline the app file. This requires node.js, inliner (via npm), and all its dependencies. Then, execute compress.sh, or run inliner -vi app.html > app_small.html in your Terminal.
To build the help instructions, you'll need pandoc. Navigate to the protocol/ directory and execute either build.sh or build.ps1, depending on your operating system. This will create a HTML and PDF file of the landmarking protocol.
- Better review interface
- Rip out dependencies on jQuery and bootstrap
- Streamline HIT posting
This research is supported by an Encyclopedia of Life Rubenstein Fellowship to Jonathan Chang.