Home

Awesome

mturk-text2shape

This repository contains the code we used for collecting model descriptions for ShapeNet for the Text2Shape project.

Getting started

Get mturk-rails and follow directions to install ruby on rails and setup mturk-rails according to the mturk-rails/README.md

git clone git@github.com:smartscenes/mturk-rails.git

Link experiments to mturk-rails

scripts/setup.sh mturk-rails
bundle exec rake mturk:develop[model2desc]
bundle exec rake mturk:develop[verify_model_desc]

Add to mturk-rails/config/routes.rb

  get 'experiments/model2desc', to: 'experiments/model2desc#index'
  get 'experiments/model2desc/results', to: 'experiments/model2desc#results'

  get 'experiments/verify_model_desc', to: 'experiments/verify_model_desc#index'
  get 'experiments/verify_model_desc/results', to: 'experiments/verify_model_desc#results'

Basic command to run task

To run task on Amazon MTurk

Use the mturk-rails UI to look and get your results as csv.

You can also do the following to dump the state of the database.

You can use the Amazon MTurk interface to approve and reject workers. When done, use the following to approve and pay workers and remove the hit from Amazon MTurk.

Directory structure

Basic directory structure

- experiments                     # Mturk Experiments 
  - model2desc                    # Show a image or 3d model and ask for a description
  - verify_model_desc             # Takens model2desc output and ask user to select model matching description  
- scripts/setup.sh                # Script to symlink experiment files to main mturk-rails

Each experiment has the following (following the rails file structure):

<experiment name>
  - app
    - controller                  # App controller logic 
    - views                       # View (webpage) with instructions and UI elements
  - config                        # Configuration (task description, turker pay, csv file to use)
  - javascript                    # Javascript code for the front end
  - public                        # CSV data for the task goes here