Skip to content

[CVPR 2017] AMT chat interface code used to collect the Visual Dialog dataset

Notifications You must be signed in to change notification settings

batra-mlp-lab/visdial-amt-chat

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

VisDial AMT Chat

Source for the two-person chat interface used to collect the VisDial dataset (arxiv.org/abs/1611.08669) on Amazon Mechanical Turk.

VisDial AMT Interface

If you find this code useful, consider citing our work:

@inproceedings{visdial,
  title={{V}isual {D}ialog},
  author={Abhishek Das and Satwik Kottur and Khushi Gupta and Avi Singh
    and Deshraj Yadav and Jos\'e M.F. Moura and Devi Parikh and Dhruv Batra},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2017}
}

How things work

The real-time chat interface is built using Node.js and Socket.io. We use Redis to maintain a pool of images for live HITs and all data finally resides in a MySQL database.

A database table stores images from the COCO dataset each with a randomly picked caption. A batch of images from this table are then pushed to a Redis list for launching HITs. The web server corresponding to the chat interface pairs two AMT workers, assigns them roles (questioner or answerer) and shows corresponding interface, picks an image from the Redis list to collect data on and saves their conversation in the database, also marking that image as 'complete' once the HIT is done. This happens in parallel so workers aren't left waiting, and the server ensures workers have unique ids. Disconnects are handled gracefully — remaining worker is asked to continue asking questions or providing facts (captions) up to 10 messages. Once the HITs are complete, scripts in mturk_scripts/approve can be used to review, approve, reject HITs and pay workers.

Node.js

Pops images from the Redis list, retrieves relevant data from the MySQL tables, renders the chat interface and pairs workers on AMT, saves submitted data to MySQL tables

  • Install MySQL, Redis, Node
  • Install dependencies using npm install (from the nodejs folder)
  • Copy over example.config.js to config.js, and set MySQL and Redis credentials
  • Create a symbolic link from static/dataset to /path/to/mscoco/images/
  • Update app path (from vhost) in index.html line 276
  • Running node index.js should now serve the interface at 127.0.0.1:5000

MTurk Scripts

Scripts to set up MySQL database, populate Redis list, and approve/reject HITs

  • Copy over example.config.json to config.json, and set MySQL credentials. from_timestamp is unix timestamp before launching a batch of HITs
  • createDatabase.py and fillDatabase.py create MySQL tables, populate it with COCO images and captions and generate the Redis list for a batch of HITs
  • createHits.py launches HITs on AMT

Approving/Rejecting HITs (mturk_scripts/approve)

  • Copy over example.config.json to config.json, and set MySQL credentials
  • Copy over example.constants.py to constants.py, and set AMT credentials
  • reviewHits.py gets completed HITs and saves them to amthitsQues.csv and amthitsAns.csv for review
  • Once HITs have been reviewed in the CSV files (by changing 'notApprove' to 'reviewReject' or 'approve'), run approveHits.py to mark approved HITs for payment and reviewRejectedHits.py, rejectHits.py to reject HITs, and finally payWorkers.py to process payments

Contributors

Acknowledgements

Parts of this code (MTurk scripts) are adapted from @jcjohnson's simple-amt project.

License

BSD