mitmedialab / ajl.ai

A web application for crowdsourcing image annotations.
GNU Affero General Public License v3.0
16 stars 9 forks source link

Create baseline visualization on home page #25

Open boazsender opened 7 years ago

boazsender commented 7 years ago

Primary goal: motivate citizens to participate.

Show users how their participation improves their participation

boazsender commented 7 years ago

From @tafsiri: show a head to head of this systems labels against the NIST standard.

boazsender commented 7 years ago

from @tafsiri: build it up in three stages?

1) What is the existing data

2) What are the actions that are done to faces based on these data

3) What impact can you, and have you had by being a part of this annotator community.

boazsender commented 7 years ago

A different idea from @vlandham:

enter your demographics and see how you are represented in a set of histograms of datasets by demographics.

Example:

Then additional UI something like flowing datas to enter in personal data, such as:

those faces matching this description light up in the distribution. Additional meta data about how much of the dataset vs average across all datasets is presented.

tafsiri commented 7 years ago

Btw another thing that occurred to me on the way home (which is probably obvious to you). Is that i've been thinking about the dataset like its a training dataset when its actually a (held out) test dataset (is that correct)? Which could allow for another framing as well around "how even in measuring the accuracy of a given tool its only taking X number of people from this demographic into account", which is maybe what you were going for and i just understood.