Open boazsender opened 7 years ago
UPDATES
What are we doing? Changes: friendlier text
Updated Text We are using the power of the crowd to tag images in IMDB-WIKI, the largest publicly available face data set for researchers, in order to increase data transparency.
Why are we doing this? Changes: friendlier text
Updated Text Algorithms used to recognize faces are trained and tested with data sets. A lack of diversity in these data sets can lead to biased systems. We need your help to checking for bias by tagging training images with information about age, gender, and ethnicity. Your tags will create insights for developing more inclusive data sets.
How can I submit feedback? (Lower-cased question to match others) Changes - Included link to feedback form https://goo.gl/forms/uHTUeua2DOgZLdkk1 and corrected typo
Updated Text We welcome your feedback on the goals of this project, on the labels we are using, and on usability of this app. Please submit your feedback using our project's feedback form.
What if there's no person in the image? _Changes: Current site shows &lquo;flag&rquo; lets replace with report image
Updated Text Our data set is based on images from IMDB-WIKI, and have been auto-cropped. During the course of tagging you may come across images with an unclear subject, no subject at all, or inappropriate content. Use the report image button to help us refine our data set. Thank you!
Destiny Unveiled: Facial Data Transparency Project -> Facial Data Transparency Project
If data is destiny, how inclusive is our future data-centric technology? How representative are the datasets being used to train machines to analyze faces? Help us find out by tagging training images. - >
Help us check for bias by tagging pictures of celebrities with demographic labels.
delete: Make a difference Simple Actions Matter Be Heard
FAQ: What are we doing? We are annotating IMDB-Wiki, the largest publicly available face-data set for researchers, in order to increase data transparency.
Why are we doing this? Algorithms used to recognize faces are trained and tested with data sets. A lack of diversity in these data sets can lead to biased systems. We currently have no idea how representative our computer vision research data sets are. We're annotating this data set in order to asses demographic representation, develop a new evaluation metric, and create insights for better systems.
How did you choose your demographic labels? The demographic labels we selected are reflective of existing conventions in western classification systems. We acknowledge that specific demographic categorizations are imprecise and reductive. We choose to gather perceptions of identity because the social constructs of ethnicity and gender have real-world impact. Our goal in gathering these tags is ultimately to create systems that work well for everybody and do not disproportionately harm a specific group. We welcome your feedback on additional labels to include or alternative ways to assess the diversity of these data sets.
What if there's no person in the image? Our data set is based on images from IMDB, and have been auto-cropped. During the course of annotating you may come across images with an unclear subject, no subject at all, or inappropriate content. We ask you to "flag" these images using the flag button underneath each image to help us refine our data set. Thank you!
How long will this take? Annotating each image should take 30 or less seconds on each image. Your first batch should take less than 5 minutes.
Why do I have to submit 12 images in order to get started? We require 12 annotations to start using the system to get a sense of how you annotate. Once you submit 12, you'll be able to do smaller batches of 3 at a time.
How Can I Submit Feedback? We welcome your feedback on the goals of this project, on the labels we are using, and on usability of this app. Please submit your feedback using our project's feedback from.
Consent Form AJL.AI is part of an ongoing MIT scientific research project. Your decision to participate in annotating images is voluntary. We do not collect personally identifiable data with your annotations. The data we will have, in addition to your responses, is the time at which you completed annotating an image.
The results of the research may be presented at scientific meetings or published in scientific journals. Selecting the 'I Agree' button indicates that you are at least 18 years of age and agree to annotate voluntarily.
[I Agree] [I Disagree]