With colleagues Erik-Learned Miller, Jamie Morgenstern, and Joy Buolamwini we make the case that benchmarks fall short in assessing Face Recognition Technologies (FRTs) for deployment and we propose a more holistic approach.

This demo attemps to make it difficult for a model to predict gender from an image by modifying it so that this task becomes harder while retaining most image information.

This demo turns textual descriptions into a scene generated automatically by stitching objects sequentially on a plain background step-by-step using sequence generation neural networks.