Explanations for Occluded Images

causaLens Principal Investigator Hana Chockler’s research paper “Explanations for Occluded Images” has been accepted to ICCV, the premier computer vision conference. The paper, published in collaboration with researchers from Amazon Science, leverages causal analysis to produce more reliable explanations of image classifiers. In particular, causality-based techniques produce better explanations where the input image depicts a partially occluded object.

You can view Hana presenting the paper at ICCV: