We introduce a novel privacy-preserving methodology for performing Visual Question Answering on the edge. Our method constructs a symbolic representation of the visual scene, using a low-complexity computer vision model that jointly predicts classes, attributes and predicates. This symbolic representation is non-differentiable, which means it cannot be used to recover the original image, thereby keeping the original image private. Our proposed hybrid solution uses a vision model which is more than 25 times smaller than the current state-of-the-art (SOTA) vision models, and 100 times smaller than end-to-end SOTA VQA models. We report detailed error analysis and discuss the trade-offs of using a distilled vision model and a symbolic representation of the visual scene.
@article{arxiv.2202.07712,
title = {Privacy Preserving Visual Question Answering},
author = {Cristian-Paul Bara and Qing Ping and Abhinav Mathur and Govind Thattai and Rohith MV and Gaurav S. Sukhatme},
journal= {arXiv preprint arXiv:2202.07712},
year = {2022}
}