FOOD: Facial Authentication and Out-of-Distribution Detection with Short-Range FMCW Radar

TL;DR

This work introduces FOOD, a reconstruction-based framework for short-range 60 GHz FMCW radar–based facial authentication with integrated out-of-distribution detection. The architecture combines a main convolutional encoder with three decoders for ID classes and linear leaves (common and private) to enable robust OOD detection via seven reconstruction losses. On a radar dataset with $190{,}126$ ID frames and $15{,}818$ OOD frames, FOOD achieves an ID accuracy of $98.07\%$ and an OOD AUROC of $98.50\%$, outperforming state-of-the-art OOD detectors. The approach demonstrates strong classification and OOD-detection performance while maintaining a scalable design to accommodate more ID faces.

Abstract

This paper proposes a short-range FMCW radar-based facial authentication and out-of-distribution (OOD) detection framework. Our pipeline jointly estimates the correct classes for the in-distribution (ID) samples and detects the OOD samples to prevent their inaccurate prediction. Our reconstruction-based architecture consists of a main convolutional block with one encoder and multi-decoder configuration, and intermediate linear encoder-decoder parts. Together, these elements form an accurate human face classifier and a robust OOD detector. For our dataset, gathered using a 60 GHz short-range FMCW radar, our network achieves an average classification accuracy of 98.07% in identifying in-distribution human faces. As an OOD detector, it achieves an average Area Under the Receiver Operating Characteristic (AUROC) curve of 98.50% and an average False Positive Rate at 95% True Positive Rate (FPR95) of 6.20%. Also, our extensive experiments show that the proposed approach outperforms previous OOD detectors in terms of common OOD detection metrics.

Figures (2)

• Figure 1: The figure presents the high-level structure of FOOD and the zoom-in version of the highlighted section. Here we have a main convolutional one-encoder multi-decoder part (MP) and intermediate linear encoder decoder parts common leaf (CL) and private leaves (PLs). The encoder of MP encodes the ID input data and from top to bottom its decoders reconstruct the ID input PER$_1$, PER$_2$, and PER$_3$, respectively. CL has a simple linear encoder-decoder network and is responsible for OOD detection. PLs are ID class specific and also have linear encoder-decoder network each. They have a considerable effect on both classification and OOD detection. MSE function is used for the calculation of each loss.
• Figure 2: Confusion matrix to demonstrate the classification performance of FOOD.