EFM3D: A Benchmark for Measuring Progress Towards 3D Egocentric Foundation Models

TL;DR

This paper introduces EFM3D, a benchmark for evaluating 3D egocentric perception models using high-quality data, and proposes EVL, a baseline model that leverages multiple modalities and outperforms existing methods.

Contribution

The paper presents the first benchmark for 3D egocentric perception and introduces EVL, a novel baseline model utilizing all egocentric modalities.

Findings

EVF3D benchmark enables standardized evaluation of 3D egocentric models.

EVL baseline outperforms existing methods on the EFM3D benchmark.

Large simulated dataset improves model training and performance.

Abstract

The advent of wearable computers enables a new source of context for AI that is embedded in egocentric sensor data. This new egocentric data comes equipped with fine-grained 3D location information and thus presents the opportunity for a novel class of spatial foundation models that are rooted in 3D space. To measure progress on what we term Egocentric Foundation Models (EFMs) we establish EFM3D, a benchmark with two core 3D egocentric perception tasks. EFM3D is the first benchmark for 3D object detection and surface regression on high quality annotated egocentric data of Project Aria. We propose Egocentric Voxel Lifting (EVL), a baseline for 3D EFMs. EVL leverages all available egocentric modalities and inherits foundational capabilities from 2D foundation models. This model, trained on a large simulated dataset, outperforms existing methods on the EFM3D benchmark.