MRD: Using Physically Based Differentiable Rendering to Probe Vision Models for 3D Scene Understanding

TL;DR

This paper introduces MRD, a differentiable rendering approach that probes vision models' implicit understanding of 3D scene properties by finding physically different scenes that produce similar model activations.

Contribution

MRD provides a physically grounded method to analyze vision models' sensitivity and invariance to 3D scene attributes, advancing interpretability of deep learning models.

Findings

High similarity in model activation between target and optimized scenes.

Qualitative reconstructions reveal models' sensitivities to physical scene attributes.

MRD enables analysis of how physical scene parameters influence model responses.

Abstract

While deep learning methods have achieved impressive success in many vision benchmarks, it remains difficult to understand and explain the representations and decisions of these models. Though vision models are typically trained on 2D inputs, they are often assumed to develop an implicit representation of the underlying 3D scene (for example, showing tolerance to partial occlusion, or the ability to reason about relative depth). Here, we introduce MRD (metamers rendered differentiably), an approach that uses physically based differentiable rendering to probe vision models' implicit understanding of generative 3D scene properties, by finding 3D scene parameters that are physically different but produce the same model activation (i.e. are model metamers). Unlike previous pixel-based methods for evaluating model representations, these reconstruction results are always grounded in physical scene descriptions. This means we can, for example, probe a model's sensitivity to object shape while holding material and lighting constant. As a proof-of-principle, we assess multiple models in their ability to recover scene parameters of geometry (shape) and bidirectional reflectance distribution function (material). The results show high similarity in model activation between target and optimized scenes, with varying visual results. Qualitatively, these reconstructions help investigate the physical scene attributes to which models are sensitive or invariant. MRD holds promise for advancing our understanding of both computer and human vision by enabling analysis of how physical scene parameters drive changes in model responses.