Dynamic Inverse Rendering for Enhanced Material-Lighting Decomposition

ECCV 2026

Raza Yunus1,3      Benjamin Ummenhofer2      Jan Eric Lenssen3      Eddy Ilg1,†
1University of Technology Nuremberg   2Intel   3MPI for Informatics, SIC
Now at Google

Key idea: Observing an object in motion, e.g. in a hand-held capture, provides stronger constraints for material-lighting decomposition than a static object capture.

Abstract

Decomposing outgoing surface radiance into material and illumination during inverse rendering is essential for applications such as relighting and augmented reality, yet it is severely ill-posed since multiple combinations can result in the same observed colour. Capturing an object under multiple lighting conditions usually helps resolve this ambiguity as it constrains the optimization towards correct solutions. In this work, we explore the potential of reconstructing rigidly moving objects—which provides observations of diverse light-surface interactions—to resolve the material-lighting ambiguity in inverse rendering. For this purpose, we introduce a relightable approach that marries object tracking and reconstruction with inverse rendering for general rigidly moving objects. Our experimental analysis on synthetic data demonstrates that motion can be an advantage for disentangling material and lighting: the reconstructed material is significantly more accurate when the object is observed under rigid motion than when it is static. Moreover, results on RGB videos of real hand-held objects show that our pipeline preserves this advantage even under noisy real-world conditions.

Contributions

1 Motion as a constraint

We show that observing significant object motion provides stronger material-lighting constraints than static object capture.

2 Relightable 4D pipeline

We jointly handle object pose tracking, geometry reconstruction and physically based material-lighting estimation.

3 Controlled benchmark

We introduce synthetic captures with ground-truth materials and relit views across static dome, turntable, and hand-held settings.

Method

1. Progressive initialization

Progressive optimization with a neural SDF representation to get coarse initial estimates for object poses and geometry.

2. Global refinement

Refinement with an SDF-based 3D Gaussian representation to capture detailed geometry for inverse rendering.

3. Dynamic inverse rendering

Physically-based rendering from posed Gaussians to obtain accurate material-lighting decomposition from a moving object.

Overview of the three-stage dynamic inverse rendering method.

Results

Synthetic capture settings

Qualitative: From static Static capture to turntable Turntable capture to hand-held Hand-held capture captures, albedo and lighting become progressively less entangled, producing better relights across objects.

Synthetic qualitative comparison across static, turntable, and hand-held capture.

Quantitative: Hand-held rotations consistently improve albedo-light disentanglement and relighting quality for both the original (specular) and diffuse variants of our dataset, even outperforming the static capture (which uses ground-truth poses) when the object poses are estimated.

Synthetic quantitative comparison across static, turntable, and hand-held capture.

Real captures

Real hand-held sequences produce cleaner albedo and more plausible relighting than static captures under similar render quality.

Real capture comparison for static multiview and hand-held settings.

Analysis

Roughness variation

Hand-held capture recovers sharper illumination, even for diffuse objects where static capture is under-constrained.

Environment reconstruction under varying roughness.

Illumination complexity

Under complex lighting, static albedo quality degrades while hand-held capture maintains high quality.

Effect of illumination complexity on decomposition.

View sampling

25 hand-held views outperform 400 static views, showing that the decomposition gain is independent of view sampling.

View sampling ablation comparing static and hand-held views.

Rotation diversity

Increasing diversity in object rotations improves disentanglement for complex environments.

Effect of diversity in hand-held rotations on decomposition.

BibTeX


      @inproceedings{yunus2026dynamic,
        author = {Yunus, Raza and Ummenhofer, Benjamin and Lenssen, Jan Eric and Ilg, Eddy},
        title = {Dynamic Inverse Rendering for Enhanced Material-Lighting Decomposition},
        year = {2026},
        booktitle = {European Conference on Computer Vision (ECCV)},
      }