TraverseWork with us
Menu

Traverse Research

Neural material representation

Investigating the quality, efficiency and memory trade-offs of neural bidirectional texture functions.

NeuBTF research imagery published by Traverse Research.
NeuBTF research imagery published by Traverse Research.
Partner
In-house research
Disciplines
Machine learning, Rendering
Platforms
PC

Problem

A bidirectional texture function represents how a material’s appearance changes with lighting and viewing direction. That richness creates a representation problem: preserving the appearance while making evaluation and storage practical.

Constraints

Quality, efficiency and memory use have to be considered together. An offline reconstruction is not the same as an integrated rendering implementation, so the technique needs to be evaluated in both contexts.

What we did

Traverse investigated enhancements to NeuMIP-based bidirectional texture functions. The work was demonstrated in Mitsuba 2 and in the in-house Breda real-time framework, connecting the representation to a rendering environment.

Results

The public project page includes the publication, source code and the authors’ material-quality and compression comparisons.

Read the published research

Traverse Research + Traverse Games

Let’s turn a good question into working software.

Start a conversation