TraverseResearchWork with us
Menu

Traverse Research

Neural reconstruction and upscaling

Super sampling, denoising and upscaling evaluated inside a renderer, alongside the cost and quality of the complete frame.

Neural reconstruction schematic: inputs, inference and image output.
Neural reconstruction schematic: inputs, inference and image output.

Reconstruction belongs to the frame

An upscaler or neural reconstruction technique needs more than an inference dispatch. The renderer must supply the inputs, manage resources, integrate the output and assess the resulting image under its actual workload.

Traverse’s programme includes Arm Neural Super Sampling integration, neural post-processing and denoising investigations. Public bindings such as arm-neural-graphics-rs and XeSS-related tooling connect vendor interfaces to the Rust rendering stack.

A repeatable comparison environment

Evolve’s 2026 programme adds upscaling quality and performance workloads, including XeSS and FSR integrations. Those benchmarks create a way to examine reconstruction alongside rendering rather than compare disconnected demonstration images.

Arm integration account · Evolve.

A few useful answers.

Does Traverse author the vendor upscalers?

The work described here is integration and evaluation. Vendor technologies retain their own authorship.

Is inference cost the only measure?

No. Input preparation, integration, image quality and the complete rendering workload also matter.

Is there a concrete integration?

We integrated Arm Neural Super Sampling into Breda and Evolve, connecting the SDK to the renderer’s inputs, resource flow and output.

Traverse Research + Traverse Games

Let’s turn a good question into working software.

Start a conversation