← All resources

Neural Super Sampling

Auto-discovered

Library / Codebase Advanced

Arm Limited

Visit resource ↗

Description

Neural Super Sampling is an Arm model card on Hugging Face for NSS, a temporal super sampling network aimed at mobile GPUs that reconstructs higher-resolution frames from lower-resolution temporal inputs.

It covers NSS v1 high, mid, and low quality modes, VGF packages for Vulkan ML Extensions, Unreal plugin and Neural Graphics SDK integration paths, scenario-runner demos that upsample 540p to 1080p, retraining through Neural Graphics Model Gym, and limits such as needing motion vectors and history, plus padding when dimensions are not divisible by 8.

It is for advanced mobile, XR, and engine graphics engineers integrating neural upscaling on Neural Accelerators. Vulkan ML Extensions or the Unreal NSS plugin are the expected integration routes.

Weights and example scenarios are freely downloadable from the model repository under Arm's AI Model Community License, with linked learning paths for Vulkan and Unreal.

It is useful when you need a mobile-oriented temporal upscaler with quality knobs and engine examples rather than a desktop-only solution.