Miris Spatial Streaming Lands as a FiftyOne Plugin for Voxel51 Physical AI Workflows

Miris has shipped a plugin that streams its radiance field backed spatial content directly into FiftyOne, the open source dataset toolkit from Voxel51, putting photorealistic 3D scenes inside the same tab where computer vision teams already inspect their labels and annotations.

The plugin works on both the open source FiftyOne release and the enterprise platform. A team updates their FiftyOne build, adds the Miris plugin, drops in a viewer key, and any Miris conditioned asset streams into FiftyOne's native 3D viewer with labels, annotations, and metadata rendered in context.

Miris's earlier public demonstration, showed its patented content delivery network streaming true spatial content to AR/VR clients with an SDK targeting Unity and Meta's Horizon OS. The FiftyOne integration moves that streaming into a data curation loop and adds a web path. The Miris Public Beta is free and now includes full SDK access for the web through three.js.

A self driving team that reconstructs an urban intersection from a drive can review it without downloading gigabytes to a workstation or provisioning a pixel-streaming GPU. The scene streams into FiftyOne alongside the dataset, where the team compares scene versions and curates training samples.

The plugin ships with a sample automated 3D annotation workflow that runs computer vision against Miris streams using FiftyOne's existing tooling, generating bounding boxes and sub-object labels. Miris argues this is hard to do cleanly on sparse point clouds or low-resolution meshes because there is not enough visual signal for a model to lock onto, and that photorealistic streams give a CV model what it needs.

The workflow runs against FiftyOne's existing sample 3D dataset, so no custom data is required to try it. Voxel51 is showing the integration live in the FiftyOne booth at CVPR, and points teams to its Physical AI Workbench for building reconstruction pipelines.

Learn more here.