Seedera

Scalable data infrastructure for world models.

Engine-generated environments, with camera pose on every frame and ground-truth layers on sampled frames.

Request data

Seed 1303

Layout

Built by a team from Vision Pro ByteDance

Pose6-DoFcamera position and rotation, every frame

Depth16-bitwhole centimetres, planar

Segmentation9 classessky to props

NormalsWorld space8-bit RGB

Depth, segmentation and normals are produced for sampled frames.

One frame, four layers

Segmentation of a sample frame, one colour per class — SegmentationStill

Depth of a sample frame, false colour — DepthStill

Surface normals of a sample frame — NormalsStill

Four layers of one frame from one of our environments. Depth and segmentation colours are a visualisation, not the raw values. Segmentation ground-to-road labels are an estimate.