Procedurally generated indoor scenes with ground-truth path-traced renders and precomputed 3-slat VAE latents, built for training [RenderFormer]-style neural renderers. Each sample is one scene observed from 14 camera poses along an orbit. This is a [WebDataset]. Each sample carries: The latents hold three roles sharing one coordinate grid per frame (a unified-proxy Frames are concatenated along the token axis with a frameoffsets index rather than stored separately, so frame i is coords[frameoffsets[i]:frameoffsets[i+1]]. Token count varies slightly per frame (~2750) because light proxies are voxelized in per frame. Fields arrive as raw bytes. Decode them with: coords are integer…
Independent publisher
Sicheng Zhou
eternity304
Models in Library0
Datasets in Library1
Models on Hugging Face—
Followers—
Compare the published evidence
This profile connects the 1 model and dataset record shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Published tasks: Image to image.
Reported licenses: cc-by-4.0. Open a record to review its specific license and conditions.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.