Modelpublic

pci-lab/worldflow3d

sourceHugging Faceotherupdated 3mo agoView on Hugging Face

0likes6downloads

Model Card

WorldFlow3D — Front3D (indoor) models

Layout-conditioned 3D indoor-scene generation (3D-FRONT), as a coarse → color refinement flow-matching cascade. Use with the `worldflow3d` package (pip install worldflow3d).

License. These models are trained on the 3D-FRONT dataset and are released for non-commercial / research use under the 3D-FRONT dataset terms. The worldflow3d code is Apache-2.0, but that license does not grant rights to these 3D-FRONT-derived weights. See the 3D-FRONT dataset terms.

The Waymo (outdoor) models are trained on the Waymo Open Dataset and live in a separate repo, `pci-lab/worldflow3d-waymo`, under the Waymo Dataset License (non-commercial).

Cascade stages

Subfolder	Role
`front3d-coarse`	coarse layout-conditioned generation (direct-diffusion)
`front3d-color`	source-flow + color refinement (UNet in=4, color mesh sidecar)

Usage

python

from worldflow3d import WorldFlow3DPipeline

pipe = WorldFlow3DPipeline.from_hub(
    "pci-lab/worldflow3d", stage="front3d-coarse",
    refinement_stages=["front3d-color"], device="cuda",
)

See the GitHub repo for full docs and the generate_indoor CLI.