Scenario
← All Models

This model is deprecated. No direct replacement is recommended yet.

3D

PartCrafter

Use it ↗

PartCrafter by PKU turns a single image into up to 16 separate, semantically distinct 3D meshes in one pass, no segmentation required.

PartCrafter prioritizes modular reconstruction through a Local-Global Denoising Transformer architecture. It simultaneously synthesizes 2 to 16 semantically meaningful and geometrically distinct 3D meshes from one RGB image without pre-segmentation. The model utilizes a compositional latent space where each part is represented by disentangled tokens and identity embeddings. It reconstructs occluded or hidden geometry with high structural coherence.

More models from Academia