OmniLatent introduces unified latent for 3D part modeling
Original title: OmniLatent: Unified Latent for 3D Part Modeling
OmniLatent enables single-image 3D part decomposition by learning a shared latent space for geometry, deformation, appearance, and part IDs.
OmniLatent is a unified latent modeling framework for 3D part decomposition. It encodes geometry, deformation, appearance, and binary part IDs into a shared latent space to support data-scarce part learning with geometric and visual priors. Given one input image, the model generates geometry, deformation, texture, and part IDs in one pass, producing a part-aware 3D asset without separate segmentation.
- Shared latent space jointly encodes geometry, deformation, appearance, and binary part IDs.
- Point-prompt adapter maps arbitrary 3D point prompts into latent tokens for segmentation.
- Amodal part completion model completes partial masks using dynamic token allocation.
- Produces semantically meaningful part segmentations with precise boundaries.
- Vendors
- Hugging Face
- AI models
- OmniLatent
From the original article
1 min read at huggingface.co
3D part decomposition is a central capability for editable asset creation, yet existing workflows typically segment a completed mesh as a post-processing step, making the result sensitive to geometric artifacts, texture ambiguity, and limited part-level annotations. We present OmniLatent, a unified latent modeling framework for 3D part decomposition.
By Yunhan Yang, Yunfei Zhao, Zeqiang Lai · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.