Scenes as Objects, Not Primitives: Instance-Structured 3D Tokenization from Unposed Views

2026-06-28

AI Summary

Existing feed-forward 3D reconstruction methods produce unstructured point clouds or Gaussian sets, making it difficult to capture the instance-level structure of objects in a scene. This work proposes a feed-forward framework that directly generates instance-centric 3D token groups from pose-free multi-view images. Each group comprises an instance token encoding object identity and multiple anchor tokens representing local geometry and appearance, which are subsequently decoded into 3D Gaussians. By designating object instances as native primitives within the feed-forward representation, the method achieves disentanglement of identity and appearance through a two-level token decomposition, enabling instance-level editing and efficient retrieval without requiring 3D annotations. Experiments demonstrate that the approach outperforms per-scene optimization baselines in category-agnostic instance segmentation while maintaining strong novel view synthesis quality, and further supports open-vocabulary 3D retrieval based on instance counts.

Abstract

A 3D scene is understood through its objects, not the primitives that compose them. Yet feed-forward reconstruction methods output dense, unstructured sets of points or Gaussians, leaving object-level structure to be recovered after the fact. We propose a feed-forward framework that decomposes a scene into instance-structured 3D token groups directly from unposed multi-view images -- compact object-centric units from which reconstruction, segmentation, and manipulation all follow. Each token group pairs an instance token capturing entity-level identity with anchor tokens that encode local geometry and appearance, which are decoded into a set of 3D Gaussians. This two-level factorization decouples object identity from local appearance, making object instances a native interface of the representation rather than a derived product. The token groups are learned through differentiable rendering with joint reconstruction and segmentation supervision, requiring no 3D annotations. Our feed-forward model surpasses per-scene optimization baselines in class-agnostic instance segmentation while remaining competitive in novel view synthesis. Beyond these metrics, the same token groups directly unlock instance-level scene editing -- removing, translating, or inserting objects by operating on their groups -- as well as efficient open-vocabulary 3D instance retrieval, where retrieval complexity scales with the number of instances rather than primitives.

Problem

• 3D scene representation
• instance segmentation
• object-centric representation
• unposed multi-view images
• structured 3D tokenization

Innovation

• instance-structured 3D representation
• feed-forward 3D reconstruction
• object-centric tokenization
• differentiable rendering
• 3D instance editing