From Part to Whole: 3D Generative World Model with an Adaptive Structural Hierarchy

ICME 26 2026 44.6 method

TLDR

A part-to-whole 3D generative world model that adaptively discovers latent structural slots from single images for cross-category shape generation.

Reasoning

Strengths: novel adaptive slot-gating and prototype bank for compact, cross-category shape representation. Weaknesses: abstract lacks experimental validation, benchmarks, or comparisons; unclear generalization to real-world data.

Read-first score

Read-first score 44.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 19.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=ablation,experiment,metric

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Topical relevance 42%

27.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 529.

Keyword Scores

generative world model

world model

world simulator

interactive world model

video world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Adaptive part-whole hierarchy for single-image 3D generation that autonomously discovers latent structural slots via soft compositional masks from image tokens
Adaptive slot-gating mechanism that dynamically determines slot-wise activation probabilities and consolidates redundant slots for compact yet expressive structure
Learnable, class-agnostic prototype bank that enables cross-category shape sharing and denoising through universal geometric prototypes
Lightweight 3D denoiser that reconstructs geometry and appearance via unified diffusion objectives

Methodology

The model learns an adaptive part-whole hierarchy in a flexible 3D latent space. It uses an adaptive slot-gating mechanism to infer soft compositional masks directly from image tokens, dynamically determining slot activation probabilities and consolidating redundant slots. Each distilled slot is aligned to a learnable, class-agnostic prototype bank for cross-category shape sharing, and a lightweight 3D denoiser reconstructs geometry and appearance via unified diffusion objectives.

Key Results

The method achieves consistent gains in cross-category transfer and part-count extrapolation, and ablations confirm the complementary benefits of the prototype bank for shape-prior sharing and slot-gating for structural adaptation.

Resources

Paper